aboutsummaryrefslogtreecommitdiff
path: root/gdb/ch-exp.c
diff options
context:
space:
mode:
authornobody <>2002-08-24 00:41:00 +0000
committernobody <>2002-08-24 00:41:00 +0000
commit13c00c23454f3278917d26d1c0f337bb6ebec579 (patch)
tree26476238215a235e57fbb0ff2d33a8173e308a99 /gdb/ch-exp.c
parentb4e397edc3619279f3e370d5c76551a7f9e843d7 (diff)
downloadgdb-13c00c23454f3278917d26d1c0f337bb6ebec579.zip
gdb-13c00c23454f3278917d26d1c0f337bb6ebec579.tar.gz
gdb-13c00c23454f3278917d26d1c0f337bb6ebec579.tar.bz2
This commit was manufactured by cvs2svn to create branch 'readline_4_3readline_4_3-import-branchpoint
-import-branch'. Sprout from kseitz_interps-20020528-branch 2002-06-19 16:55:29 UTC nobody 'This commit was manufactured by cvs2svn to create branch' Cherrypick from kseitz_interps-20020528-branch 2002-07-22 19:56:08 UTC nobody 'This commit was manufactured by cvs2svn to create branch': bfd/cpu-ip2k.c bfd/vaxbsd.c gdb/config/i386/nm-x86-64linux.h gdb/config/i386/tm-x86-64linux.h gdb/config/vax/tm-vaxbsd.h gdb/gdb_locale.h gdb/i386obsd-nat.c include/elf/ip2k.h opcodes/ip2k-asm.c opcodes/ip2k-desc.c opcodes/ip2k-desc.h opcodes/ip2k-dis.c opcodes/ip2k-ibld.c opcodes/ip2k-opc.c opcodes/ip2k-opc.h Cherrypick from kseitz_interps-20020528-branch 2002-08-09 15:39:20 UTC nobody 'This commit was manufactured by cvs2svn to create branch': bfd/elf32-ppcqnx.c bfd/elfarmqnx-nabi.c bfd/elfn32-mips.c gdb/config/powerpc/aix432.mh gdb/gdb_obstack.h gdb/mips-irix-tdep.c gdb/solib-irix.c include/gdb/sim-h8300.h opcodes/po/pt_BR.po Cherrypick from kseitz_interps-20020528-branch 2002-05-28 18:43:09 UTC nobody 'This commit was manufactured by cvs2svn to create branch': gdb/event-loop.c gdb/event-top.h gdb/mi/mi-cmds.c gdb/mi/mi-cmds.h gdb/mi/mi-main.c gdb/wrapper.c gdb/wrapper.h Cherrypick from master 2002-08-24 00:40:59 UTC Mark Kettenis <kettenis@gnu.org> '* valprint.c (print_longest) [CC_HAS_LONG_LONG &&': ChangeLog MAINTAINERS Makefile.in bfd/ChangeLog bfd/ChangeLog-0001 bfd/Makefile.am bfd/Makefile.in bfd/aout-adobe.c bfd/aout-ns32k.c bfd/aout-target.h bfd/aout-tic30.c bfd/aoutf1.h bfd/aoutx.h bfd/archive.c bfd/archures.c bfd/armnetbsd.c bfd/bfd-in.h bfd/bfd-in2.h bfd/bfd.c bfd/binary.c bfd/bout.c bfd/cache.c bfd/coff-a29k.c bfd/coff-alpha.c bfd/coff-i386.c bfd/coff-mips.c bfd/coff-or32.c bfd/coff-rs6000.c bfd/coff64-rs6000.c bfd/coffcode.h bfd/coffgen.c bfd/cofflink.c bfd/config.bfd bfd/configure bfd/configure.in bfd/corefile.c bfd/cpu-h8300.c bfd/cpu-ns32k.c bfd/cpu-powerpc.c bfd/doc/ChangeLog bfd/doc/Makefile.am bfd/doc/Makefile.in bfd/doc/bfdint.texi bfd/doc/header.sed bfd/dwarf2.c bfd/ecoff.c bfd/ecofflink.c bfd/elf-bfd.h bfd/elf-eh-frame.c bfd/elf-hppa.h bfd/elf-m10200.c bfd/elf-m10300.c bfd/elf-strtab.c bfd/elf.c bfd/elf32-arc.c bfd/elf32-arm.h bfd/elf32-avr.c bfd/elf32-cris.c bfd/elf32-d10v.c bfd/elf32-d30v.c bfd/elf32-dlx.c bfd/elf32-fr30.c bfd/elf32-frv.c bfd/elf32-h8300.c bfd/elf32-hppa.c bfd/elf32-i370.c bfd/elf32-i386.c bfd/elf32-i386qnx.c bfd/elf32-i860.c bfd/elf32-ip2k.c bfd/elf32-m32r.c bfd/elf32-m68hc11.c bfd/elf32-m68hc12.c bfd/elf32-m68k.c bfd/elf32-mcore.c bfd/elf32-mips.c bfd/elf32-openrisc.c bfd/elf32-or32.c bfd/elf32-ppc.c bfd/elf32-qnx.h bfd/elf32-s390.c bfd/elf32-sh.c bfd/elf32-sh64.c bfd/elf32-shqnx.c bfd/elf32-sparc.c bfd/elf32-v850.c bfd/elf32-vax.c bfd/elf32-xstormy16.c bfd/elf64-alpha.c bfd/elf64-hppa.c bfd/elf64-mips.c bfd/elf64-mmix.c bfd/elf64-ppc.c bfd/elf64-ppc.h bfd/elf64-s390.c bfd/elf64-sh64.c bfd/elf64-sparc.c bfd/elf64-x86-64.c bfd/elfarm-nabi.c bfd/elfarm-oabi.c bfd/elfcode.h bfd/elfcore.h bfd/elflink.c bfd/elflink.h bfd/elfxx-ia64.c bfd/elfxx-mips.c bfd/elfxx-mips.h bfd/elfxx-target.h bfd/format.c bfd/hash.c bfd/i386linux.c bfd/i386lynx.c bfd/i386msdos.c bfd/i386os9k.c bfd/ieee.c bfd/ihex.c bfd/libaout.h bfd/libbfd-in.h bfd/libbfd.c bfd/libbfd.h bfd/libcoff-in.h bfd/libcoff.h bfd/libxcoff.h bfd/linker.c bfd/m68klinux.c bfd/merge.c bfd/mmo.c bfd/nlm-target.h bfd/nlm32-alpha.c bfd/nlm32-i386.c bfd/nlm32-ppc.c bfd/nlm32-sparc.c bfd/nlmcode.h bfd/ns32k.h bfd/ns32knetbsd.c bfd/oasys.c bfd/opncls.c bfd/pdp11.c bfd/pe-i386.c bfd/peXXigen.c bfd/pei-i386.c bfd/peicode.h bfd/po/SRC-POTFILES.in bfd/po/es.po bfd/po/fr.po bfd/po/sv.po bfd/po/tr.po bfd/ppcboot.c bfd/reloc.c bfd/reloc16.c bfd/section.c bfd/som.c bfd/sparclinux.c bfd/srec.c bfd/stabs.c bfd/sunos.c bfd/syms.c bfd/targets.c bfd/tekhex.c bfd/vaxnetbsd.c bfd/versados.c bfd/version.h bfd/vms-gsd.c bfd/vms-misc.c bfd/vms.c bfd/xcoff-target.h bfd/xcofflink.c config-ml.in config.guess config.sub config/ChangeLog configure configure.in contrib/ChangeLog contrib/texi2pod.pl etc/ChangeLog etc/texi2pod.pl gdb/ChangeLog gdb/MAINTAINERS gdb/Makefile.in gdb/NEWS gdb/PROBLEMS gdb/README gdb/abug-rom.c gdb/ada-exp.y gdb/ada-lang.c gdb/ada-lang.h gdb/ada-tasks.c gdb/ada-typeprint.c gdb/ada-valprint.c gdb/aix-thread.c gdb/alpha-osf1-tdep.c gdb/alpha-tdep.c gdb/arc-tdep.c gdb/arch-utils.c gdb/arch-utils.h gdb/arm-linux-tdep.c gdb/arm-tdep.c gdb/avr-tdep.c gdb/ax-general.c gdb/bcache.c gdb/bcache.h gdb/blockframe.c gdb/breakpoint.c gdb/breakpoint.h gdb/buildsym.c gdb/buildsym.h gdb/c-exp.y gdb/c-lang.c gdb/c-typeprint.c gdb/ch-exp.c gdb/ch-lang.c gdb/ch-lang.h gdb/ch-typeprint.c gdb/ch-valprint.c gdb/cli/cli-cmds.c gdb/cli/cli-decode.c gdb/cli/cli-decode.h gdb/cli/cli-dump.c gdb/cli/cli-script.c gdb/cli/cli-setshow.c gdb/coffread.c gdb/command.h gdb/config.in gdb/config/alpha/alpha.mt gdb/config/alpha/nm-linux.h gdb/config/alpha/tm-alphalinux.h gdb/config/arc/arc.mt gdb/config/arc/tm-arc.h gdb/config/arm/nm-linux.h gdb/config/arm/tm-linux.h gdb/config/arm/xm-nbsd.h gdb/config/d30v/d30v.mt gdb/config/d30v/tm-d30v.h gdb/config/djgpp/fnchange.lst gdb/config/fr30/fr30.mt gdb/config/fr30/tm-fr30.h gdb/config/frv/frv.mt gdb/config/frv/tm-frv.h gdb/config/h8300/tm-h8300.h gdb/config/h8500/tm-h8500.h gdb/config/i386/i386aix.mh gdb/config/i386/i386aix.mt gdb/config/i386/i386aout.mt gdb/config/i386/i386gnu.mh gdb/config/i386/i386gnu.mt gdb/config/i386/i386m3.mh gdb/config/i386/i386m3.mt gdb/config/i386/i386mach.mh gdb/config/i386/i386nw.mt gdb/config/i386/i386os9k.mt gdb/config/i386/i386sco5.mt gdb/config/i386/i386sol2.mh gdb/config/i386/i386sol2.mt gdb/config/i386/i386v.mt gdb/config/i386/nm-fbsd.h gdb/config/i386/nm-i386.h gdb/config/i386/nm-i386aix.h gdb/config/i386/nm-i386bsd.h gdb/config/i386/nm-i386gnu.h gdb/config/i386/nm-i386lynx.h gdb/config/i386/nm-i386mach.h gdb/config/i386/nm-i386sco.h gdb/config/i386/nm-i386sco5.h gdb/config/i386/nm-i386sol2.h gdb/config/i386/nm-i386v.h gdb/config/i386/nm-i386v4.h gdb/config/i386/nm-i386v42mp.h gdb/config/i386/nm-linux.h gdb/config/i386/nm-m3.h gdb/config/i386/nm-ptx4.h gdb/config/i386/obsd.mh gdb/config/i386/tm-cygwin.h gdb/config/i386/tm-fbsd.h gdb/config/i386/tm-go32.h gdb/config/i386/tm-i386.h gdb/config/i386/tm-i386aix.h gdb/config/i386/tm-i386lynx.h gdb/config/i386/tm-i386m3.h gdb/config/i386/tm-i386mk.h gdb/config/i386/tm-i386sol2.h gdb/config/i386/tm-i386v4.h gdb/config/i386/tm-linux.h gdb/config/i386/tm-nbsd.h gdb/config/i386/tm-nbsdaout.h gdb/config/i386/tm-ptx.h gdb/config/i386/tm-ptx4.h gdb/config/i386/tm-symmetry.h gdb/config/i386/tm-vxworks.h gdb/config/i386/x86-64linux.mh gdb/config/i386/x86-64linux.mt gdb/config/i386/xm-i386aix.h gdb/config/i386/xm-i386m3.h gdb/config/i386/xm-i386mach.h gdb/config/i386/xm-i386mk.h gdb/config/i386/xm-i386v4.h gdb/config/i386/xm-nbsd.h gdb/config/i386/xm-ptx.h gdb/config/i386/xm-ptx4.h gdb/config/i960/mon960.mt gdb/config/i960/nindy960.mt gdb/config/i960/tm-i960.h gdb/config/i960/tm-mon960.h gdb/config/i960/tm-nindy960.h gdb/config/i960/tm-vx960.h gdb/config/i960/vxworks960.mt gdb/config/ia64/ia64.mt gdb/config/ia64/nm-aix.h gdb/config/ia64/nm-linux.h gdb/config/ia64/tm-aix.h gdb/config/ia64/tm-linux.h gdb/config/ia64/xm-aix.h gdb/config/m32r/tm-m32r.h gdb/config/m68k/apollo68b.mh gdb/config/m68k/apollo68b.mt gdb/config/m68k/apollo68v.mh gdb/config/m68k/hp300bsd.mh gdb/config/m68k/hp300bsd.mt gdb/config/m68k/hp300hpux.mh gdb/config/m68k/hp300hpux.mt gdb/config/m68k/nm-linux.h gdb/config/m68k/nm-m68klynx.h gdb/config/m68k/nm-sysv4.h gdb/config/m68k/tm-delta68.h gdb/config/m68k/tm-linux.h gdb/config/m68k/tm-m68k.h gdb/config/m68k/tm-m68klynx.h gdb/config/m68k/tm-m68kv4.h gdb/config/m68k/tm-sun2os4.h gdb/config/m68k/tm-sun3os4.h gdb/config/m68k/tm-vx68.h gdb/config/m68k/xm-m68kv4.h gdb/config/m68k/xm-nbsd.h gdb/config/m88k/delta88.mh gdb/config/m88k/delta88.mt gdb/config/m88k/delta88v4.mh gdb/config/m88k/delta88v4.mt gdb/config/m88k/m88k.mh gdb/config/m88k/m88k.mt gdb/config/m88k/nm-delta88v4.h gdb/config/m88k/nm-m88k.h gdb/config/m88k/tm-delta88.h gdb/config/m88k/tm-delta88v4.h gdb/config/m88k/tm-m88k.h gdb/config/m88k/xm-delta88.h gdb/config/m88k/xm-delta88v4.h gdb/config/m88k/xm-dgux.h gdb/config/mcore/mcore.mt gdb/config/mips/irix5.mt gdb/config/mips/irix6.mh gdb/config/mips/irix6.mt gdb/config/mips/nm-irix5.h gdb/config/mips/nm-linux.h gdb/config/mips/tm-embed.h gdb/config/mips/tm-irix5.h gdb/config/mips/tm-irix6.h gdb/config/mips/tm-linux.h gdb/config/mips/tm-mips.h gdb/config/mips/tm-mips64.h gdb/config/mips/tm-mipsm3.h gdb/config/mips/tm-mipsv4.h gdb/config/mips/tm-nbsd.h gdb/config/mips/tm-vxmips.h gdb/config/mips/xm-irix5.h gdb/config/mips/xm-mipsv4.h gdb/config/mn10200/tm-mn10200.h gdb/config/ns32k/xm-nbsd.h gdb/config/pa/hppa.mt gdb/config/pa/nm-hppao.h gdb/config/pa/tm-hppa.h gdb/config/pa/tm-hppa64.h gdb/config/powerpc/aix.mt gdb/config/powerpc/linux.mh gdb/config/powerpc/linux.mt gdb/config/powerpc/nbsd.mh gdb/config/powerpc/nbsd.mt gdb/config/powerpc/nm-linux.h gdb/config/powerpc/ppc-eabi.mt gdb/config/powerpc/ppc-sim.mt gdb/config/powerpc/ppcle-eabi.mt gdb/config/powerpc/ppcle-sim.mt gdb/config/powerpc/tm-linux.h gdb/config/powerpc/tm-vxworks.h gdb/config/powerpc/vxworks.mt gdb/config/powerpc/xm-aix.h gdb/config/rs6000/aix4.mt gdb/config/rs6000/nm-rs6000ly.h gdb/config/rs6000/rs6000.mt gdb/config/rs6000/rs6000lynx.mt gdb/config/rs6000/tm-rs6000.h gdb/config/rs6000/tm-rs6000ly.h gdb/config/rs6000/xm-aix4.h gdb/config/sh/tm-linux.h gdb/config/sparc/nbsdaout.mh gdb/config/sparc/nbsdelf.mh gdb/config/sparc/nm-linux.h gdb/config/sparc/nm-nbsd.h gdb/config/sparc/nm-sparclynx.h gdb/config/sparc/nm-sun4sol2.h gdb/config/sparc/tm-linux.h gdb/config/sparc/tm-nbsd.h gdb/config/sparc/tm-sp64.h gdb/config/sparc/tm-sp64linux.h gdb/config/sparc/tm-sp64sim.h gdb/config/sparc/tm-sparc.h gdb/config/sparc/tm-sparclet.h gdb/config/sparc/tm-sparclynx.h gdb/config/sparc/tm-sun4os4.h gdb/config/sparc/tm-sun4sol2.h gdb/config/sparc/tm-vxsparc.h gdb/config/sparc/xm-sun4sol2.h gdb/config/tm-linux.h gdb/config/vax/nm-vax.h gdb/config/vax/tm-vax.h gdb/config/vax/vax.mt gdb/config/z8k/tm-z8k.h gdb/configure gdb/configure.host gdb/configure.in gdb/configure.tgt gdb/corefile.c gdb/cp-valprint.c gdb/cpu32bug-rom.c gdb/cris-tdep.c gdb/cxux-nat.c gdb/d10v-tdep.c gdb/d30v-tdep.c gdb/dbug-rom.c gdb/dbxread.c gdb/defs.h gdb/doc/ChangeLog gdb/doc/gdb.texinfo gdb/doc/gdbint.texinfo gdb/doc/stabs.texinfo gdb/doublest.c gdb/dst.h gdb/dstread.c gdb/dwarf2cfi.c gdb/dwarf2read.c gdb/dwarfread.c gdb/elfread.c gdb/eval.c gdb/event-top.c gdb/expprint.c gdb/expression.h gdb/f-exp.y gdb/f-typeprint.c gdb/fbsd-proc.c gdb/findvar.c gdb/fr30-tdep.c gdb/frame.c gdb/frame.h gdb/frv-tdep.c gdb/gcore.c gdb/gdb-events.c gdb/gdb-events.h gdb/gdb-events.sh gdb/gdb_indent.sh gdb/gdbarch.c gdb/gdbarch.h gdb/gdbarch.sh gdb/gdbinit.in gdb/gdbserver/ChangeLog gdb/gdbserver/Makefile.in gdb/gdbserver/acconfig.h gdb/gdbserver/aclocal.m4 gdb/gdbserver/config.in gdb/gdbserver/configure gdb/gdbserver/configure.in gdb/gdbserver/configure.srv gdb/gdbserver/gdbreplay.c gdb/gdbserver/i387-fp.c gdb/gdbserver/i387-fp.h gdb/gdbserver/inferiors.c gdb/gdbserver/linux-arm-low.c gdb/gdbserver/linux-i386-low.c gdb/gdbserver/linux-low.c gdb/gdbserver/linux-low.h gdb/gdbserver/linux-mips-low.c gdb/gdbserver/linux-ppc-low.c gdb/gdbserver/linux-sh-low.c gdb/gdbserver/linux-x86-64-low.c gdb/gdbserver/regcache.c gdb/gdbserver/regcache.h gdb/gdbserver/remote-utils.c gdb/gdbserver/server.c gdb/gdbserver/server.h gdb/gdbserver/target.c gdb/gdbserver/target.h gdb/gdbtypes.c gdb/gdbtypes.h gdb/gnu-v2-abi.c gdb/gnu-v3-abi.c gdb/h8300-tdep.c gdb/h8500-tdep.c gdb/hpread.c gdb/i386-linux-nat.c gdb/i386-linux-tdep.c gdb/i386-nat.c gdb/i386-sol2-tdep.c gdb/i386-tdep.c gdb/i386-tdep.h gdb/i386aix-nat.c gdb/i386b-nat.c gdb/i386bsd-nat.c gdb/i386bsd-tdep.c gdb/i386fbsd-nat.c gdb/i386gnu-nat.c gdb/i386gnu-tdep.c gdb/i386ly-tdep.c gdb/i386m3-nat.c gdb/i386mach-nat.c gdb/i386nbsd-tdep.c gdb/i386v-nat.c gdb/i386v4-nat.c gdb/i387-tdep.c gdb/i387-tdep.h gdb/i960-tdep.c gdb/ia64-tdep.c gdb/infcmd.c gdb/inferior.h gdb/infrun.c gdb/irix5-nat.c gdb/jv-exp.y gdb/jv-lang.c gdb/language.c gdb/language.h gdb/linux-proc.c gdb/m2-exp.y gdb/m32r-tdep.c gdb/m68hc11-tdep.c gdb/m68k-tdep.c gdb/m68klinux-nat.c gdb/m88k-nat.c gdb/m88k-tdep.c gdb/macrocmd.c gdb/macroexp.c gdb/macroscope.c gdb/macrotab.c gdb/macrotab.h gdb/main.c gdb/maint.c gdb/mcore-tdep.c gdb/mdebugread.c gdb/memattr.c gdb/mi/ChangeLog gdb/mi/gdbmi.texinfo gdb/mi/mi-cmd-disas.c gdb/mi/mi-cmd-var.c gdb/minsyms.c gdb/mips-linux-tdep.c gdb/mips-tdep.c gdb/mn10200-tdep.c gdb/mn10300-tdep.c gdb/mon960-rom.c gdb/monitor.c gdb/monitor.h gdb/ns32k-tdep.c gdb/objfiles.c gdb/objfiles.h gdb/os9kread.c gdb/osabi.c gdb/osabi.h gdb/p-exp.y gdb/p-typeprint.c gdb/p-valprint.c gdb/pa64solib.c gdb/parse.c gdb/parser-defs.h gdb/ppc-linux-nat.c gdb/ppc-linux-tdep.c gdb/ppc-tdep.h gdb/ppcnbsd-nat.c gdb/ppcnbsd-tdep.c gdb/printcmd.c gdb/proc-api.c gdb/procfs.c gdb/rdi-share/devsw.c gdb/rdi-share/host.h gdb/rdi-share/hostchan.c gdb/rdi-share/hostchan.h gdb/rdi-share/serdrv.c gdb/rdi-share/serpardr.c gdb/rdi-share/unixcomm.c gdb/regcache.c gdb/regcache.h gdb/remote-bug.c gdb/remote-es.c gdb/remote-est.c gdb/remote-nindy.c gdb/remote-nrom.c gdb/remote-os9k.c gdb/remote-rdi.c gdb/remote-rdp.c gdb/remote-sim.c gdb/remote-vx960.c gdb/remote.c gdb/rom68k-rom.c gdb/rs6000-nat.c gdb/rs6000-tdep.c gdb/s390-nat.c gdb/s390-tdep.c gdb/ser-e7kpc.c gdb/sh-tdep.c gdb/signals/signals.c gdb/solib-osf.c gdb/solib-svr4.c gdb/somsolib.c gdb/source.c gdb/sparc-tdep.c gdb/sparcl-tdep.c gdb/stabsread.c gdb/stabsread.h gdb/stack.c gdb/std-regs.c gdb/symfile.c gdb/symmisc.c gdb/symtab.c gdb/symtab.h gdb/target.c gdb/target.h gdb/testsuite/ChangeLog gdb/testsuite/Makefile.in gdb/testsuite/config/gdbserver.exp gdb/testsuite/configure gdb/testsuite/configure.in gdb/testsuite/gdb.arch/altivec-abi.exp gdb/testsuite/gdb.asm/asm-source.exp gdb/testsuite/gdb.base/attach.exp gdb/testsuite/gdb.base/call-ar-st.exp gdb/testsuite/gdb.base/d10v.ld gdb/testsuite/gdb.base/ending-run.exp gdb/testsuite/gdb.base/interrupt.exp gdb/testsuite/gdb.base/long_long.exp gdb/testsuite/gdb.base/overlays.exp gdb/testsuite/gdb.base/page.exp gdb/testsuite/gdb.base/readline.exp gdb/testsuite/gdb.base/return.exp gdb/testsuite/gdb.base/selftest.exp gdb/testsuite/gdb.base/shlib-call.exp gdb/testsuite/gdb.chill/ChangeLog gdb/testsuite/gdb.chill/builtins.exp gdb/testsuite/gdb.chill/callch.exp gdb/testsuite/gdb.chill/chillvars.exp gdb/testsuite/gdb.chill/enum.exp gdb/testsuite/gdb.chill/gch1041.exp gdb/testsuite/gdb.chill/gch1272.exp gdb/testsuite/gdb.chill/gch1280.exp gdb/testsuite/gdb.chill/gch922.exp gdb/testsuite/gdb.chill/gch981.exp gdb/testsuite/gdb.chill/misc.exp gdb/testsuite/gdb.chill/powerset.exp gdb/testsuite/gdb.chill/pr-4975.exp gdb/testsuite/gdb.chill/pr-5016.exp gdb/testsuite/gdb.chill/pr-5020.exp gdb/testsuite/gdb.chill/pr-5022.exp gdb/testsuite/gdb.chill/pr-5646.exp gdb/testsuite/gdb.chill/pr-5984.exp gdb/testsuite/gdb.chill/pr-6292.exp gdb/testsuite/gdb.chill/pr-6632.exp gdb/testsuite/gdb.chill/pr-8134.exp gdb/testsuite/gdb.chill/pr-8136.exp gdb/testsuite/gdb.chill/pr-8405.exp gdb/testsuite/gdb.chill/pr-8742.exp gdb/testsuite/gdb.chill/pr-8894.exp gdb/testsuite/gdb.chill/pr-9095.exp gdb/testsuite/gdb.chill/pr-9946.exp gdb/testsuite/gdb.chill/result.exp gdb/testsuite/gdb.chill/string.exp gdb/testsuite/gdb.chill/tests1.exp gdb/testsuite/gdb.chill/tests2.exp gdb/testsuite/gdb.chill/tuples.exp gdb/testsuite/gdb.chill/xstruct.exp gdb/testsuite/gdb.fortran/exprs.exp gdb/testsuite/gdb.fortran/types.exp gdb/testsuite/gdb.hp/configure gdb/testsuite/gdb.hp/configure.in gdb/testsuite/gdb.java/jv-print.exp gdb/testsuite/gdb.threads/print-threads.c gdb/testsuite/gdb.threads/print-threads.exp gdb/testsuite/gdb.threads/schedlock.c gdb/testsuite/gdb.threads/schedlock.exp gdb/testsuite/lib/gdb.exp gdb/thread.c gdb/top.c gdb/tracepoint.c gdb/tui/ChangeLog gdb/tui/tui.c gdb/typeprint.c gdb/ui-file.c gdb/ui-file.h gdb/ui-out.c gdb/utils.c gdb/uw-thread.c gdb/v850-tdep.c gdb/valarith.c gdb/valops.c gdb/valprint.c gdb/value.h gdb/values.c gdb/varobj.c gdb/vax-tdep.c gdb/vax-tdep.h gdb/version.in gdb/win32-nat.c gdb/wince.c gdb/x86-64-linux-nat.c gdb/x86-64-tdep.c gdb/x86-64-tdep.h gdb/xcoffread.c gdb/xstormy16-tdep.c gdb/z8k-tdep.c include/ChangeLog include/bfdlink.h include/demangle.h include/dis-asm.h include/elf/ChangeLog include/elf/alpha.h include/elf/common.h include/elf/dwarf2.h include/elf/i370.h include/elf/m68hc11.h include/elf/sh.h include/elf/vax.h include/fibheap.h include/gdb/ChangeLog include/gdb/remote-sim.h include/gdb/sim-d10v.h include/gdb/sim-sh.h include/getopt.h include/hashtab.h include/libiberty.h include/opcode/ChangeLog include/opcode/a29k.h include/opcode/convex.h include/opcode/dlx.h include/opcode/i386.h include/opcode/m68hc11.h include/opcode/mips.h include/opcode/or32.h include/opcode/ppc.h include/partition.h include/sort.h include/splay-tree.h libiberty/ChangeLog libiberty/Makefile.in libiberty/configure libiberty/configure.in libiberty/cp-demangle.c libiberty/cplus-dem.c libiberty/hashtab.c libiberty/lbasename.c libiberty/regex.c libiberty/testsuite/demangle-expected ltcf-cxx.sh opcodes/ChangeLog opcodes/Makefile.am opcodes/Makefile.in opcodes/a29k-dis.c opcodes/configure opcodes/configure.in opcodes/d10v-opc.c opcodes/disassemble.c opcodes/h8300-dis.c opcodes/ia64-asmtab.c opcodes/ia64-opc-b.c opcodes/m68hc11-dis.c opcodes/m68hc11-opc.c opcodes/m68k-dis.c opcodes/mips-dis.c opcodes/mips-opc.c opcodes/or32-dis.c opcodes/po/POTFILES.in opcodes/po/es.po opcodes/po/fr.po opcodes/po/id.po opcodes/po/sv.po opcodes/po/tr.po opcodes/ppc-dis.c opcodes/ppc-opc.c opcodes/sparc-dis.c opcodes/z8k-dis.c opcodes/z8k-opc.h opcodes/z8kgen.c readline/ChangeLog.gdb readline/doc/rluserman.texinfo readline/examples/rlfe.c readline/rlprivate.h readline/rlshell.h readline/support/config.guess readline/support/config.sub readline/xmalloc.h sim/ChangeLog sim/Makefile.in sim/README-HACKING sim/arm/ChangeLog sim/arm/Makefile.in sim/arm/armcopro.c sim/arm/armemu.c sim/arm/armos.c sim/arm/armos.h sim/arm/configure sim/arm/wrapper.c sim/common/ChangeLog sim/common/Make-common.in sim/common/aclocal.m4 sim/common/callback.c sim/common/gennltvals.sh sim/common/gentmap.c sim/common/hw-events.c sim/common/nltvals.def sim/common/run-sim.h sim/common/run.c sim/common/sim-basics.h sim/common/sim-events.c sim/common/sim-fpu.c sim/common/sim-load.c sim/common/sim-memopt.c sim/common/sim-resume.c sim/common/syscall.c sim/configure sim/configure.in sim/d10v/ChangeLog sim/d10v/Makefile.in sim/d10v/configure sim/d10v/d10v_sim.h sim/d10v/gencode.c sim/d10v/interp.c sim/d30v/ChangeLog sim/d30v/Makefile.in sim/d30v/alu.h sim/d30v/configure sim/d30v/cpu.c sim/d30v/cpu.h sim/d30v/d30v-insns sim/d30v/dc-short sim/d30v/engine.c sim/d30v/ic-d30v sim/d30v/sim-calls.c sim/d30v/sim-main.h sim/erc32/ChangeLog sim/erc32/interf.c sim/erc32/sis.h sim/fr30/ChangeLog sim/fr30/Makefile.in sim/fr30/README sim/fr30/TODO sim/fr30/arch.c sim/fr30/arch.h sim/fr30/configure sim/fr30/configure.in sim/fr30/cpu.c sim/fr30/cpu.h sim/fr30/cpuall.h sim/fr30/decode.c sim/fr30/decode.h sim/fr30/devices.c sim/fr30/fr30-sim.h sim/fr30/fr30.c sim/fr30/mloop.in sim/fr30/model.c sim/fr30/sem-switch.c sim/fr30/sem.c sim/fr30/sim-if.c sim/fr30/sim-main.h sim/fr30/traps.c sim/h8300/ChangeLog sim/h8300/Makefile.in sim/h8300/compile.c sim/h8500/ChangeLog sim/h8500/compile.c sim/i960/ChangeLog sim/igen/ChangeLog sim/igen/Makefile.in sim/igen/compare_igen_models sim/igen/configure sim/igen/gen-engine.c sim/igen/gen-icache.c sim/igen/gen.c sim/igen/igen.c sim/igen/lf.c sim/m32r/ChangeLog sim/m68hc11/ChangeLog sim/m68hc11/Makefile.in sim/m68hc11/configure sim/m68hc11/dv-m68hc11.c sim/m68hc11/dv-m68hc11eepr.c sim/m68hc11/gencode.c sim/m68hc11/interp.c sim/m68hc11/m68hc11_sim.c sim/m68hc11/sim-main.h sim/mcore/ChangeLog sim/mcore/interp.c sim/mips/ChangeLog sim/mips/Makefile.in sim/mips/configure sim/mips/configure.in sim/mips/cp1.c sim/mips/interp.c sim/mips/mips.igen sim/mips/sim-main.c sim/mips/sim-main.h sim/mn10200/ChangeLog sim/mn10200/Makefile.in sim/mn10200/mn10200_sim.h sim/mn10300/ChangeLog sim/mn10300/Makefile.in sim/mn10300/configure sim/mn10300/mn10300_sim.h sim/mn10300/tconfig.in sim/ppc/ChangeLog sim/ppc/Makefile.in sim/ppc/gen-idecode.c sim/ppc/gen-model.c sim/ppc/lf.c sim/ppc/main.c sim/ppc/psim.c sim/ppc/sim_calls.c sim/sh/ChangeLog sim/sh/Makefile.in sim/sh/interp.c sim/testsuite/ChangeLog sim/v850/ChangeLog sim/v850/configure sim/v850/simops.c sim/z8k/ChangeLog sim/z8k/Makefile.in sim/z8k/iface.c sim/z8k/support.c sim/z8k/writecode.c texinfo/texinfo.tex Cherrypick from FSF 2002-08-23 22:02:32 UTC Elena Zannoni <ezannoni@kwikemart.cygnus.com> 'import of readline-4.3': readline/examples/excallback.c Delete: gdb/29k-share/README gdb/29k-share/udi/udi2go32.c gdb/29k-share/udi/udiids.h gdb/29k-share/udi/udip2soc.c gdb/29k-share/udi/udiphcfg.h gdb/29k-share/udi/udiphunix.h gdb/29k-share/udi/udiproc.h gdb/29k-share/udi/udipt29k.h gdb/29k-share/udi/udiptcfg.h gdb/29k-share/udi/udisoc.h gdb/29k-share/udi/udr.c gdb/29k-share/udi_soc gdb/ada-exp.tab.c gdb/ada-lex.c gdb/cli/cli-interp.c gdb/config/i386/nm-gnu.h gdb/config/i386/nm-x86-64.h gdb/config/i386/obsd.mt gdb/config/i386/tm-i386gnu.h gdb/config/i386/tm-i386nw.h gdb/config/i386/tm-i386sco5.h gdb/config/i386/tm-i386v.h gdb/config/i386/tm-obsd.h gdb/config/i386/xm-i386gnu.h gdb/config/mcore/tm-mcore.h gdb/config/romp/rtbsd.mh gdb/config/romp/xm-rtbsd.h gdb/config/sparc/nbsdaout.mt gdb/config/sparc/nbsdelf.mt gdb/config/sparc/tm-nbsdaout.h gdb/config/sparc/xm-nbsd.h gdb/interps.c gdb/interps.h gdb/mi/mi-events.c gdb/mi/mi-interp.c gdb/mi/mi.h gdb/remote-vx29k.c gdb/testsuite/gdb.hp/gdb.threads-hp/Makefile.in gdb/testsuite/gdb.hp/gdb.threads-hp/configure gdb/testsuite/gdb.hp/gdb.threads-hp/configure.in include/callback.h include/remote-sim.h sim/tic80/ChangeLog sim/tic80/Makefile.in sim/tic80/acconfig.h sim/tic80/alu.h sim/tic80/config.in sim/tic80/configure sim/tic80/configure.in sim/tic80/cpu.h sim/tic80/interp.c sim/tic80/misc.c sim/tic80/sim-calls.c sim/tic80/sim-main.h sim/tic80/tic80.dc sim/tic80/tic80.ic sim/tic80/tic80.igen sim/w65/ChangeLog sim/w65/Makefile.in sim/w65/acconfig.h sim/w65/config.in sim/w65/configure sim/w65/configure.in sim/w65/gencode.c sim/w65/interp.c sim/w65/interp.h sim/w65/run.c sim/z8k/list.c
Diffstat (limited to 'gdb/ch-exp.c')
-rw-r--r--gdb/ch-exp.c4466
1 files changed, 2233 insertions, 2233 deletions
diff --git a/gdb/ch-exp.c b/gdb/ch-exp.c
index e96a6f8..d588ec2 100644
--- a/gdb/ch-exp.c
+++ b/gdb/ch-exp.c
@@ -1,2233 +1,2233 @@
-/* Parser for GNU CHILL (CCITT High-Level Language) -*- C -*-
- Copyright 1992, 1993, 1995, 1996, 1997, 1999, 2000, 2001
- Free Software Foundation, Inc.
-
- This file is part of GDB.
-
- This program is free software; you can redistribute it and/or modify
- it under the terms of the GNU General Public License as published by
- the Free Software Foundation; either version 2 of the License, or
- (at your option) any later version.
-
- This program is distributed in the hope that it will be useful,
- but WITHOUT ANY WARRANTY; without even the implied warranty of
- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- GNU General Public License for more details.
-
- You should have received a copy of the GNU General Public License
- along with this program; if not, write to the Free Software
- Foundation, Inc., 59 Temple Place - Suite 330,
- Boston, MA 02111-1307, USA. */
-
-/* Parse a Chill expression from text in a string,
- and return the result as a struct expression pointer.
- That structure contains arithmetic operations in reverse polish,
- with constants represented by operations that are followed by special data.
- See expression.h for the details of the format.
- What is important here is that it can be built up sequentially
- during the process of parsing; the lower levels of the tree always
- come first in the result.
-
- Note that the language accepted by this parser is more liberal
- than the one accepted by an actual Chill compiler. For example, the
- language rule that a simple name string can not be one of the reserved
- simple name strings is not enforced (e.g "case" is not treated as a
- reserved name). Another example is that Chill is a strongly typed
- language, and certain expressions that violate the type constraints
- may still be evaluated if gdb can do so in a meaningful manner, while
- such expressions would be rejected by the compiler. The reason for
- this more liberal behavior is the philosophy that the debugger
- is intended to be a tool that is used by the programmer when things
- go wrong, and as such, it should provide as few artificial barriers
- to it's use as possible. If it can do something meaningful, even
- something that violates language contraints that are enforced by the
- compiler, it should do so without complaint.
-
- */
-
-#include "defs.h"
-#include "gdb_string.h"
-#include <ctype.h>
-#include "expression.h"
-#include "language.h"
-#include "value.h"
-#include "parser-defs.h"
-#include "ch-lang.h"
-#include "bfd.h" /* Required by objfiles.h. */
-#include "symfile.h" /* Required by objfiles.h. */
-#include "objfiles.h" /* For have_full_symbols and have_partial_symbols */
-
-#ifdef __GNUC__
-#define INLINE __inline__
-#endif
-
-typedef union
-
- {
- LONGEST lval;
- ULONGEST ulval;
- struct
- {
- LONGEST val;
- struct type *type;
- }
- typed_val;
- double dval;
- struct symbol *sym;
- struct type *tval;
- struct stoken sval;
- struct ttype tsym;
- struct symtoken ssym;
- }
-YYSTYPE;
-
-enum ch_terminal
- {
- END_TOKEN = 0,
- /* '\001' ... '\xff' come first. */
- OPEN_PAREN = '(',
- TOKEN_NOT_READ = 999,
- INTEGER_LITERAL,
- BOOLEAN_LITERAL,
- CHARACTER_LITERAL,
- FLOAT_LITERAL,
- GENERAL_PROCEDURE_NAME,
- LOCATION_NAME,
- EMPTINESS_LITERAL,
- CHARACTER_STRING_LITERAL,
- BIT_STRING_LITERAL,
- TYPENAME,
- DOT_FIELD_NAME, /* '.' followed by <field name> */
- CASE,
- OF,
- ESAC,
- LOGIOR,
- ORIF,
- LOGXOR,
- LOGAND,
- ANDIF,
- NOTEQUAL,
- GEQ,
- LEQ,
- IN,
- SLASH_SLASH,
- MOD,
- REM,
- NOT,
- POINTER,
- RECEIVE,
- UP,
- IF,
- THEN,
- ELSE,
- FI,
- ELSIF,
- ILLEGAL_TOKEN,
- NUM,
- PRED,
- SUCC,
- ABS,
- CARD,
- MAX_TOKEN,
- MIN_TOKEN,
- ADDR_TOKEN,
- SIZE,
- UPPER,
- LOWER,
- LENGTH,
- ARRAY,
- GDB_VARIABLE,
- GDB_ASSIGNMENT
- };
-
-/* Forward declarations. */
-
-static void write_lower_upper_value (enum exp_opcode, struct type *);
-static enum ch_terminal match_bitstring_literal (void);
-static enum ch_terminal match_integer_literal (void);
-static enum ch_terminal match_character_literal (void);
-static enum ch_terminal match_string_literal (void);
-static enum ch_terminal match_float_literal (void);
-static int decode_integer_literal (LONGEST *, char **);
-static int decode_integer_value (int, char **, LONGEST *);
-static char *match_simple_name_string (void);
-static void growbuf_by_size (int);
-static void parse_case_label (void);
-static void parse_untyped_expr (void);
-static void parse_if_expression (void);
-static void parse_if_expression_body (void);
-static void parse_else_alternative (void);
-static void parse_then_alternative (void);
-static void parse_expr (void);
-static void parse_operand0 (void);
-static void parse_operand1 (void);
-static void parse_operand2 (void);
-static void parse_operand3 (void);
-static void parse_operand4 (void);
-static void parse_operand5 (void);
-static void parse_operand6 (void);
-static void parse_primval (void);
-static void parse_tuple (struct type *);
-static void parse_opt_element_list (struct type *);
-static void parse_tuple_element (struct type *);
-static void parse_named_record_element (void);
-static void parse_call (void);
-static struct type *parse_mode_or_normal_call (void);
-#if 0
-static struct type *parse_mode_call (void);
-#endif
-static void parse_unary_call (void);
-static int parse_opt_untyped_expr (void);
-static int expect (enum ch_terminal, char *);
-static enum ch_terminal ch_lex (void);
-INLINE static enum ch_terminal PEEK_TOKEN (void);
-static enum ch_terminal peek_token_ (int);
-static void forward_token_ (void);
-static void require (enum ch_terminal);
-static int check_token (enum ch_terminal);
-
-#define MAX_LOOK_AHEAD 2
-static enum ch_terminal terminal_buffer[MAX_LOOK_AHEAD + 1] =
-{
- TOKEN_NOT_READ, TOKEN_NOT_READ, TOKEN_NOT_READ};
-static YYSTYPE yylval;
-static YYSTYPE val_buffer[MAX_LOOK_AHEAD + 1];
-
-/*int current_token, lookahead_token; */
-
-INLINE static enum ch_terminal
-PEEK_TOKEN (void)
-{
- if (terminal_buffer[0] == TOKEN_NOT_READ)
- {
- terminal_buffer[0] = ch_lex ();
- val_buffer[0] = yylval;
- }
- return terminal_buffer[0];
-}
-#define PEEK_LVAL() val_buffer[0]
-#define PEEK_TOKEN1() peek_token_(1)
-#define PEEK_TOKEN2() peek_token_(2)
-static enum ch_terminal
-peek_token_ (int i)
-{
- if (i > MAX_LOOK_AHEAD)
- internal_error (__FILE__, __LINE__,
- "too much lookahead");
- if (terminal_buffer[i] == TOKEN_NOT_READ)
- {
- terminal_buffer[i] = ch_lex ();
- val_buffer[i] = yylval;
- }
- return terminal_buffer[i];
-}
-
-#if 0
-
-static void
-pushback_token (enum ch_terminal code, YYSTYPE node)
-{
- int i;
- if (terminal_buffer[MAX_LOOK_AHEAD] != TOKEN_NOT_READ)
- internal_error (__FILE__, __LINE__,
- "cannot pushback token");
- for (i = MAX_LOOK_AHEAD; i > 0; i--)
- {
- terminal_buffer[i] = terminal_buffer[i - 1];
- val_buffer[i] = val_buffer[i - 1];
- }
- terminal_buffer[0] = code;
- val_buffer[0] = node;
-}
-
-#endif
-
-static void
-forward_token_ (void)
-{
- int i;
- for (i = 0; i < MAX_LOOK_AHEAD; i++)
- {
- terminal_buffer[i] = terminal_buffer[i + 1];
- val_buffer[i] = val_buffer[i + 1];
- }
- terminal_buffer[MAX_LOOK_AHEAD] = TOKEN_NOT_READ;
-}
-#define FORWARD_TOKEN() forward_token_()
-
-/* Skip the next token.
- if it isn't TOKEN, the parser is broken. */
-
-static void
-require (enum ch_terminal token)
-{
- if (PEEK_TOKEN () != token)
- {
- internal_error (__FILE__, __LINE__,
- "expected token %d", (int) token);
- }
- FORWARD_TOKEN ();
-}
-
-static int
-check_token (enum ch_terminal token)
-{
- if (PEEK_TOKEN () != token)
- return 0;
- FORWARD_TOKEN ();
- return 1;
-}
-
-/* return 0 if expected token was not found,
- else return 1.
- */
-static int
-expect (enum ch_terminal token, char *message)
-{
- if (PEEK_TOKEN () != token)
- {
- if (message)
- error (message);
- else if (token < 256)
- error ("syntax error - expected a '%c' here \"%s\"", token, lexptr);
- else
- error ("syntax error");
- return 0;
- }
- else
- FORWARD_TOKEN ();
- return 1;
-}
-
-#if 0
-/* Parse a name string. If ALLOW_ALL is 1, ALL is allowed as a postfix. */
-
-static tree
-parse_opt_name_string (int allow_all)
-{
- int token = PEEK_TOKEN ();
- tree name;
- if (token != NAME)
- {
- if (token == ALL && allow_all)
- {
- FORWARD_TOKEN ();
- return ALL_POSTFIX;
- }
- return NULL_TREE;
- }
- name = PEEK_LVAL ();
- for (;;)
- {
- FORWARD_TOKEN ();
- token = PEEK_TOKEN ();
- if (token != '!')
- return name;
- FORWARD_TOKEN ();
- token = PEEK_TOKEN ();
- if (token == ALL && allow_all)
- return get_identifier3 (IDENTIFIER_POINTER (name), "!", "*");
- if (token != NAME)
- {
- if (pass == 1)
- error ("'%s!' is not followed by an identifier",
- IDENTIFIER_POINTER (name));
- return name;
- }
- name = get_identifier3 (IDENTIFIER_POINTER (name),
- "!", IDENTIFIER_POINTER (PEEK_LVAL ()));
- }
-}
-
-static tree
-parse_simple_name_string (void)
-{
- int token = PEEK_TOKEN ();
- tree name;
- if (token != NAME)
- {
- error ("expected a name here");
- return error_mark_node;
- }
- name = PEEK_LVAL ();
- FORWARD_TOKEN ();
- return name;
-}
-
-static tree
-parse_name_string (void)
-{
- tree name = parse_opt_name_string (0);
- if (name)
- return name;
- if (pass == 1)
- error ("expected a name string here");
- return error_mark_node;
-}
-
-/* Matches: <name_string>
- Returns if pass 1: the identifier.
- Returns if pass 2: a decl or value for identifier. */
-
-static tree
-parse_name (void)
-{
- tree name = parse_name_string ();
- if (pass == 1 || ignoring)
- return name;
- else
- {
- tree decl = lookup_name (name);
- if (decl == NULL_TREE)
- {
- error ("`%s' undeclared", IDENTIFIER_POINTER (name));
- return error_mark_node;
- }
- else if (TREE_CODE (TREE_TYPE (decl)) == ERROR_MARK)
- return error_mark_node;
- else if (TREE_CODE (decl) == CONST_DECL)
- return DECL_INITIAL (decl);
- else if (TREE_CODE (TREE_TYPE (decl)) == REFERENCE_TYPE)
- return convert_from_reference (decl);
- else
- return decl;
- }
-}
-#endif
-
-#if 0
-static void
-pushback_paren_expr (tree expr)
-{
- if (pass == 1 && !ignoring)
- expr = build1 (PAREN_EXPR, NULL_TREE, expr);
- pushback_token (EXPR, expr);
-}
-#endif
-
-/* Matches: <case label> */
-
-static void
-parse_case_label (void)
-{
- if (check_token (ELSE))
- error ("ELSE in tuples labels not implemented");
- /* Does not handle the case of a mode name. FIXME */
- parse_expr ();
- if (check_token (':'))
- {
- parse_expr ();
- write_exp_elt_opcode (BINOP_RANGE);
- }
-}
-
-static int
-parse_opt_untyped_expr (void)
-{
- switch (PEEK_TOKEN ())
- {
- case ',':
- case ':':
- case ')':
- return 0;
- default:
- parse_untyped_expr ();
- return 1;
- }
-}
-
-static void
-parse_unary_call (void)
-{
- FORWARD_TOKEN ();
- expect ('(', NULL);
- parse_expr ();
- expect (')', NULL);
-}
-
-/* Parse NAME '(' MODENAME ')'. */
-
-#if 0
-
-static struct type *
-parse_mode_call (void)
-{
- struct type *type;
- FORWARD_TOKEN ();
- expect ('(', NULL);
- if (PEEK_TOKEN () != TYPENAME)
- error ("expect MODENAME here `%s'", lexptr);
- type = PEEK_LVAL ().tsym.type;
- FORWARD_TOKEN ();
- expect (')', NULL);
- return type;
-}
-
-#endif
-
-static struct type *
-parse_mode_or_normal_call (void)
-{
- struct type *type;
- FORWARD_TOKEN ();
- expect ('(', NULL);
- if (PEEK_TOKEN () == TYPENAME)
- {
- type = PEEK_LVAL ().tsym.type;
- FORWARD_TOKEN ();
- }
- else
- {
- parse_expr ();
- type = NULL;
- }
- expect (')', NULL);
- return type;
-}
-
-/* Parse something that looks like a function call.
- Assume we have parsed the function, and are at the '('. */
-
-static void
-parse_call (void)
-{
- int arg_count;
- require ('(');
- /* This is to save the value of arglist_len
- being accumulated for each dimension. */
- start_arglist ();
- if (parse_opt_untyped_expr ())
- {
- int tok = PEEK_TOKEN ();
- arglist_len = 1;
- if (tok == UP || tok == ':')
- {
- FORWARD_TOKEN ();
- parse_expr ();
- expect (')', "expected ')' to terminate slice");
- end_arglist ();
- write_exp_elt_opcode (tok == UP ? TERNOP_SLICE_COUNT
- : TERNOP_SLICE);
- return;
- }
- while (check_token (','))
- {
- parse_untyped_expr ();
- arglist_len++;
- }
- }
- else
- arglist_len = 0;
- expect (')', NULL);
- arg_count = end_arglist ();
- write_exp_elt_opcode (MULTI_SUBSCRIPT);
- write_exp_elt_longcst (arg_count);
- write_exp_elt_opcode (MULTI_SUBSCRIPT);
-}
-
-static void
-parse_named_record_element (void)
-{
- struct stoken label;
- char buf[256];
-
- label = PEEK_LVAL ().sval;
- sprintf (buf, "expected a field name here `%s'", lexptr);
- expect (DOT_FIELD_NAME, buf);
- if (check_token (','))
- parse_named_record_element ();
- else if (check_token (':'))
- parse_expr ();
- else
- error ("syntax error near `%s' in named record tuple element", lexptr);
- write_exp_elt_opcode (OP_LABELED);
- write_exp_string (label);
- write_exp_elt_opcode (OP_LABELED);
-}
-
-/* Returns one or more TREE_LIST nodes, in reverse order. */
-
-static void
-parse_tuple_element (struct type *type)
-{
- if (PEEK_TOKEN () == DOT_FIELD_NAME)
- {
- /* Parse a labelled structure tuple. */
- parse_named_record_element ();
- return;
- }
-
- if (check_token ('('))
- {
- if (check_token ('*'))
- {
- expect (')', "missing ')' after '*' case label list");
- if (type)
- {
- if (TYPE_CODE (type) == TYPE_CODE_ARRAY)
- {
- /* do this as a range from low to high */
- struct type *range_type = TYPE_FIELD_TYPE (type, 0);
- LONGEST low_bound, high_bound;
- if (get_discrete_bounds (range_type, &low_bound, &high_bound) < 0)
- error ("cannot determine bounds for (*)");
- /* lower bound */
- write_exp_elt_opcode (OP_LONG);
- write_exp_elt_type (range_type);
- write_exp_elt_longcst (low_bound);
- write_exp_elt_opcode (OP_LONG);
- /* upper bound */
- write_exp_elt_opcode (OP_LONG);
- write_exp_elt_type (range_type);
- write_exp_elt_longcst (high_bound);
- write_exp_elt_opcode (OP_LONG);
- write_exp_elt_opcode (BINOP_RANGE);
- }
- else
- error ("(*) in invalid context");
- }
- else
- error ("(*) only possible with modename in front of tuple (mode[..])");
- }
- else
- {
- parse_case_label ();
- while (check_token (','))
- {
- parse_case_label ();
- write_exp_elt_opcode (BINOP_COMMA);
- }
- expect (')', NULL);
- }
- }
- else
- parse_untyped_expr ();
- if (check_token (':'))
- {
- /* A powerset range or a labeled Array. */
- parse_untyped_expr ();
- write_exp_elt_opcode (BINOP_RANGE);
- }
-}
-
-/* Matches: a COMMA-separated list of tuple elements.
- Returns a list (of TREE_LIST nodes). */
-static void
-parse_opt_element_list (struct type *type)
-{
- arglist_len = 0;
- if (PEEK_TOKEN () == ']')
- return;
- for (;;)
- {
- parse_tuple_element (type);
- arglist_len++;
- if (PEEK_TOKEN () == ']')
- break;
- if (!check_token (','))
- error ("bad syntax in tuple");
- }
-}
-
-/* Parses: '[' elements ']'
- If modename is non-NULL it prefixed the tuple. */
-
-static void
-parse_tuple (struct type *mode)
-{
- struct type *type;
- if (mode)
- type = check_typedef (mode);
- else
- type = 0;
- require ('[');
- start_arglist ();
- parse_opt_element_list (type);
- expect (']', "missing ']' after tuple");
- write_exp_elt_opcode (OP_ARRAY);
- write_exp_elt_longcst ((LONGEST) 0);
- write_exp_elt_longcst ((LONGEST) end_arglist () - 1);
- write_exp_elt_opcode (OP_ARRAY);
- if (type)
- {
- if (TYPE_CODE (type) != TYPE_CODE_ARRAY
- && TYPE_CODE (type) != TYPE_CODE_STRUCT
- && TYPE_CODE (type) != TYPE_CODE_SET)
- error ("invalid tuple mode");
- write_exp_elt_opcode (UNOP_CAST);
- write_exp_elt_type (mode);
- write_exp_elt_opcode (UNOP_CAST);
- }
-}
-
-static void
-parse_primval (void)
-{
- struct type *type;
- enum exp_opcode op;
- char *op_name;
- switch (PEEK_TOKEN ())
- {
- case INTEGER_LITERAL:
- case CHARACTER_LITERAL:
- write_exp_elt_opcode (OP_LONG);
- write_exp_elt_type (PEEK_LVAL ().typed_val.type);
- write_exp_elt_longcst (PEEK_LVAL ().typed_val.val);
- write_exp_elt_opcode (OP_LONG);
- FORWARD_TOKEN ();
- break;
- case BOOLEAN_LITERAL:
- write_exp_elt_opcode (OP_BOOL);
- write_exp_elt_longcst ((LONGEST) PEEK_LVAL ().ulval);
- write_exp_elt_opcode (OP_BOOL);
- FORWARD_TOKEN ();
- break;
- case FLOAT_LITERAL:
- write_exp_elt_opcode (OP_DOUBLE);
- write_exp_elt_type (builtin_type_double);
- write_exp_elt_dblcst (PEEK_LVAL ().dval);
- write_exp_elt_opcode (OP_DOUBLE);
- FORWARD_TOKEN ();
- break;
- case EMPTINESS_LITERAL:
- write_exp_elt_opcode (OP_LONG);
- write_exp_elt_type (lookup_pointer_type (builtin_type_void));
- write_exp_elt_longcst (0);
- write_exp_elt_opcode (OP_LONG);
- FORWARD_TOKEN ();
- break;
- case CHARACTER_STRING_LITERAL:
- write_exp_elt_opcode (OP_STRING);
- write_exp_string (PEEK_LVAL ().sval);
- write_exp_elt_opcode (OP_STRING);
- FORWARD_TOKEN ();
- break;
- case BIT_STRING_LITERAL:
- write_exp_elt_opcode (OP_BITSTRING);
- write_exp_bitstring (PEEK_LVAL ().sval);
- write_exp_elt_opcode (OP_BITSTRING);
- FORWARD_TOKEN ();
- break;
- case ARRAY:
- FORWARD_TOKEN ();
- /* This is pseudo-Chill, similar to C's '(TYPE[])EXPR'
- which casts to an artificial array. */
- expect ('(', NULL);
- expect (')', NULL);
- if (PEEK_TOKEN () != TYPENAME)
- error ("missing MODENAME after ARRAY()");
- type = PEEK_LVAL ().tsym.type;
- FORWARD_TOKEN ();
- expect ('(', NULL);
- parse_expr ();
- expect (')', "missing right parenthesis");
- type = create_array_type ((struct type *) NULL, type,
- create_range_type ((struct type *) NULL,
- builtin_type_int, 0, 0));
- TYPE_ARRAY_UPPER_BOUND_TYPE (type) = BOUND_CANNOT_BE_DETERMINED;
- write_exp_elt_opcode (UNOP_CAST);
- write_exp_elt_type (type);
- write_exp_elt_opcode (UNOP_CAST);
- break;
-#if 0
- case CONST:
- case EXPR:
- val = PEEK_LVAL ();
- FORWARD_TOKEN ();
- break;
-#endif
- case '(':
- FORWARD_TOKEN ();
- parse_expr ();
- expect (')', "missing right parenthesis");
- break;
- case '[':
- parse_tuple (NULL);
- break;
- case GENERAL_PROCEDURE_NAME:
- case LOCATION_NAME:
- write_exp_elt_opcode (OP_VAR_VALUE);
- write_exp_elt_block (NULL);
- write_exp_elt_sym (PEEK_LVAL ().ssym.sym);
- write_exp_elt_opcode (OP_VAR_VALUE);
- FORWARD_TOKEN ();
- break;
- case GDB_VARIABLE: /* gdb specific */
- FORWARD_TOKEN ();
- break;
- case NUM:
- parse_unary_call ();
- write_exp_elt_opcode (UNOP_CAST);
- write_exp_elt_type (builtin_type_int);
- write_exp_elt_opcode (UNOP_CAST);
- break;
- case CARD:
- parse_unary_call ();
- write_exp_elt_opcode (UNOP_CARD);
- break;
- case MAX_TOKEN:
- parse_unary_call ();
- write_exp_elt_opcode (UNOP_CHMAX);
- break;
- case MIN_TOKEN:
- parse_unary_call ();
- write_exp_elt_opcode (UNOP_CHMIN);
- break;
- case PRED:
- op_name = "PRED";
- goto unimplemented_unary_builtin;
- case SUCC:
- op_name = "SUCC";
- goto unimplemented_unary_builtin;
- case ABS:
- op_name = "ABS";
- goto unimplemented_unary_builtin;
- unimplemented_unary_builtin:
- parse_unary_call ();
- error ("not implemented: %s builtin function", op_name);
- break;
- case ADDR_TOKEN:
- parse_unary_call ();
- write_exp_elt_opcode (UNOP_ADDR);
- break;
- case SIZE:
- type = parse_mode_or_normal_call ();
- if (type)
- {
- write_exp_elt_opcode (OP_LONG);
- write_exp_elt_type (builtin_type_int);
- CHECK_TYPEDEF (type);
- write_exp_elt_longcst ((LONGEST) TYPE_LENGTH (type));
- write_exp_elt_opcode (OP_LONG);
- }
- else
- write_exp_elt_opcode (UNOP_SIZEOF);
- break;
- case LOWER:
- op = UNOP_LOWER;
- goto lower_upper;
- case UPPER:
- op = UNOP_UPPER;
- goto lower_upper;
- lower_upper:
- type = parse_mode_or_normal_call ();
- write_lower_upper_value (op, type);
- break;
- case LENGTH:
- parse_unary_call ();
- write_exp_elt_opcode (UNOP_LENGTH);
- break;
- case TYPENAME:
- type = PEEK_LVAL ().tsym.type;
- FORWARD_TOKEN ();
- switch (PEEK_TOKEN ())
- {
- case '[':
- parse_tuple (type);
- break;
- case '(':
- FORWARD_TOKEN ();
- parse_expr ();
- expect (')', "missing right parenthesis");
- write_exp_elt_opcode (UNOP_CAST);
- write_exp_elt_type (type);
- write_exp_elt_opcode (UNOP_CAST);
- break;
- default:
- error ("typename in invalid context");
- }
- break;
-
- default:
- error ("invalid expression syntax at `%s'", lexptr);
- }
- for (;;)
- {
- switch (PEEK_TOKEN ())
- {
- case DOT_FIELD_NAME:
- write_exp_elt_opcode (STRUCTOP_STRUCT);
- write_exp_string (PEEK_LVAL ().sval);
- write_exp_elt_opcode (STRUCTOP_STRUCT);
- FORWARD_TOKEN ();
- continue;
- case POINTER:
- FORWARD_TOKEN ();
- if (PEEK_TOKEN () == TYPENAME)
- {
- type = PEEK_LVAL ().tsym.type;
- write_exp_elt_opcode (UNOP_CAST);
- write_exp_elt_type (lookup_pointer_type (type));
- write_exp_elt_opcode (UNOP_CAST);
- FORWARD_TOKEN ();
- }
- write_exp_elt_opcode (UNOP_IND);
- continue;
- case OPEN_PAREN:
- parse_call ();
- continue;
- case CHARACTER_STRING_LITERAL:
- case CHARACTER_LITERAL:
- case BIT_STRING_LITERAL:
- /* Handle string repetition. (See comment in parse_operand5.) */
- parse_primval ();
- write_exp_elt_opcode (MULTI_SUBSCRIPT);
- write_exp_elt_longcst (1);
- write_exp_elt_opcode (MULTI_SUBSCRIPT);
- continue;
- case END_TOKEN:
- case TOKEN_NOT_READ:
- case INTEGER_LITERAL:
- case BOOLEAN_LITERAL:
- case FLOAT_LITERAL:
- case GENERAL_PROCEDURE_NAME:
- case LOCATION_NAME:
- case EMPTINESS_LITERAL:
- case TYPENAME:
- case CASE:
- case OF:
- case ESAC:
- case LOGIOR:
- case ORIF:
- case LOGXOR:
- case LOGAND:
- case ANDIF:
- case NOTEQUAL:
- case GEQ:
- case LEQ:
- case IN:
- case SLASH_SLASH:
- case MOD:
- case REM:
- case NOT:
- case RECEIVE:
- case UP:
- case IF:
- case THEN:
- case ELSE:
- case FI:
- case ELSIF:
- case ILLEGAL_TOKEN:
- case NUM:
- case PRED:
- case SUCC:
- case ABS:
- case CARD:
- case MAX_TOKEN:
- case MIN_TOKEN:
- case ADDR_TOKEN:
- case SIZE:
- case UPPER:
- case LOWER:
- case LENGTH:
- case ARRAY:
- case GDB_VARIABLE:
- case GDB_ASSIGNMENT:
- break;
- }
- break;
- }
- return;
-}
-
-static void
-parse_operand6 (void)
-{
- if (check_token (RECEIVE))
- {
- parse_primval ();
- error ("not implemented: RECEIVE expression");
- }
- else if (check_token (POINTER))
- {
- parse_primval ();
- write_exp_elt_opcode (UNOP_ADDR);
- }
- else
- parse_primval ();
-}
-
-static void
-parse_operand5 (void)
-{
- enum exp_opcode op;
- /* We are supposed to be looking for a <string repetition operator>,
- but in general we can't distinguish that from a parenthesized
- expression. This is especially difficult if we allow the
- string operand to be a constant expression (as requested by
- some users), and not just a string literal.
- Consider: LPRN expr RPRN LPRN expr RPRN
- Is that a function call or string repetition?
- Instead, we handle string repetition in parse_primval,
- and build_generalized_call. */
- switch (PEEK_TOKEN ())
- {
- case NOT:
- op = UNOP_LOGICAL_NOT;
- break;
- case '-':
- op = UNOP_NEG;
- break;
- default:
- op = OP_NULL;
- }
- if (op != OP_NULL)
- FORWARD_TOKEN ();
- parse_operand6 ();
- if (op != OP_NULL)
- write_exp_elt_opcode (op);
-}
-
-static void
-parse_operand4 (void)
-{
- enum exp_opcode op;
- parse_operand5 ();
- for (;;)
- {
- switch (PEEK_TOKEN ())
- {
- case '*':
- op = BINOP_MUL;
- break;
- case '/':
- op = BINOP_DIV;
- break;
- case MOD:
- op = BINOP_MOD;
- break;
- case REM:
- op = BINOP_REM;
- break;
- default:
- return;
- }
- FORWARD_TOKEN ();
- parse_operand5 ();
- write_exp_elt_opcode (op);
- }
-}
-
-static void
-parse_operand3 (void)
-{
- enum exp_opcode op;
- parse_operand4 ();
- for (;;)
- {
- switch (PEEK_TOKEN ())
- {
- case '+':
- op = BINOP_ADD;
- break;
- case '-':
- op = BINOP_SUB;
- break;
- case SLASH_SLASH:
- op = BINOP_CONCAT;
- break;
- default:
- return;
- }
- FORWARD_TOKEN ();
- parse_operand4 ();
- write_exp_elt_opcode (op);
- }
-}
-
-static void
-parse_operand2 (void)
-{
- enum exp_opcode op;
- parse_operand3 ();
- for (;;)
- {
- if (check_token (IN))
- {
- parse_operand3 ();
- write_exp_elt_opcode (BINOP_IN);
- }
- else
- {
- switch (PEEK_TOKEN ())
- {
- case '>':
- op = BINOP_GTR;
- break;
- case GEQ:
- op = BINOP_GEQ;
- break;
- case '<':
- op = BINOP_LESS;
- break;
- case LEQ:
- op = BINOP_LEQ;
- break;
- case '=':
- op = BINOP_EQUAL;
- break;
- case NOTEQUAL:
- op = BINOP_NOTEQUAL;
- break;
- default:
- return;
- }
- FORWARD_TOKEN ();
- parse_operand3 ();
- write_exp_elt_opcode (op);
- }
- }
-}
-
-static void
-parse_operand1 (void)
-{
- enum exp_opcode op;
- parse_operand2 ();
- for (;;)
- {
- switch (PEEK_TOKEN ())
- {
- case LOGAND:
- op = BINOP_BITWISE_AND;
- break;
- case ANDIF:
- op = BINOP_LOGICAL_AND;
- break;
- default:
- return;
- }
- FORWARD_TOKEN ();
- parse_operand2 ();
- write_exp_elt_opcode (op);
- }
-}
-
-static void
-parse_operand0 (void)
-{
- enum exp_opcode op;
- parse_operand1 ();
- for (;;)
- {
- switch (PEEK_TOKEN ())
- {
- case LOGIOR:
- op = BINOP_BITWISE_IOR;
- break;
- case LOGXOR:
- op = BINOP_BITWISE_XOR;
- break;
- case ORIF:
- op = BINOP_LOGICAL_OR;
- break;
- default:
- return;
- }
- FORWARD_TOKEN ();
- parse_operand1 ();
- write_exp_elt_opcode (op);
- }
-}
-
-static void
-parse_expr (void)
-{
- parse_operand0 ();
- if (check_token (GDB_ASSIGNMENT))
- {
- parse_expr ();
- write_exp_elt_opcode (BINOP_ASSIGN);
- }
-}
-
-static void
-parse_then_alternative (void)
-{
- expect (THEN, "missing 'THEN' in 'IF' expression");
- parse_expr ();
-}
-
-static void
-parse_else_alternative (void)
-{
- if (check_token (ELSIF))
- parse_if_expression_body ();
- else if (check_token (ELSE))
- parse_expr ();
- else
- error ("missing ELSE/ELSIF in IF expression");
-}
-
-/* Matches: <boolean expression> <then alternative> <else alternative> */
-
-static void
-parse_if_expression_body (void)
-{
- parse_expr ();
- parse_then_alternative ();
- parse_else_alternative ();
- write_exp_elt_opcode (TERNOP_COND);
-}
-
-static void
-parse_if_expression (void)
-{
- require (IF);
- parse_if_expression_body ();
- expect (FI, "missing 'FI' at end of conditional expression");
-}
-
-/* An <untyped_expr> is a superset of <expr>. It also includes
- <conditional expressions> and untyped <tuples>, whose types
- are not given by their constituents. Hence, these are only
- allowed in certain contexts that expect a certain type.
- You should call convert() to fix up the <untyped_expr>. */
-
-static void
-parse_untyped_expr (void)
-{
- switch (PEEK_TOKEN ())
- {
- case IF:
- parse_if_expression ();
- return;
- case CASE:
- error ("not implemented: CASE expression");
- case '(':
- switch (PEEK_TOKEN1 ())
- {
- case IF:
- case CASE:
- goto skip_lprn;
- case '[':
- skip_lprn:
- FORWARD_TOKEN ();
- parse_untyped_expr ();
- expect (')', "missing ')'");
- return;
- default:;
- /* fall through */
- }
- default:
- parse_operand0 ();
- }
-}
-
-int
-chill_parse (void)
-{
- terminal_buffer[0] = TOKEN_NOT_READ;
- if (PEEK_TOKEN () == TYPENAME && PEEK_TOKEN1 () == END_TOKEN)
- {
- write_exp_elt_opcode (OP_TYPE);
- write_exp_elt_type (PEEK_LVAL ().tsym.type);
- write_exp_elt_opcode (OP_TYPE);
- FORWARD_TOKEN ();
- }
- else
- parse_expr ();
- if (terminal_buffer[0] != END_TOKEN)
- {
- if (comma_terminates && terminal_buffer[0] == ',')
- lexptr--; /* Put the comma back. */
- else
- error ("Junk after end of expression.");
- }
- return 0;
-}
-
-
-/* Implementation of a dynamically expandable buffer for processing input
- characters acquired through lexptr and building a value to return in
- yylval. */
-
-static char *tempbuf; /* Current buffer contents */
-static int tempbufsize; /* Size of allocated buffer */
-static int tempbufindex; /* Current index into buffer */
-
-#define GROWBY_MIN_SIZE 64 /* Minimum amount to grow buffer by */
-
-#define CHECKBUF(size) \
- do { \
- if (tempbufindex + (size) >= tempbufsize) \
- { \
- growbuf_by_size (size); \
- } \
- } while (0);
-
-/* Grow the static temp buffer if necessary, including allocating the first one
- on demand. */
-
-static void
-growbuf_by_size (int count)
-{
- int growby;
-
- growby = max (count, GROWBY_MIN_SIZE);
- tempbufsize += growby;
- if (tempbuf == NULL)
- {
- tempbuf = (char *) xmalloc (tempbufsize);
- }
- else
- {
- tempbuf = (char *) xrealloc (tempbuf, tempbufsize);
- }
-}
-
-/* Try to consume a simple name string token. If successful, returns
- a pointer to a nullbyte terminated copy of the name that can be used
- in symbol table lookups. If not successful, returns NULL. */
-
-static char *
-match_simple_name_string (void)
-{
- char *tokptr = lexptr;
-
- if (isalpha (*tokptr) || *tokptr == '_')
- {
- char *result;
- do
- {
- tokptr++;
- }
- while (isalnum (*tokptr) || (*tokptr == '_'));
- yylval.sval.ptr = lexptr;
- yylval.sval.length = tokptr - lexptr;
- lexptr = tokptr;
- result = copy_name (yylval.sval);
- return result;
- }
- return (NULL);
-}
-
-/* Start looking for a value composed of valid digits as set by the base
- in use. Note that '_' characters are valid anywhere, in any quantity,
- and are simply ignored. Since we must find at least one valid digit,
- or reject this token as an integer literal, we keep track of how many
- digits we have encountered. */
-
-static int
-decode_integer_value (int base, char **tokptrptr, LONGEST *ivalptr)
-{
- char *tokptr = *tokptrptr;
- int temp;
- int digits = 0;
-
- while (*tokptr != '\0')
- {
- temp = *tokptr;
- if (isupper (temp))
- temp = tolower (temp);
- tokptr++;
- switch (temp)
- {
- case '_':
- continue;
- case '0':
- case '1':
- case '2':
- case '3':
- case '4':
- case '5':
- case '6':
- case '7':
- case '8':
- case '9':
- temp -= '0';
- break;
- case 'a':
- case 'b':
- case 'c':
- case 'd':
- case 'e':
- case 'f':
- temp -= 'a';
- temp += 10;
- break;
- default:
- temp = base;
- break;
- }
- if (temp < base)
- {
- digits++;
- *ivalptr *= base;
- *ivalptr += temp;
- }
- else
- {
- /* Found something not in domain for current base. */
- tokptr--; /* Unconsume what gave us indigestion. */
- break;
- }
- }
-
- /* If we didn't find any digits, then we don't have a valid integer
- value, so reject the entire token. Otherwise, update the lexical
- scan pointer, and return non-zero for success. */
-
- if (digits == 0)
- {
- return (0);
- }
- else
- {
- *tokptrptr = tokptr;
- return (1);
- }
-}
-
-static int
-decode_integer_literal (LONGEST *valptr, char **tokptrptr)
-{
- char *tokptr = *tokptrptr;
- int base = 0;
- LONGEST ival = 0;
- int explicit_base = 0;
-
- /* Look for an explicit base specifier, which is optional. */
-
- switch (*tokptr)
- {
- case 'd':
- case 'D':
- explicit_base++;
- base = 10;
- tokptr++;
- break;
- case 'b':
- case 'B':
- explicit_base++;
- base = 2;
- tokptr++;
- break;
- case 'h':
- case 'H':
- explicit_base++;
- base = 16;
- tokptr++;
- break;
- case 'o':
- case 'O':
- explicit_base++;
- base = 8;
- tokptr++;
- break;
- default:
- base = 10;
- break;
- }
-
- /* If we found an explicit base ensure that the character after the
- explicit base is a single quote. */
-
- if (explicit_base && (*tokptr++ != '\''))
- {
- return (0);
- }
-
- /* Attempt to decode whatever follows as an integer value in the
- indicated base, updating the token pointer in the process and
- computing the value into ival. Also, if we have an explicit
- base, then the next character must not be a single quote, or we
- have a bitstring literal, so reject the entire token in this case.
- Otherwise, update the lexical scan pointer, and return non-zero
- for success. */
-
- if (!decode_integer_value (base, &tokptr, &ival))
- {
- return (0);
- }
- else if (explicit_base && (*tokptr == '\''))
- {
- return (0);
- }
- else
- {
- *valptr = ival;
- *tokptrptr = tokptr;
- return (1);
- }
-}
-
-/* If it wasn't for the fact that floating point values can contain '_'
- characters, we could just let strtod do all the hard work by letting it
- try to consume as much of the current token buffer as possible and
- find a legal conversion. Unfortunately we need to filter out the '_'
- characters before calling strtod, which we do by copying the other
- legal chars to a local buffer to be converted. However since we also
- need to keep track of where the last unconsumed character in the input
- buffer is, we have transfer only as many characters as may compose a
- legal floating point value. */
-
-static enum ch_terminal
-match_float_literal (void)
-{
- char *tokptr = lexptr;
- char *buf;
- char *copy;
- double dval;
- extern double strtod ();
-
- /* Make local buffer in which to build the string to convert. This is
- required because underscores are valid in chill floating point numbers
- but not in the string passed to strtod to convert. The string will be
- no longer than our input string. */
-
- copy = buf = (char *) alloca (strlen (tokptr) + 1);
-
- /* Transfer all leading digits to the conversion buffer, discarding any
- underscores. */
-
- while (isdigit (*tokptr) || *tokptr == '_')
- {
- if (*tokptr != '_')
- {
- *copy++ = *tokptr;
- }
- tokptr++;
- }
-
- /* Now accept either a '.', or one of [eEdD]. Dot is legal regardless
- of whether we found any leading digits, and we simply accept it and
- continue on to look for the fractional part and/or exponent. One of
- [eEdD] is legal only if we have seen digits, and means that there
- is no fractional part. If we find neither of these, then this is
- not a floating point number, so return failure. */
-
- switch (*tokptr++)
- {
- case '.':
- /* Accept and then look for fractional part and/or exponent. */
- *copy++ = '.';
- break;
-
- case 'e':
- case 'E':
- case 'd':
- case 'D':
- if (copy == buf)
- {
- return (0);
- }
- *copy++ = 'e';
- goto collect_exponent;
- break;
-
- default:
- return (0);
- break;
- }
-
- /* We found a '.', copy any fractional digits to the conversion buffer, up
- to the first nondigit, non-underscore character. */
-
- while (isdigit (*tokptr) || *tokptr == '_')
- {
- if (*tokptr != '_')
- {
- *copy++ = *tokptr;
- }
- tokptr++;
- }
-
- /* Look for an exponent, which must start with one of [eEdD]. If none
- is found, jump directly to trying to convert what we have collected
- so far. */
-
- switch (*tokptr)
- {
- case 'e':
- case 'E':
- case 'd':
- case 'D':
- *copy++ = 'e';
- tokptr++;
- break;
- default:
- goto convert_float;
- break;
- }
-
- /* Accept an optional '-' or '+' following one of [eEdD]. */
-
-collect_exponent:
- if (*tokptr == '+' || *tokptr == '-')
- {
- *copy++ = *tokptr++;
- }
-
- /* Now copy an exponent into the conversion buffer. Note that at the
- moment underscores are *not* allowed in exponents. */
-
- while (isdigit (*tokptr))
- {
- *copy++ = *tokptr++;
- }
-
- /* If we transfered any chars to the conversion buffer, try to interpret its
- contents as a floating point value. If any characters remain, then we
- must not have a valid floating point string. */
-
-convert_float:
- *copy = '\0';
- if (copy != buf)
- {
- dval = strtod (buf, &copy);
- if (*copy == '\0')
- {
- yylval.dval = dval;
- lexptr = tokptr;
- return (FLOAT_LITERAL);
- }
- }
- return (0);
-}
-
-/* Recognize a string literal. A string literal is a sequence
- of characters enclosed in matching single or double quotes, except that
- a single character inside single quotes is a character literal, which
- we reject as a string literal. To embed the terminator character inside
- a string, it is simply doubled (I.E. "this""is""one""string") */
-
-static enum ch_terminal
-match_string_literal (void)
-{
- char *tokptr = lexptr;
- int in_ctrlseq = 0;
- LONGEST ival;
-
- for (tempbufindex = 0, tokptr++; *tokptr != '\0'; tokptr++)
- {
- CHECKBUF (1);
- tryagain:;
- if (in_ctrlseq)
- {
- /* skip possible whitespaces */
- while ((*tokptr == ' ' || *tokptr == '\t') && *tokptr)
- tokptr++;
- if (*tokptr == ')')
- {
- in_ctrlseq = 0;
- tokptr++;
- goto tryagain;
- }
- else if (*tokptr != ',')
- error ("Invalid control sequence");
- tokptr++;
- /* skip possible whitespaces */
- while ((*tokptr == ' ' || *tokptr == '\t') && *tokptr)
- tokptr++;
- if (!decode_integer_literal (&ival, &tokptr))
- error ("Invalid control sequence");
- tokptr--;
- }
- else if (*tokptr == *lexptr)
- {
- if (*(tokptr + 1) == *lexptr)
- {
- ival = *tokptr++;
- }
- else
- {
- break;
- }
- }
- else if (*tokptr == '^')
- {
- if (*(tokptr + 1) == '(')
- {
- in_ctrlseq = 1;
- tokptr += 2;
- if (!decode_integer_literal (&ival, &tokptr))
- error ("Invalid control sequence");
- tokptr--;
- }
- else if (*(tokptr + 1) == '^')
- ival = *tokptr++;
- else
- error ("Invalid control sequence");
- }
- else
- ival = *tokptr;
- tempbuf[tempbufindex++] = ival;
- }
- if (in_ctrlseq)
- error ("Invalid control sequence");
-
- if (*tokptr == '\0' /* no terminator */
- || (tempbufindex == 1 && *tokptr == '\'')) /* char literal */
- {
- return (0);
- }
- else
- {
- tempbuf[tempbufindex] = '\0';
- yylval.sval.ptr = tempbuf;
- yylval.sval.length = tempbufindex;
- lexptr = ++tokptr;
- return (CHARACTER_STRING_LITERAL);
- }
-}
-
-/* Recognize a character literal. A character literal is single character
- or a control sequence, enclosed in single quotes. A control sequence
- is a comma separated list of one or more integer literals, enclosed
- in parenthesis and introduced with a circumflex character.
-
- EX: 'a' '^(7)' '^(7,8)'
-
- As a GNU chill extension, the syntax C'xx' is also recognized as a
- character literal, where xx is a hex value for the character.
-
- Note that more than a single character, enclosed in single quotes, is
- a string literal.
-
- Returns CHARACTER_LITERAL if a match is found.
- */
-
-static enum ch_terminal
-match_character_literal (void)
-{
- char *tokptr = lexptr;
- LONGEST ival = 0;
-
- if ((*tokptr == 'c' || *tokptr == 'C') && (*(tokptr + 1) == '\''))
- {
- /* We have a GNU chill extension form, so skip the leading "C'",
- decode the hex value, and then ensure that we have a trailing
- single quote character. */
- tokptr += 2;
- if (!decode_integer_value (16, &tokptr, &ival) || (*tokptr != '\''))
- {
- return (0);
- }
- tokptr++;
- }
- else if (*tokptr == '\'')
- {
- tokptr++;
-
- /* Determine which form we have, either a control sequence or the
- single character form. */
-
- if (*tokptr == '^')
- {
- if (*(tokptr + 1) == '(')
- {
- /* Match and decode a control sequence. Return zero if we don't
- find a valid integer literal, or if the next unconsumed character
- after the integer literal is not the trailing ')'. */
- tokptr += 2;
- if (!decode_integer_literal (&ival, &tokptr) || (*tokptr++ != ')'))
- {
- return (0);
- }
- }
- else if (*(tokptr + 1) == '^')
- {
- ival = *tokptr;
- tokptr += 2;
- }
- else
- /* fail */
- error ("Invalid control sequence");
- }
- else if (*tokptr == '\'')
- {
- /* this must be duplicated */
- ival = *tokptr;
- tokptr += 2;
- }
- else
- {
- ival = *tokptr++;
- }
-
- /* The trailing quote has not yet been consumed. If we don't find
- it, then we have no match. */
-
- if (*tokptr++ != '\'')
- {
- return (0);
- }
- }
- else
- {
- /* Not a character literal. */
- return (0);
- }
- yylval.typed_val.val = ival;
- yylval.typed_val.type = builtin_type_chill_char;
- lexptr = tokptr;
- return (CHARACTER_LITERAL);
-}
-
-/* Recognize an integer literal, as specified in Z.200 sec 5.2.4.2.
- Note that according to 5.2.4.2, a single "_" is also a valid integer
- literal, however GNU-chill requires there to be at least one "digit"
- in any integer literal. */
-
-static enum ch_terminal
-match_integer_literal (void)
-{
- char *tokptr = lexptr;
- LONGEST ival;
-
- if (!decode_integer_literal (&ival, &tokptr))
- {
- return (0);
- }
- else
- {
- yylval.typed_val.val = ival;
-#if defined(CC_HAS_LONG_LONG)
- if (ival > (LONGEST) 2147483647U || ival < -(LONGEST) 2147483648U)
- yylval.typed_val.type = builtin_type_long_long;
- else
-#endif
- yylval.typed_val.type = builtin_type_int;
- lexptr = tokptr;
- return (INTEGER_LITERAL);
- }
-}
-
-/* Recognize a bit-string literal, as specified in Z.200 sec 5.2.4.8
- Note that according to 5.2.4.8, a single "_" is also a valid bit-string
- literal, however GNU-chill requires there to be at least one "digit"
- in any bit-string literal. */
-
-static enum ch_terminal
-match_bitstring_literal (void)
-{
- register char *tokptr = lexptr;
- int bitoffset = 0;
- int bitcount = 0;
- int bits_per_char;
- int digit;
-
- tempbufindex = 0;
- CHECKBUF (1);
- tempbuf[0] = 0;
-
- /* Look for the required explicit base specifier. */
-
- switch (*tokptr++)
- {
- case 'b':
- case 'B':
- bits_per_char = 1;
- break;
- case 'o':
- case 'O':
- bits_per_char = 3;
- break;
- case 'h':
- case 'H':
- bits_per_char = 4;
- break;
- default:
- return (0);
- break;
- }
-
- /* Ensure that the character after the explicit base is a single quote. */
-
- if (*tokptr++ != '\'')
- {
- return (0);
- }
-
- while (*tokptr != '\0' && *tokptr != '\'')
- {
- digit = *tokptr;
- if (isupper (digit))
- digit = tolower (digit);
- tokptr++;
- switch (digit)
- {
- case '_':
- continue;
- case '0':
- case '1':
- case '2':
- case '3':
- case '4':
- case '5':
- case '6':
- case '7':
- case '8':
- case '9':
- digit -= '0';
- break;
- case 'a':
- case 'b':
- case 'c':
- case 'd':
- case 'e':
- case 'f':
- digit -= 'a';
- digit += 10;
- break;
- default:
- /* this is not a bitstring literal, probably an integer */
- return 0;
- }
- if (digit >= 1 << bits_per_char)
- {
- /* Found something not in domain for current base. */
- error ("Too-large digit in bitstring or integer.");
- }
- else
- {
- /* Extract bits from digit, packing them into the bitstring byte. */
- int k = TARGET_BYTE_ORDER == BFD_ENDIAN_BIG ? bits_per_char - 1 : 0;
- for (; TARGET_BYTE_ORDER == BFD_ENDIAN_BIG ? k >= 0 : k < bits_per_char;
- TARGET_BYTE_ORDER == BFD_ENDIAN_BIG ? k-- : k++)
- {
- bitcount++;
- if (digit & (1 << k))
- {
- tempbuf[tempbufindex] |=
- (TARGET_BYTE_ORDER == BFD_ENDIAN_BIG)
- ? (1 << (HOST_CHAR_BIT - 1 - bitoffset))
- : (1 << bitoffset);
- }
- bitoffset++;
- if (bitoffset == HOST_CHAR_BIT)
- {
- bitoffset = 0;
- tempbufindex++;
- CHECKBUF (1);
- tempbuf[tempbufindex] = 0;
- }
- }
- }
- }
-
- /* Verify that we consumed everything up to the trailing single quote,
- and that we found some bits (IE not just underbars). */
-
- if (*tokptr++ != '\'')
- {
- return (0);
- }
- else
- {
- yylval.sval.ptr = tempbuf;
- yylval.sval.length = bitcount;
- lexptr = tokptr;
- return (BIT_STRING_LITERAL);
- }
-}
-
-struct token
-{
- char *operator;
- int token;
-};
-
-static const struct token idtokentab[] =
-{
- {"array", ARRAY},
- {"length", LENGTH},
- {"lower", LOWER},
- {"upper", UPPER},
- {"andif", ANDIF},
- {"pred", PRED},
- {"succ", SUCC},
- {"card", CARD},
- {"size", SIZE},
- {"orif", ORIF},
- {"num", NUM},
- {"abs", ABS},
- {"max", MAX_TOKEN},
- {"min", MIN_TOKEN},
- {"mod", MOD},
- {"rem", REM},
- {"not", NOT},
- {"xor", LOGXOR},
- {"and", LOGAND},
- {"in", IN},
- {"or", LOGIOR},
- {"up", UP},
- {"addr", ADDR_TOKEN},
- {"null", EMPTINESS_LITERAL}
-};
-
-static const struct token tokentab2[] =
-{
- {":=", GDB_ASSIGNMENT},
- {"//", SLASH_SLASH},
- {"->", POINTER},
- {"/=", NOTEQUAL},
- {"<=", LEQ},
- {">=", GEQ}
-};
-
-/* Read one token, getting characters through lexptr. */
-/* This is where we will check to make sure that the language and the
- operators used are compatible. */
-
-static enum ch_terminal
-ch_lex (void)
-{
- unsigned int i;
- enum ch_terminal token;
- char *inputname;
- struct symbol *sym;
-
- /* Skip over any leading whitespace. */
- while (isspace (*lexptr))
- {
- lexptr++;
- }
- /* Look for special single character cases which can't be the first
- character of some other multicharacter token. */
- switch (*lexptr)
- {
- case '\0':
- return END_TOKEN;
- case ',':
- case '=':
- case ';':
- case '!':
- case '+':
- case '*':
- case '(':
- case ')':
- case '[':
- case ']':
- return (*lexptr++);
- }
- /* Look for characters which start a particular kind of multicharacter
- token, such as a character literal, register name, convenience
- variable name, string literal, etc. */
- switch (*lexptr)
- {
- case '\'':
- case '\"':
- /* First try to match a string literal, which is any
- sequence of characters enclosed in matching single or double
- quotes, except that a single character inside single quotes
- is a character literal, so we have to catch that case also. */
- token = match_string_literal ();
- if (token != 0)
- {
- return (token);
- }
- if (*lexptr == '\'')
- {
- token = match_character_literal ();
- if (token != 0)
- {
- return (token);
- }
- }
- break;
- case 'C':
- case 'c':
- token = match_character_literal ();
- if (token != 0)
- {
- return (token);
- }
- break;
- case '$':
- yylval.sval.ptr = lexptr;
- do
- {
- lexptr++;
- }
- while (isalnum (*lexptr) || *lexptr == '_' || *lexptr == '$');
- yylval.sval.length = lexptr - yylval.sval.ptr;
- write_dollar_variable (yylval.sval);
- return GDB_VARIABLE;
- break;
- }
- /* See if it is a special token of length 2. */
- for (i = 0; i < sizeof (tokentab2) / sizeof (tokentab2[0]); i++)
- {
- if (STREQN (lexptr, tokentab2[i].operator, 2))
- {
- lexptr += 2;
- return (tokentab2[i].token);
- }
- }
- /* Look for single character cases which which could be the first
- character of some other multicharacter token, but aren't, or we
- would already have found it. */
- switch (*lexptr)
- {
- case '-':
- case ':':
- case '/':
- case '<':
- case '>':
- return (*lexptr++);
- }
- /* Look for a float literal before looking for an integer literal, so
- we match as much of the input stream as possible. */
- token = match_float_literal ();
- if (token != 0)
- {
- return (token);
- }
- token = match_bitstring_literal ();
- if (token != 0)
- {
- return (token);
- }
- token = match_integer_literal ();
- if (token != 0)
- {
- return (token);
- }
-
- /* Try to match a simple name string, and if a match is found, then
- further classify what sort of name it is and return an appropriate
- token. Note that attempting to match a simple name string consumes
- the token from lexptr, so we can't back out if we later find that
- we can't classify what sort of name it is. */
-
- inputname = match_simple_name_string ();
-
- if (inputname != NULL)
- {
- char *simplename = (char *) alloca (strlen (inputname) + 1);
-
- char *dptr = simplename, *sptr = inputname;
- for (; *sptr; sptr++)
- *dptr++ = isupper (*sptr) ? tolower (*sptr) : *sptr;
- *dptr = '\0';
-
- /* See if it is a reserved identifier. */
- for (i = 0; i < sizeof (idtokentab) / sizeof (idtokentab[0]); i++)
- {
- if (STREQ (simplename, idtokentab[i].operator))
- {
- return (idtokentab[i].token);
- }
- }
-
- /* Look for other special tokens. */
- if (STREQ (simplename, "true"))
- {
- yylval.ulval = 1;
- return (BOOLEAN_LITERAL);
- }
- if (STREQ (simplename, "false"))
- {
- yylval.ulval = 0;
- return (BOOLEAN_LITERAL);
- }
-
- sym = lookup_symbol (inputname, expression_context_block,
- VAR_NAMESPACE, (int *) NULL,
- (struct symtab **) NULL);
- if (sym == NULL && strcmp (inputname, simplename) != 0)
- {
- sym = lookup_symbol (simplename, expression_context_block,
- VAR_NAMESPACE, (int *) NULL,
- (struct symtab **) NULL);
- }
- if (sym != NULL)
- {
- yylval.ssym.stoken.ptr = NULL;
- yylval.ssym.stoken.length = 0;
- yylval.ssym.sym = sym;
- yylval.ssym.is_a_field_of_this = 0; /* FIXME, C++'ism */
- switch (SYMBOL_CLASS (sym))
- {
- case LOC_BLOCK:
- /* Found a procedure name. */
- return (GENERAL_PROCEDURE_NAME);
- case LOC_STATIC:
- /* Found a global or local static variable. */
- return (LOCATION_NAME);
- case LOC_REGISTER:
- case LOC_ARG:
- case LOC_REF_ARG:
- case LOC_REGPARM:
- case LOC_REGPARM_ADDR:
- case LOC_LOCAL:
- case LOC_LOCAL_ARG:
- case LOC_BASEREG:
- case LOC_BASEREG_ARG:
- if (innermost_block == NULL
- || contained_in (block_found, innermost_block))
- {
- innermost_block = block_found;
- }
- return (LOCATION_NAME);
- break;
- case LOC_CONST:
- case LOC_LABEL:
- return (LOCATION_NAME);
- break;
- case LOC_TYPEDEF:
- yylval.tsym.type = SYMBOL_TYPE (sym);
- return TYPENAME;
- case LOC_UNDEF:
- case LOC_CONST_BYTES:
- case LOC_OPTIMIZED_OUT:
- error ("Symbol \"%s\" names no location.", inputname);
- break;
- default:
- internal_error (__FILE__, __LINE__,
- "unhandled SYMBOL_CLASS in ch_lex()");
- break;
- }
- }
- else if (!have_full_symbols () && !have_partial_symbols ())
- {
- error ("No symbol table is loaded. Use the \"file\" command.");
- }
- else
- {
- error ("No symbol \"%s\" in current context.", inputname);
- }
- }
-
- /* Catch single character tokens which are not part of some
- longer token. */
-
- switch (*lexptr)
- {
- case '.': /* Not float for example. */
- lexptr++;
- while (isspace (*lexptr))
- lexptr++;
- inputname = match_simple_name_string ();
- if (!inputname)
- return '.';
- return DOT_FIELD_NAME;
- }
-
- return (ILLEGAL_TOKEN);
-}
-
-static void
-write_lower_upper_value (enum exp_opcode opcode, /* Either UNOP_LOWER or UNOP_UPPER */
- struct type *type)
-{
- if (type == NULL)
- write_exp_elt_opcode (opcode);
- else
- {
- struct type *result_type;
- LONGEST val = type_lower_upper (opcode, type, &result_type);
- write_exp_elt_opcode (OP_LONG);
- write_exp_elt_type (result_type);
- write_exp_elt_longcst (val);
- write_exp_elt_opcode (OP_LONG);
- }
-}
-
-void
-chill_error (char *msg)
-{
- /* Never used. */
-}
+// OBSOLETE /* Parser for GNU CHILL (CCITT High-Level Language) -*- C -*-
+// OBSOLETE Copyright 1992, 1993, 1995, 1996, 1997, 1999, 2000, 2001
+// OBSOLETE Free Software Foundation, Inc.
+// OBSOLETE
+// OBSOLETE This file is part of GDB.
+// OBSOLETE
+// OBSOLETE This program is free software; you can redistribute it and/or modify
+// OBSOLETE it under the terms of the GNU General Public License as published by
+// OBSOLETE the Free Software Foundation; either version 2 of the License, or
+// OBSOLETE (at your option) any later version.
+// OBSOLETE
+// OBSOLETE This program is distributed in the hope that it will be useful,
+// OBSOLETE but WITHOUT ANY WARRANTY; without even the implied warranty of
+// OBSOLETE MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+// OBSOLETE GNU General Public License for more details.
+// OBSOLETE
+// OBSOLETE You should have received a copy of the GNU General Public License
+// OBSOLETE along with this program; if not, write to the Free Software
+// OBSOLETE Foundation, Inc., 59 Temple Place - Suite 330,
+// OBSOLETE Boston, MA 02111-1307, USA. */
+// OBSOLETE
+// OBSOLETE /* Parse a Chill expression from text in a string,
+// OBSOLETE and return the result as a struct expression pointer.
+// OBSOLETE That structure contains arithmetic operations in reverse polish,
+// OBSOLETE with constants represented by operations that are followed by special data.
+// OBSOLETE See expression.h for the details of the format.
+// OBSOLETE What is important here is that it can be built up sequentially
+// OBSOLETE during the process of parsing; the lower levels of the tree always
+// OBSOLETE come first in the result.
+// OBSOLETE
+// OBSOLETE Note that the language accepted by this parser is more liberal
+// OBSOLETE than the one accepted by an actual Chill compiler. For example, the
+// OBSOLETE language rule that a simple name string can not be one of the reserved
+// OBSOLETE simple name strings is not enforced (e.g "case" is not treated as a
+// OBSOLETE reserved name). Another example is that Chill is a strongly typed
+// OBSOLETE language, and certain expressions that violate the type constraints
+// OBSOLETE may still be evaluated if gdb can do so in a meaningful manner, while
+// OBSOLETE such expressions would be rejected by the compiler. The reason for
+// OBSOLETE this more liberal behavior is the philosophy that the debugger
+// OBSOLETE is intended to be a tool that is used by the programmer when things
+// OBSOLETE go wrong, and as such, it should provide as few artificial barriers
+// OBSOLETE to it's use as possible. If it can do something meaningful, even
+// OBSOLETE something that violates language contraints that are enforced by the
+// OBSOLETE compiler, it should do so without complaint.
+// OBSOLETE
+// OBSOLETE */
+// OBSOLETE
+// OBSOLETE #include "defs.h"
+// OBSOLETE #include "gdb_string.h"
+// OBSOLETE #include <ctype.h>
+// OBSOLETE #include "expression.h"
+// OBSOLETE #include "language.h"
+// OBSOLETE #include "value.h"
+// OBSOLETE #include "parser-defs.h"
+// OBSOLETE #include "ch-lang.h"
+// OBSOLETE #include "bfd.h" /* Required by objfiles.h. */
+// OBSOLETE #include "symfile.h" /* Required by objfiles.h. */
+// OBSOLETE #include "objfiles.h" /* For have_full_symbols and have_partial_symbols */
+// OBSOLETE
+// OBSOLETE #ifdef __GNUC__
+// OBSOLETE #define INLINE __inline__
+// OBSOLETE #endif
+// OBSOLETE
+// OBSOLETE typedef union
+// OBSOLETE
+// OBSOLETE {
+// OBSOLETE LONGEST lval;
+// OBSOLETE ULONGEST ulval;
+// OBSOLETE struct
+// OBSOLETE {
+// OBSOLETE LONGEST val;
+// OBSOLETE struct type *type;
+// OBSOLETE }
+// OBSOLETE typed_val;
+// OBSOLETE double dval;
+// OBSOLETE struct symbol *sym;
+// OBSOLETE struct type *tval;
+// OBSOLETE struct stoken sval;
+// OBSOLETE struct ttype tsym;
+// OBSOLETE struct symtoken ssym;
+// OBSOLETE }
+// OBSOLETE YYSTYPE;
+// OBSOLETE
+// OBSOLETE enum ch_terminal
+// OBSOLETE {
+// OBSOLETE END_TOKEN = 0,
+// OBSOLETE /* '\001' ... '\xff' come first. */
+// OBSOLETE OPEN_PAREN = '(',
+// OBSOLETE TOKEN_NOT_READ = 999,
+// OBSOLETE INTEGER_LITERAL,
+// OBSOLETE BOOLEAN_LITERAL,
+// OBSOLETE CHARACTER_LITERAL,
+// OBSOLETE FLOAT_LITERAL,
+// OBSOLETE GENERAL_PROCEDURE_NAME,
+// OBSOLETE LOCATION_NAME,
+// OBSOLETE EMPTINESS_LITERAL,
+// OBSOLETE CHARACTER_STRING_LITERAL,
+// OBSOLETE BIT_STRING_LITERAL,
+// OBSOLETE TYPENAME,
+// OBSOLETE DOT_FIELD_NAME, /* '.' followed by <field name> */
+// OBSOLETE CASE,
+// OBSOLETE OF,
+// OBSOLETE ESAC,
+// OBSOLETE LOGIOR,
+// OBSOLETE ORIF,
+// OBSOLETE LOGXOR,
+// OBSOLETE LOGAND,
+// OBSOLETE ANDIF,
+// OBSOLETE NOTEQUAL,
+// OBSOLETE GEQ,
+// OBSOLETE LEQ,
+// OBSOLETE IN,
+// OBSOLETE SLASH_SLASH,
+// OBSOLETE MOD,
+// OBSOLETE REM,
+// OBSOLETE NOT,
+// OBSOLETE POINTER,
+// OBSOLETE RECEIVE,
+// OBSOLETE UP,
+// OBSOLETE IF,
+// OBSOLETE THEN,
+// OBSOLETE ELSE,
+// OBSOLETE FI,
+// OBSOLETE ELSIF,
+// OBSOLETE ILLEGAL_TOKEN,
+// OBSOLETE NUM,
+// OBSOLETE PRED,
+// OBSOLETE SUCC,
+// OBSOLETE ABS,
+// OBSOLETE CARD,
+// OBSOLETE MAX_TOKEN,
+// OBSOLETE MIN_TOKEN,
+// OBSOLETE ADDR_TOKEN,
+// OBSOLETE SIZE,
+// OBSOLETE UPPER,
+// OBSOLETE LOWER,
+// OBSOLETE LENGTH,
+// OBSOLETE ARRAY,
+// OBSOLETE GDB_VARIABLE,
+// OBSOLETE GDB_ASSIGNMENT
+// OBSOLETE };
+// OBSOLETE
+// OBSOLETE /* Forward declarations. */
+// OBSOLETE
+// OBSOLETE static void write_lower_upper_value (enum exp_opcode, struct type *);
+// OBSOLETE static enum ch_terminal match_bitstring_literal (void);
+// OBSOLETE static enum ch_terminal match_integer_literal (void);
+// OBSOLETE static enum ch_terminal match_character_literal (void);
+// OBSOLETE static enum ch_terminal match_string_literal (void);
+// OBSOLETE static enum ch_terminal match_float_literal (void);
+// OBSOLETE static int decode_integer_literal (LONGEST *, char **);
+// OBSOLETE static int decode_integer_value (int, char **, LONGEST *);
+// OBSOLETE static char *match_simple_name_string (void);
+// OBSOLETE static void growbuf_by_size (int);
+// OBSOLETE static void parse_case_label (void);
+// OBSOLETE static void parse_untyped_expr (void);
+// OBSOLETE static void parse_if_expression (void);
+// OBSOLETE static void parse_if_expression_body (void);
+// OBSOLETE static void parse_else_alternative (void);
+// OBSOLETE static void parse_then_alternative (void);
+// OBSOLETE static void parse_expr (void);
+// OBSOLETE static void parse_operand0 (void);
+// OBSOLETE static void parse_operand1 (void);
+// OBSOLETE static void parse_operand2 (void);
+// OBSOLETE static void parse_operand3 (void);
+// OBSOLETE static void parse_operand4 (void);
+// OBSOLETE static void parse_operand5 (void);
+// OBSOLETE static void parse_operand6 (void);
+// OBSOLETE static void parse_primval (void);
+// OBSOLETE static void parse_tuple (struct type *);
+// OBSOLETE static void parse_opt_element_list (struct type *);
+// OBSOLETE static void parse_tuple_element (struct type *);
+// OBSOLETE static void parse_named_record_element (void);
+// OBSOLETE static void parse_call (void);
+// OBSOLETE static struct type *parse_mode_or_normal_call (void);
+// OBSOLETE #if 0
+// OBSOLETE static struct type *parse_mode_call (void);
+// OBSOLETE #endif
+// OBSOLETE static void parse_unary_call (void);
+// OBSOLETE static int parse_opt_untyped_expr (void);
+// OBSOLETE static int expect (enum ch_terminal, char *);
+// OBSOLETE static enum ch_terminal ch_lex (void);
+// OBSOLETE INLINE static enum ch_terminal PEEK_TOKEN (void);
+// OBSOLETE static enum ch_terminal peek_token_ (int);
+// OBSOLETE static void forward_token_ (void);
+// OBSOLETE static void require (enum ch_terminal);
+// OBSOLETE static int check_token (enum ch_terminal);
+// OBSOLETE
+// OBSOLETE #define MAX_LOOK_AHEAD 2
+// OBSOLETE static enum ch_terminal terminal_buffer[MAX_LOOK_AHEAD + 1] =
+// OBSOLETE {
+// OBSOLETE TOKEN_NOT_READ, TOKEN_NOT_READ, TOKEN_NOT_READ};
+// OBSOLETE static YYSTYPE yylval;
+// OBSOLETE static YYSTYPE val_buffer[MAX_LOOK_AHEAD + 1];
+// OBSOLETE
+// OBSOLETE /*int current_token, lookahead_token; */
+// OBSOLETE
+// OBSOLETE INLINE static enum ch_terminal
+// OBSOLETE PEEK_TOKEN (void)
+// OBSOLETE {
+// OBSOLETE if (terminal_buffer[0] == TOKEN_NOT_READ)
+// OBSOLETE {
+// OBSOLETE terminal_buffer[0] = ch_lex ();
+// OBSOLETE val_buffer[0] = yylval;
+// OBSOLETE }
+// OBSOLETE return terminal_buffer[0];
+// OBSOLETE }
+// OBSOLETE #define PEEK_LVAL() val_buffer[0]
+// OBSOLETE #define PEEK_TOKEN1() peek_token_(1)
+// OBSOLETE #define PEEK_TOKEN2() peek_token_(2)
+// OBSOLETE static enum ch_terminal
+// OBSOLETE peek_token_ (int i)
+// OBSOLETE {
+// OBSOLETE if (i > MAX_LOOK_AHEAD)
+// OBSOLETE internal_error (__FILE__, __LINE__,
+// OBSOLETE "too much lookahead");
+// OBSOLETE if (terminal_buffer[i] == TOKEN_NOT_READ)
+// OBSOLETE {
+// OBSOLETE terminal_buffer[i] = ch_lex ();
+// OBSOLETE val_buffer[i] = yylval;
+// OBSOLETE }
+// OBSOLETE return terminal_buffer[i];
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE #if 0
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE pushback_token (enum ch_terminal code, YYSTYPE node)
+// OBSOLETE {
+// OBSOLETE int i;
+// OBSOLETE if (terminal_buffer[MAX_LOOK_AHEAD] != TOKEN_NOT_READ)
+// OBSOLETE internal_error (__FILE__, __LINE__,
+// OBSOLETE "cannot pushback token");
+// OBSOLETE for (i = MAX_LOOK_AHEAD; i > 0; i--)
+// OBSOLETE {
+// OBSOLETE terminal_buffer[i] = terminal_buffer[i - 1];
+// OBSOLETE val_buffer[i] = val_buffer[i - 1];
+// OBSOLETE }
+// OBSOLETE terminal_buffer[0] = code;
+// OBSOLETE val_buffer[0] = node;
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE #endif
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE forward_token_ (void)
+// OBSOLETE {
+// OBSOLETE int i;
+// OBSOLETE for (i = 0; i < MAX_LOOK_AHEAD; i++)
+// OBSOLETE {
+// OBSOLETE terminal_buffer[i] = terminal_buffer[i + 1];
+// OBSOLETE val_buffer[i] = val_buffer[i + 1];
+// OBSOLETE }
+// OBSOLETE terminal_buffer[MAX_LOOK_AHEAD] = TOKEN_NOT_READ;
+// OBSOLETE }
+// OBSOLETE #define FORWARD_TOKEN() forward_token_()
+// OBSOLETE
+// OBSOLETE /* Skip the next token.
+// OBSOLETE if it isn't TOKEN, the parser is broken. */
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE require (enum ch_terminal token)
+// OBSOLETE {
+// OBSOLETE if (PEEK_TOKEN () != token)
+// OBSOLETE {
+// OBSOLETE internal_error (__FILE__, __LINE__,
+// OBSOLETE "expected token %d", (int) token);
+// OBSOLETE }
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE static int
+// OBSOLETE check_token (enum ch_terminal token)
+// OBSOLETE {
+// OBSOLETE if (PEEK_TOKEN () != token)
+// OBSOLETE return 0;
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE return 1;
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* return 0 if expected token was not found,
+// OBSOLETE else return 1.
+// OBSOLETE */
+// OBSOLETE static int
+// OBSOLETE expect (enum ch_terminal token, char *message)
+// OBSOLETE {
+// OBSOLETE if (PEEK_TOKEN () != token)
+// OBSOLETE {
+// OBSOLETE if (message)
+// OBSOLETE error (message);
+// OBSOLETE else if (token < 256)
+// OBSOLETE error ("syntax error - expected a '%c' here \"%s\"", token, lexptr);
+// OBSOLETE else
+// OBSOLETE error ("syntax error");
+// OBSOLETE return 0;
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE return 1;
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE #if 0
+// OBSOLETE /* Parse a name string. If ALLOW_ALL is 1, ALL is allowed as a postfix. */
+// OBSOLETE
+// OBSOLETE static tree
+// OBSOLETE parse_opt_name_string (int allow_all)
+// OBSOLETE {
+// OBSOLETE int token = PEEK_TOKEN ();
+// OBSOLETE tree name;
+// OBSOLETE if (token != NAME)
+// OBSOLETE {
+// OBSOLETE if (token == ALL && allow_all)
+// OBSOLETE {
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE return ALL_POSTFIX;
+// OBSOLETE }
+// OBSOLETE return NULL_TREE;
+// OBSOLETE }
+// OBSOLETE name = PEEK_LVAL ();
+// OBSOLETE for (;;)
+// OBSOLETE {
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE token = PEEK_TOKEN ();
+// OBSOLETE if (token != '!')
+// OBSOLETE return name;
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE token = PEEK_TOKEN ();
+// OBSOLETE if (token == ALL && allow_all)
+// OBSOLETE return get_identifier3 (IDENTIFIER_POINTER (name), "!", "*");
+// OBSOLETE if (token != NAME)
+// OBSOLETE {
+// OBSOLETE if (pass == 1)
+// OBSOLETE error ("'%s!' is not followed by an identifier",
+// OBSOLETE IDENTIFIER_POINTER (name));
+// OBSOLETE return name;
+// OBSOLETE }
+// OBSOLETE name = get_identifier3 (IDENTIFIER_POINTER (name),
+// OBSOLETE "!", IDENTIFIER_POINTER (PEEK_LVAL ()));
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE static tree
+// OBSOLETE parse_simple_name_string (void)
+// OBSOLETE {
+// OBSOLETE int token = PEEK_TOKEN ();
+// OBSOLETE tree name;
+// OBSOLETE if (token != NAME)
+// OBSOLETE {
+// OBSOLETE error ("expected a name here");
+// OBSOLETE return error_mark_node;
+// OBSOLETE }
+// OBSOLETE name = PEEK_LVAL ();
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE return name;
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE static tree
+// OBSOLETE parse_name_string (void)
+// OBSOLETE {
+// OBSOLETE tree name = parse_opt_name_string (0);
+// OBSOLETE if (name)
+// OBSOLETE return name;
+// OBSOLETE if (pass == 1)
+// OBSOLETE error ("expected a name string here");
+// OBSOLETE return error_mark_node;
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Matches: <name_string>
+// OBSOLETE Returns if pass 1: the identifier.
+// OBSOLETE Returns if pass 2: a decl or value for identifier. */
+// OBSOLETE
+// OBSOLETE static tree
+// OBSOLETE parse_name (void)
+// OBSOLETE {
+// OBSOLETE tree name = parse_name_string ();
+// OBSOLETE if (pass == 1 || ignoring)
+// OBSOLETE return name;
+// OBSOLETE else
+// OBSOLETE {
+// OBSOLETE tree decl = lookup_name (name);
+// OBSOLETE if (decl == NULL_TREE)
+// OBSOLETE {
+// OBSOLETE error ("`%s' undeclared", IDENTIFIER_POINTER (name));
+// OBSOLETE return error_mark_node;
+// OBSOLETE }
+// OBSOLETE else if (TREE_CODE (TREE_TYPE (decl)) == ERROR_MARK)
+// OBSOLETE return error_mark_node;
+// OBSOLETE else if (TREE_CODE (decl) == CONST_DECL)
+// OBSOLETE return DECL_INITIAL (decl);
+// OBSOLETE else if (TREE_CODE (TREE_TYPE (decl)) == REFERENCE_TYPE)
+// OBSOLETE return convert_from_reference (decl);
+// OBSOLETE else
+// OBSOLETE return decl;
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE #endif
+// OBSOLETE
+// OBSOLETE #if 0
+// OBSOLETE static void
+// OBSOLETE pushback_paren_expr (tree expr)
+// OBSOLETE {
+// OBSOLETE if (pass == 1 && !ignoring)
+// OBSOLETE expr = build1 (PAREN_EXPR, NULL_TREE, expr);
+// OBSOLETE pushback_token (EXPR, expr);
+// OBSOLETE }
+// OBSOLETE #endif
+// OBSOLETE
+// OBSOLETE /* Matches: <case label> */
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE parse_case_label (void)
+// OBSOLETE {
+// OBSOLETE if (check_token (ELSE))
+// OBSOLETE error ("ELSE in tuples labels not implemented");
+// OBSOLETE /* Does not handle the case of a mode name. FIXME */
+// OBSOLETE parse_expr ();
+// OBSOLETE if (check_token (':'))
+// OBSOLETE {
+// OBSOLETE parse_expr ();
+// OBSOLETE write_exp_elt_opcode (BINOP_RANGE);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE static int
+// OBSOLETE parse_opt_untyped_expr (void)
+// OBSOLETE {
+// OBSOLETE switch (PEEK_TOKEN ())
+// OBSOLETE {
+// OBSOLETE case ',':
+// OBSOLETE case ':':
+// OBSOLETE case ')':
+// OBSOLETE return 0;
+// OBSOLETE default:
+// OBSOLETE parse_untyped_expr ();
+// OBSOLETE return 1;
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE parse_unary_call (void)
+// OBSOLETE {
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE expect ('(', NULL);
+// OBSOLETE parse_expr ();
+// OBSOLETE expect (')', NULL);
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Parse NAME '(' MODENAME ')'. */
+// OBSOLETE
+// OBSOLETE #if 0
+// OBSOLETE
+// OBSOLETE static struct type *
+// OBSOLETE parse_mode_call (void)
+// OBSOLETE {
+// OBSOLETE struct type *type;
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE expect ('(', NULL);
+// OBSOLETE if (PEEK_TOKEN () != TYPENAME)
+// OBSOLETE error ("expect MODENAME here `%s'", lexptr);
+// OBSOLETE type = PEEK_LVAL ().tsym.type;
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE expect (')', NULL);
+// OBSOLETE return type;
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE #endif
+// OBSOLETE
+// OBSOLETE static struct type *
+// OBSOLETE parse_mode_or_normal_call (void)
+// OBSOLETE {
+// OBSOLETE struct type *type;
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE expect ('(', NULL);
+// OBSOLETE if (PEEK_TOKEN () == TYPENAME)
+// OBSOLETE {
+// OBSOLETE type = PEEK_LVAL ().tsym.type;
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE {
+// OBSOLETE parse_expr ();
+// OBSOLETE type = NULL;
+// OBSOLETE }
+// OBSOLETE expect (')', NULL);
+// OBSOLETE return type;
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Parse something that looks like a function call.
+// OBSOLETE Assume we have parsed the function, and are at the '('. */
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE parse_call (void)
+// OBSOLETE {
+// OBSOLETE int arg_count;
+// OBSOLETE require ('(');
+// OBSOLETE /* This is to save the value of arglist_len
+// OBSOLETE being accumulated for each dimension. */
+// OBSOLETE start_arglist ();
+// OBSOLETE if (parse_opt_untyped_expr ())
+// OBSOLETE {
+// OBSOLETE int tok = PEEK_TOKEN ();
+// OBSOLETE arglist_len = 1;
+// OBSOLETE if (tok == UP || tok == ':')
+// OBSOLETE {
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE parse_expr ();
+// OBSOLETE expect (')', "expected ')' to terminate slice");
+// OBSOLETE end_arglist ();
+// OBSOLETE write_exp_elt_opcode (tok == UP ? TERNOP_SLICE_COUNT
+// OBSOLETE : TERNOP_SLICE);
+// OBSOLETE return;
+// OBSOLETE }
+// OBSOLETE while (check_token (','))
+// OBSOLETE {
+// OBSOLETE parse_untyped_expr ();
+// OBSOLETE arglist_len++;
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE arglist_len = 0;
+// OBSOLETE expect (')', NULL);
+// OBSOLETE arg_count = end_arglist ();
+// OBSOLETE write_exp_elt_opcode (MULTI_SUBSCRIPT);
+// OBSOLETE write_exp_elt_longcst (arg_count);
+// OBSOLETE write_exp_elt_opcode (MULTI_SUBSCRIPT);
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE parse_named_record_element (void)
+// OBSOLETE {
+// OBSOLETE struct stoken label;
+// OBSOLETE char buf[256];
+// OBSOLETE
+// OBSOLETE label = PEEK_LVAL ().sval;
+// OBSOLETE sprintf (buf, "expected a field name here `%s'", lexptr);
+// OBSOLETE expect (DOT_FIELD_NAME, buf);
+// OBSOLETE if (check_token (','))
+// OBSOLETE parse_named_record_element ();
+// OBSOLETE else if (check_token (':'))
+// OBSOLETE parse_expr ();
+// OBSOLETE else
+// OBSOLETE error ("syntax error near `%s' in named record tuple element", lexptr);
+// OBSOLETE write_exp_elt_opcode (OP_LABELED);
+// OBSOLETE write_exp_string (label);
+// OBSOLETE write_exp_elt_opcode (OP_LABELED);
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Returns one or more TREE_LIST nodes, in reverse order. */
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE parse_tuple_element (struct type *type)
+// OBSOLETE {
+// OBSOLETE if (PEEK_TOKEN () == DOT_FIELD_NAME)
+// OBSOLETE {
+// OBSOLETE /* Parse a labelled structure tuple. */
+// OBSOLETE parse_named_record_element ();
+// OBSOLETE return;
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE if (check_token ('('))
+// OBSOLETE {
+// OBSOLETE if (check_token ('*'))
+// OBSOLETE {
+// OBSOLETE expect (')', "missing ')' after '*' case label list");
+// OBSOLETE if (type)
+// OBSOLETE {
+// OBSOLETE if (TYPE_CODE (type) == TYPE_CODE_ARRAY)
+// OBSOLETE {
+// OBSOLETE /* do this as a range from low to high */
+// OBSOLETE struct type *range_type = TYPE_FIELD_TYPE (type, 0);
+// OBSOLETE LONGEST low_bound, high_bound;
+// OBSOLETE if (get_discrete_bounds (range_type, &low_bound, &high_bound) < 0)
+// OBSOLETE error ("cannot determine bounds for (*)");
+// OBSOLETE /* lower bound */
+// OBSOLETE write_exp_elt_opcode (OP_LONG);
+// OBSOLETE write_exp_elt_type (range_type);
+// OBSOLETE write_exp_elt_longcst (low_bound);
+// OBSOLETE write_exp_elt_opcode (OP_LONG);
+// OBSOLETE /* upper bound */
+// OBSOLETE write_exp_elt_opcode (OP_LONG);
+// OBSOLETE write_exp_elt_type (range_type);
+// OBSOLETE write_exp_elt_longcst (high_bound);
+// OBSOLETE write_exp_elt_opcode (OP_LONG);
+// OBSOLETE write_exp_elt_opcode (BINOP_RANGE);
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE error ("(*) in invalid context");
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE error ("(*) only possible with modename in front of tuple (mode[..])");
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE {
+// OBSOLETE parse_case_label ();
+// OBSOLETE while (check_token (','))
+// OBSOLETE {
+// OBSOLETE parse_case_label ();
+// OBSOLETE write_exp_elt_opcode (BINOP_COMMA);
+// OBSOLETE }
+// OBSOLETE expect (')', NULL);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE parse_untyped_expr ();
+// OBSOLETE if (check_token (':'))
+// OBSOLETE {
+// OBSOLETE /* A powerset range or a labeled Array. */
+// OBSOLETE parse_untyped_expr ();
+// OBSOLETE write_exp_elt_opcode (BINOP_RANGE);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Matches: a COMMA-separated list of tuple elements.
+// OBSOLETE Returns a list (of TREE_LIST nodes). */
+// OBSOLETE static void
+// OBSOLETE parse_opt_element_list (struct type *type)
+// OBSOLETE {
+// OBSOLETE arglist_len = 0;
+// OBSOLETE if (PEEK_TOKEN () == ']')
+// OBSOLETE return;
+// OBSOLETE for (;;)
+// OBSOLETE {
+// OBSOLETE parse_tuple_element (type);
+// OBSOLETE arglist_len++;
+// OBSOLETE if (PEEK_TOKEN () == ']')
+// OBSOLETE break;
+// OBSOLETE if (!check_token (','))
+// OBSOLETE error ("bad syntax in tuple");
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Parses: '[' elements ']'
+// OBSOLETE If modename is non-NULL it prefixed the tuple. */
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE parse_tuple (struct type *mode)
+// OBSOLETE {
+// OBSOLETE struct type *type;
+// OBSOLETE if (mode)
+// OBSOLETE type = check_typedef (mode);
+// OBSOLETE else
+// OBSOLETE type = 0;
+// OBSOLETE require ('[');
+// OBSOLETE start_arglist ();
+// OBSOLETE parse_opt_element_list (type);
+// OBSOLETE expect (']', "missing ']' after tuple");
+// OBSOLETE write_exp_elt_opcode (OP_ARRAY);
+// OBSOLETE write_exp_elt_longcst ((LONGEST) 0);
+// OBSOLETE write_exp_elt_longcst ((LONGEST) end_arglist () - 1);
+// OBSOLETE write_exp_elt_opcode (OP_ARRAY);
+// OBSOLETE if (type)
+// OBSOLETE {
+// OBSOLETE if (TYPE_CODE (type) != TYPE_CODE_ARRAY
+// OBSOLETE && TYPE_CODE (type) != TYPE_CODE_STRUCT
+// OBSOLETE && TYPE_CODE (type) != TYPE_CODE_SET)
+// OBSOLETE error ("invalid tuple mode");
+// OBSOLETE write_exp_elt_opcode (UNOP_CAST);
+// OBSOLETE write_exp_elt_type (mode);
+// OBSOLETE write_exp_elt_opcode (UNOP_CAST);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE parse_primval (void)
+// OBSOLETE {
+// OBSOLETE struct type *type;
+// OBSOLETE enum exp_opcode op;
+// OBSOLETE char *op_name;
+// OBSOLETE switch (PEEK_TOKEN ())
+// OBSOLETE {
+// OBSOLETE case INTEGER_LITERAL:
+// OBSOLETE case CHARACTER_LITERAL:
+// OBSOLETE write_exp_elt_opcode (OP_LONG);
+// OBSOLETE write_exp_elt_type (PEEK_LVAL ().typed_val.type);
+// OBSOLETE write_exp_elt_longcst (PEEK_LVAL ().typed_val.val);
+// OBSOLETE write_exp_elt_opcode (OP_LONG);
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE break;
+// OBSOLETE case BOOLEAN_LITERAL:
+// OBSOLETE write_exp_elt_opcode (OP_BOOL);
+// OBSOLETE write_exp_elt_longcst ((LONGEST) PEEK_LVAL ().ulval);
+// OBSOLETE write_exp_elt_opcode (OP_BOOL);
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE break;
+// OBSOLETE case FLOAT_LITERAL:
+// OBSOLETE write_exp_elt_opcode (OP_DOUBLE);
+// OBSOLETE write_exp_elt_type (builtin_type_double);
+// OBSOLETE write_exp_elt_dblcst (PEEK_LVAL ().dval);
+// OBSOLETE write_exp_elt_opcode (OP_DOUBLE);
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE break;
+// OBSOLETE case EMPTINESS_LITERAL:
+// OBSOLETE write_exp_elt_opcode (OP_LONG);
+// OBSOLETE write_exp_elt_type (lookup_pointer_type (builtin_type_void));
+// OBSOLETE write_exp_elt_longcst (0);
+// OBSOLETE write_exp_elt_opcode (OP_LONG);
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE break;
+// OBSOLETE case CHARACTER_STRING_LITERAL:
+// OBSOLETE write_exp_elt_opcode (OP_STRING);
+// OBSOLETE write_exp_string (PEEK_LVAL ().sval);
+// OBSOLETE write_exp_elt_opcode (OP_STRING);
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE break;
+// OBSOLETE case BIT_STRING_LITERAL:
+// OBSOLETE write_exp_elt_opcode (OP_BITSTRING);
+// OBSOLETE write_exp_bitstring (PEEK_LVAL ().sval);
+// OBSOLETE write_exp_elt_opcode (OP_BITSTRING);
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE break;
+// OBSOLETE case ARRAY:
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE /* This is pseudo-Chill, similar to C's '(TYPE[])EXPR'
+// OBSOLETE which casts to an artificial array. */
+// OBSOLETE expect ('(', NULL);
+// OBSOLETE expect (')', NULL);
+// OBSOLETE if (PEEK_TOKEN () != TYPENAME)
+// OBSOLETE error ("missing MODENAME after ARRAY()");
+// OBSOLETE type = PEEK_LVAL ().tsym.type;
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE expect ('(', NULL);
+// OBSOLETE parse_expr ();
+// OBSOLETE expect (')', "missing right parenthesis");
+// OBSOLETE type = create_array_type ((struct type *) NULL, type,
+// OBSOLETE create_range_type ((struct type *) NULL,
+// OBSOLETE builtin_type_int, 0, 0));
+// OBSOLETE TYPE_ARRAY_UPPER_BOUND_TYPE (type) = BOUND_CANNOT_BE_DETERMINED;
+// OBSOLETE write_exp_elt_opcode (UNOP_CAST);
+// OBSOLETE write_exp_elt_type (type);
+// OBSOLETE write_exp_elt_opcode (UNOP_CAST);
+// OBSOLETE break;
+// OBSOLETE #if 0
+// OBSOLETE case CONST:
+// OBSOLETE case EXPR:
+// OBSOLETE val = PEEK_LVAL ();
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE break;
+// OBSOLETE #endif
+// OBSOLETE case '(':
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE parse_expr ();
+// OBSOLETE expect (')', "missing right parenthesis");
+// OBSOLETE break;
+// OBSOLETE case '[':
+// OBSOLETE parse_tuple (NULL);
+// OBSOLETE break;
+// OBSOLETE case GENERAL_PROCEDURE_NAME:
+// OBSOLETE case LOCATION_NAME:
+// OBSOLETE write_exp_elt_opcode (OP_VAR_VALUE);
+// OBSOLETE write_exp_elt_block (NULL);
+// OBSOLETE write_exp_elt_sym (PEEK_LVAL ().ssym.sym);
+// OBSOLETE write_exp_elt_opcode (OP_VAR_VALUE);
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE break;
+// OBSOLETE case GDB_VARIABLE: /* gdb specific */
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE break;
+// OBSOLETE case NUM:
+// OBSOLETE parse_unary_call ();
+// OBSOLETE write_exp_elt_opcode (UNOP_CAST);
+// OBSOLETE write_exp_elt_type (builtin_type_int);
+// OBSOLETE write_exp_elt_opcode (UNOP_CAST);
+// OBSOLETE break;
+// OBSOLETE case CARD:
+// OBSOLETE parse_unary_call ();
+// OBSOLETE write_exp_elt_opcode (UNOP_CARD);
+// OBSOLETE break;
+// OBSOLETE case MAX_TOKEN:
+// OBSOLETE parse_unary_call ();
+// OBSOLETE write_exp_elt_opcode (UNOP_CHMAX);
+// OBSOLETE break;
+// OBSOLETE case MIN_TOKEN:
+// OBSOLETE parse_unary_call ();
+// OBSOLETE write_exp_elt_opcode (UNOP_CHMIN);
+// OBSOLETE break;
+// OBSOLETE case PRED:
+// OBSOLETE op_name = "PRED";
+// OBSOLETE goto unimplemented_unary_builtin;
+// OBSOLETE case SUCC:
+// OBSOLETE op_name = "SUCC";
+// OBSOLETE goto unimplemented_unary_builtin;
+// OBSOLETE case ABS:
+// OBSOLETE op_name = "ABS";
+// OBSOLETE goto unimplemented_unary_builtin;
+// OBSOLETE unimplemented_unary_builtin:
+// OBSOLETE parse_unary_call ();
+// OBSOLETE error ("not implemented: %s builtin function", op_name);
+// OBSOLETE break;
+// OBSOLETE case ADDR_TOKEN:
+// OBSOLETE parse_unary_call ();
+// OBSOLETE write_exp_elt_opcode (UNOP_ADDR);
+// OBSOLETE break;
+// OBSOLETE case SIZE:
+// OBSOLETE type = parse_mode_or_normal_call ();
+// OBSOLETE if (type)
+// OBSOLETE {
+// OBSOLETE write_exp_elt_opcode (OP_LONG);
+// OBSOLETE write_exp_elt_type (builtin_type_int);
+// OBSOLETE CHECK_TYPEDEF (type);
+// OBSOLETE write_exp_elt_longcst ((LONGEST) TYPE_LENGTH (type));
+// OBSOLETE write_exp_elt_opcode (OP_LONG);
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE write_exp_elt_opcode (UNOP_SIZEOF);
+// OBSOLETE break;
+// OBSOLETE case LOWER:
+// OBSOLETE op = UNOP_LOWER;
+// OBSOLETE goto lower_upper;
+// OBSOLETE case UPPER:
+// OBSOLETE op = UNOP_UPPER;
+// OBSOLETE goto lower_upper;
+// OBSOLETE lower_upper:
+// OBSOLETE type = parse_mode_or_normal_call ();
+// OBSOLETE write_lower_upper_value (op, type);
+// OBSOLETE break;
+// OBSOLETE case LENGTH:
+// OBSOLETE parse_unary_call ();
+// OBSOLETE write_exp_elt_opcode (UNOP_LENGTH);
+// OBSOLETE break;
+// OBSOLETE case TYPENAME:
+// OBSOLETE type = PEEK_LVAL ().tsym.type;
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE switch (PEEK_TOKEN ())
+// OBSOLETE {
+// OBSOLETE case '[':
+// OBSOLETE parse_tuple (type);
+// OBSOLETE break;
+// OBSOLETE case '(':
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE parse_expr ();
+// OBSOLETE expect (')', "missing right parenthesis");
+// OBSOLETE write_exp_elt_opcode (UNOP_CAST);
+// OBSOLETE write_exp_elt_type (type);
+// OBSOLETE write_exp_elt_opcode (UNOP_CAST);
+// OBSOLETE break;
+// OBSOLETE default:
+// OBSOLETE error ("typename in invalid context");
+// OBSOLETE }
+// OBSOLETE break;
+// OBSOLETE
+// OBSOLETE default:
+// OBSOLETE error ("invalid expression syntax at `%s'", lexptr);
+// OBSOLETE }
+// OBSOLETE for (;;)
+// OBSOLETE {
+// OBSOLETE switch (PEEK_TOKEN ())
+// OBSOLETE {
+// OBSOLETE case DOT_FIELD_NAME:
+// OBSOLETE write_exp_elt_opcode (STRUCTOP_STRUCT);
+// OBSOLETE write_exp_string (PEEK_LVAL ().sval);
+// OBSOLETE write_exp_elt_opcode (STRUCTOP_STRUCT);
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE continue;
+// OBSOLETE case POINTER:
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE if (PEEK_TOKEN () == TYPENAME)
+// OBSOLETE {
+// OBSOLETE type = PEEK_LVAL ().tsym.type;
+// OBSOLETE write_exp_elt_opcode (UNOP_CAST);
+// OBSOLETE write_exp_elt_type (lookup_pointer_type (type));
+// OBSOLETE write_exp_elt_opcode (UNOP_CAST);
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE }
+// OBSOLETE write_exp_elt_opcode (UNOP_IND);
+// OBSOLETE continue;
+// OBSOLETE case OPEN_PAREN:
+// OBSOLETE parse_call ();
+// OBSOLETE continue;
+// OBSOLETE case CHARACTER_STRING_LITERAL:
+// OBSOLETE case CHARACTER_LITERAL:
+// OBSOLETE case BIT_STRING_LITERAL:
+// OBSOLETE /* Handle string repetition. (See comment in parse_operand5.) */
+// OBSOLETE parse_primval ();
+// OBSOLETE write_exp_elt_opcode (MULTI_SUBSCRIPT);
+// OBSOLETE write_exp_elt_longcst (1);
+// OBSOLETE write_exp_elt_opcode (MULTI_SUBSCRIPT);
+// OBSOLETE continue;
+// OBSOLETE case END_TOKEN:
+// OBSOLETE case TOKEN_NOT_READ:
+// OBSOLETE case INTEGER_LITERAL:
+// OBSOLETE case BOOLEAN_LITERAL:
+// OBSOLETE case FLOAT_LITERAL:
+// OBSOLETE case GENERAL_PROCEDURE_NAME:
+// OBSOLETE case LOCATION_NAME:
+// OBSOLETE case EMPTINESS_LITERAL:
+// OBSOLETE case TYPENAME:
+// OBSOLETE case CASE:
+// OBSOLETE case OF:
+// OBSOLETE case ESAC:
+// OBSOLETE case LOGIOR:
+// OBSOLETE case ORIF:
+// OBSOLETE case LOGXOR:
+// OBSOLETE case LOGAND:
+// OBSOLETE case ANDIF:
+// OBSOLETE case NOTEQUAL:
+// OBSOLETE case GEQ:
+// OBSOLETE case LEQ:
+// OBSOLETE case IN:
+// OBSOLETE case SLASH_SLASH:
+// OBSOLETE case MOD:
+// OBSOLETE case REM:
+// OBSOLETE case NOT:
+// OBSOLETE case RECEIVE:
+// OBSOLETE case UP:
+// OBSOLETE case IF:
+// OBSOLETE case THEN:
+// OBSOLETE case ELSE:
+// OBSOLETE case FI:
+// OBSOLETE case ELSIF:
+// OBSOLETE case ILLEGAL_TOKEN:
+// OBSOLETE case NUM:
+// OBSOLETE case PRED:
+// OBSOLETE case SUCC:
+// OBSOLETE case ABS:
+// OBSOLETE case CARD:
+// OBSOLETE case MAX_TOKEN:
+// OBSOLETE case MIN_TOKEN:
+// OBSOLETE case ADDR_TOKEN:
+// OBSOLETE case SIZE:
+// OBSOLETE case UPPER:
+// OBSOLETE case LOWER:
+// OBSOLETE case LENGTH:
+// OBSOLETE case ARRAY:
+// OBSOLETE case GDB_VARIABLE:
+// OBSOLETE case GDB_ASSIGNMENT:
+// OBSOLETE break;
+// OBSOLETE }
+// OBSOLETE break;
+// OBSOLETE }
+// OBSOLETE return;
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE parse_operand6 (void)
+// OBSOLETE {
+// OBSOLETE if (check_token (RECEIVE))
+// OBSOLETE {
+// OBSOLETE parse_primval ();
+// OBSOLETE error ("not implemented: RECEIVE expression");
+// OBSOLETE }
+// OBSOLETE else if (check_token (POINTER))
+// OBSOLETE {
+// OBSOLETE parse_primval ();
+// OBSOLETE write_exp_elt_opcode (UNOP_ADDR);
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE parse_primval ();
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE parse_operand5 (void)
+// OBSOLETE {
+// OBSOLETE enum exp_opcode op;
+// OBSOLETE /* We are supposed to be looking for a <string repetition operator>,
+// OBSOLETE but in general we can't distinguish that from a parenthesized
+// OBSOLETE expression. This is especially difficult if we allow the
+// OBSOLETE string operand to be a constant expression (as requested by
+// OBSOLETE some users), and not just a string literal.
+// OBSOLETE Consider: LPRN expr RPRN LPRN expr RPRN
+// OBSOLETE Is that a function call or string repetition?
+// OBSOLETE Instead, we handle string repetition in parse_primval,
+// OBSOLETE and build_generalized_call. */
+// OBSOLETE switch (PEEK_TOKEN ())
+// OBSOLETE {
+// OBSOLETE case NOT:
+// OBSOLETE op = UNOP_LOGICAL_NOT;
+// OBSOLETE break;
+// OBSOLETE case '-':
+// OBSOLETE op = UNOP_NEG;
+// OBSOLETE break;
+// OBSOLETE default:
+// OBSOLETE op = OP_NULL;
+// OBSOLETE }
+// OBSOLETE if (op != OP_NULL)
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE parse_operand6 ();
+// OBSOLETE if (op != OP_NULL)
+// OBSOLETE write_exp_elt_opcode (op);
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE parse_operand4 (void)
+// OBSOLETE {
+// OBSOLETE enum exp_opcode op;
+// OBSOLETE parse_operand5 ();
+// OBSOLETE for (;;)
+// OBSOLETE {
+// OBSOLETE switch (PEEK_TOKEN ())
+// OBSOLETE {
+// OBSOLETE case '*':
+// OBSOLETE op = BINOP_MUL;
+// OBSOLETE break;
+// OBSOLETE case '/':
+// OBSOLETE op = BINOP_DIV;
+// OBSOLETE break;
+// OBSOLETE case MOD:
+// OBSOLETE op = BINOP_MOD;
+// OBSOLETE break;
+// OBSOLETE case REM:
+// OBSOLETE op = BINOP_REM;
+// OBSOLETE break;
+// OBSOLETE default:
+// OBSOLETE return;
+// OBSOLETE }
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE parse_operand5 ();
+// OBSOLETE write_exp_elt_opcode (op);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE parse_operand3 (void)
+// OBSOLETE {
+// OBSOLETE enum exp_opcode op;
+// OBSOLETE parse_operand4 ();
+// OBSOLETE for (;;)
+// OBSOLETE {
+// OBSOLETE switch (PEEK_TOKEN ())
+// OBSOLETE {
+// OBSOLETE case '+':
+// OBSOLETE op = BINOP_ADD;
+// OBSOLETE break;
+// OBSOLETE case '-':
+// OBSOLETE op = BINOP_SUB;
+// OBSOLETE break;
+// OBSOLETE case SLASH_SLASH:
+// OBSOLETE op = BINOP_CONCAT;
+// OBSOLETE break;
+// OBSOLETE default:
+// OBSOLETE return;
+// OBSOLETE }
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE parse_operand4 ();
+// OBSOLETE write_exp_elt_opcode (op);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE parse_operand2 (void)
+// OBSOLETE {
+// OBSOLETE enum exp_opcode op;
+// OBSOLETE parse_operand3 ();
+// OBSOLETE for (;;)
+// OBSOLETE {
+// OBSOLETE if (check_token (IN))
+// OBSOLETE {
+// OBSOLETE parse_operand3 ();
+// OBSOLETE write_exp_elt_opcode (BINOP_IN);
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE {
+// OBSOLETE switch (PEEK_TOKEN ())
+// OBSOLETE {
+// OBSOLETE case '>':
+// OBSOLETE op = BINOP_GTR;
+// OBSOLETE break;
+// OBSOLETE case GEQ:
+// OBSOLETE op = BINOP_GEQ;
+// OBSOLETE break;
+// OBSOLETE case '<':
+// OBSOLETE op = BINOP_LESS;
+// OBSOLETE break;
+// OBSOLETE case LEQ:
+// OBSOLETE op = BINOP_LEQ;
+// OBSOLETE break;
+// OBSOLETE case '=':
+// OBSOLETE op = BINOP_EQUAL;
+// OBSOLETE break;
+// OBSOLETE case NOTEQUAL:
+// OBSOLETE op = BINOP_NOTEQUAL;
+// OBSOLETE break;
+// OBSOLETE default:
+// OBSOLETE return;
+// OBSOLETE }
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE parse_operand3 ();
+// OBSOLETE write_exp_elt_opcode (op);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE parse_operand1 (void)
+// OBSOLETE {
+// OBSOLETE enum exp_opcode op;
+// OBSOLETE parse_operand2 ();
+// OBSOLETE for (;;)
+// OBSOLETE {
+// OBSOLETE switch (PEEK_TOKEN ())
+// OBSOLETE {
+// OBSOLETE case LOGAND:
+// OBSOLETE op = BINOP_BITWISE_AND;
+// OBSOLETE break;
+// OBSOLETE case ANDIF:
+// OBSOLETE op = BINOP_LOGICAL_AND;
+// OBSOLETE break;
+// OBSOLETE default:
+// OBSOLETE return;
+// OBSOLETE }
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE parse_operand2 ();
+// OBSOLETE write_exp_elt_opcode (op);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE parse_operand0 (void)
+// OBSOLETE {
+// OBSOLETE enum exp_opcode op;
+// OBSOLETE parse_operand1 ();
+// OBSOLETE for (;;)
+// OBSOLETE {
+// OBSOLETE switch (PEEK_TOKEN ())
+// OBSOLETE {
+// OBSOLETE case LOGIOR:
+// OBSOLETE op = BINOP_BITWISE_IOR;
+// OBSOLETE break;
+// OBSOLETE case LOGXOR:
+// OBSOLETE op = BINOP_BITWISE_XOR;
+// OBSOLETE break;
+// OBSOLETE case ORIF:
+// OBSOLETE op = BINOP_LOGICAL_OR;
+// OBSOLETE break;
+// OBSOLETE default:
+// OBSOLETE return;
+// OBSOLETE }
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE parse_operand1 ();
+// OBSOLETE write_exp_elt_opcode (op);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE parse_expr (void)
+// OBSOLETE {
+// OBSOLETE parse_operand0 ();
+// OBSOLETE if (check_token (GDB_ASSIGNMENT))
+// OBSOLETE {
+// OBSOLETE parse_expr ();
+// OBSOLETE write_exp_elt_opcode (BINOP_ASSIGN);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE parse_then_alternative (void)
+// OBSOLETE {
+// OBSOLETE expect (THEN, "missing 'THEN' in 'IF' expression");
+// OBSOLETE parse_expr ();
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE parse_else_alternative (void)
+// OBSOLETE {
+// OBSOLETE if (check_token (ELSIF))
+// OBSOLETE parse_if_expression_body ();
+// OBSOLETE else if (check_token (ELSE))
+// OBSOLETE parse_expr ();
+// OBSOLETE else
+// OBSOLETE error ("missing ELSE/ELSIF in IF expression");
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Matches: <boolean expression> <then alternative> <else alternative> */
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE parse_if_expression_body (void)
+// OBSOLETE {
+// OBSOLETE parse_expr ();
+// OBSOLETE parse_then_alternative ();
+// OBSOLETE parse_else_alternative ();
+// OBSOLETE write_exp_elt_opcode (TERNOP_COND);
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE parse_if_expression (void)
+// OBSOLETE {
+// OBSOLETE require (IF);
+// OBSOLETE parse_if_expression_body ();
+// OBSOLETE expect (FI, "missing 'FI' at end of conditional expression");
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* An <untyped_expr> is a superset of <expr>. It also includes
+// OBSOLETE <conditional expressions> and untyped <tuples>, whose types
+// OBSOLETE are not given by their constituents. Hence, these are only
+// OBSOLETE allowed in certain contexts that expect a certain type.
+// OBSOLETE You should call convert() to fix up the <untyped_expr>. */
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE parse_untyped_expr (void)
+// OBSOLETE {
+// OBSOLETE switch (PEEK_TOKEN ())
+// OBSOLETE {
+// OBSOLETE case IF:
+// OBSOLETE parse_if_expression ();
+// OBSOLETE return;
+// OBSOLETE case CASE:
+// OBSOLETE error ("not implemented: CASE expression");
+// OBSOLETE case '(':
+// OBSOLETE switch (PEEK_TOKEN1 ())
+// OBSOLETE {
+// OBSOLETE case IF:
+// OBSOLETE case CASE:
+// OBSOLETE goto skip_lprn;
+// OBSOLETE case '[':
+// OBSOLETE skip_lprn:
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE parse_untyped_expr ();
+// OBSOLETE expect (')', "missing ')'");
+// OBSOLETE return;
+// OBSOLETE default:;
+// OBSOLETE /* fall through */
+// OBSOLETE }
+// OBSOLETE default:
+// OBSOLETE parse_operand0 ();
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE int
+// OBSOLETE chill_parse (void)
+// OBSOLETE {
+// OBSOLETE terminal_buffer[0] = TOKEN_NOT_READ;
+// OBSOLETE if (PEEK_TOKEN () == TYPENAME && PEEK_TOKEN1 () == END_TOKEN)
+// OBSOLETE {
+// OBSOLETE write_exp_elt_opcode (OP_TYPE);
+// OBSOLETE write_exp_elt_type (PEEK_LVAL ().tsym.type);
+// OBSOLETE write_exp_elt_opcode (OP_TYPE);
+// OBSOLETE FORWARD_TOKEN ();
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE parse_expr ();
+// OBSOLETE if (terminal_buffer[0] != END_TOKEN)
+// OBSOLETE {
+// OBSOLETE if (comma_terminates && terminal_buffer[0] == ',')
+// OBSOLETE lexptr--; /* Put the comma back. */
+// OBSOLETE else
+// OBSOLETE error ("Junk after end of expression.");
+// OBSOLETE }
+// OBSOLETE return 0;
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE
+// OBSOLETE /* Implementation of a dynamically expandable buffer for processing input
+// OBSOLETE characters acquired through lexptr and building a value to return in
+// OBSOLETE yylval. */
+// OBSOLETE
+// OBSOLETE static char *tempbuf; /* Current buffer contents */
+// OBSOLETE static int tempbufsize; /* Size of allocated buffer */
+// OBSOLETE static int tempbufindex; /* Current index into buffer */
+// OBSOLETE
+// OBSOLETE #define GROWBY_MIN_SIZE 64 /* Minimum amount to grow buffer by */
+// OBSOLETE
+// OBSOLETE #define CHECKBUF(size) \
+// OBSOLETE do { \
+// OBSOLETE if (tempbufindex + (size) >= tempbufsize) \
+// OBSOLETE { \
+// OBSOLETE growbuf_by_size (size); \
+// OBSOLETE } \
+// OBSOLETE } while (0);
+// OBSOLETE
+// OBSOLETE /* Grow the static temp buffer if necessary, including allocating the first one
+// OBSOLETE on demand. */
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE growbuf_by_size (int count)
+// OBSOLETE {
+// OBSOLETE int growby;
+// OBSOLETE
+// OBSOLETE growby = max (count, GROWBY_MIN_SIZE);
+// OBSOLETE tempbufsize += growby;
+// OBSOLETE if (tempbuf == NULL)
+// OBSOLETE {
+// OBSOLETE tempbuf = (char *) xmalloc (tempbufsize);
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE {
+// OBSOLETE tempbuf = (char *) xrealloc (tempbuf, tempbufsize);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Try to consume a simple name string token. If successful, returns
+// OBSOLETE a pointer to a nullbyte terminated copy of the name that can be used
+// OBSOLETE in symbol table lookups. If not successful, returns NULL. */
+// OBSOLETE
+// OBSOLETE static char *
+// OBSOLETE match_simple_name_string (void)
+// OBSOLETE {
+// OBSOLETE char *tokptr = lexptr;
+// OBSOLETE
+// OBSOLETE if (isalpha (*tokptr) || *tokptr == '_')
+// OBSOLETE {
+// OBSOLETE char *result;
+// OBSOLETE do
+// OBSOLETE {
+// OBSOLETE tokptr++;
+// OBSOLETE }
+// OBSOLETE while (isalnum (*tokptr) || (*tokptr == '_'));
+// OBSOLETE yylval.sval.ptr = lexptr;
+// OBSOLETE yylval.sval.length = tokptr - lexptr;
+// OBSOLETE lexptr = tokptr;
+// OBSOLETE result = copy_name (yylval.sval);
+// OBSOLETE return result;
+// OBSOLETE }
+// OBSOLETE return (NULL);
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Start looking for a value composed of valid digits as set by the base
+// OBSOLETE in use. Note that '_' characters are valid anywhere, in any quantity,
+// OBSOLETE and are simply ignored. Since we must find at least one valid digit,
+// OBSOLETE or reject this token as an integer literal, we keep track of how many
+// OBSOLETE digits we have encountered. */
+// OBSOLETE
+// OBSOLETE static int
+// OBSOLETE decode_integer_value (int base, char **tokptrptr, LONGEST *ivalptr)
+// OBSOLETE {
+// OBSOLETE char *tokptr = *tokptrptr;
+// OBSOLETE int temp;
+// OBSOLETE int digits = 0;
+// OBSOLETE
+// OBSOLETE while (*tokptr != '\0')
+// OBSOLETE {
+// OBSOLETE temp = *tokptr;
+// OBSOLETE if (isupper (temp))
+// OBSOLETE temp = tolower (temp);
+// OBSOLETE tokptr++;
+// OBSOLETE switch (temp)
+// OBSOLETE {
+// OBSOLETE case '_':
+// OBSOLETE continue;
+// OBSOLETE case '0':
+// OBSOLETE case '1':
+// OBSOLETE case '2':
+// OBSOLETE case '3':
+// OBSOLETE case '4':
+// OBSOLETE case '5':
+// OBSOLETE case '6':
+// OBSOLETE case '7':
+// OBSOLETE case '8':
+// OBSOLETE case '9':
+// OBSOLETE temp -= '0';
+// OBSOLETE break;
+// OBSOLETE case 'a':
+// OBSOLETE case 'b':
+// OBSOLETE case 'c':
+// OBSOLETE case 'd':
+// OBSOLETE case 'e':
+// OBSOLETE case 'f':
+// OBSOLETE temp -= 'a';
+// OBSOLETE temp += 10;
+// OBSOLETE break;
+// OBSOLETE default:
+// OBSOLETE temp = base;
+// OBSOLETE break;
+// OBSOLETE }
+// OBSOLETE if (temp < base)
+// OBSOLETE {
+// OBSOLETE digits++;
+// OBSOLETE *ivalptr *= base;
+// OBSOLETE *ivalptr += temp;
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE {
+// OBSOLETE /* Found something not in domain for current base. */
+// OBSOLETE tokptr--; /* Unconsume what gave us indigestion. */
+// OBSOLETE break;
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* If we didn't find any digits, then we don't have a valid integer
+// OBSOLETE value, so reject the entire token. Otherwise, update the lexical
+// OBSOLETE scan pointer, and return non-zero for success. */
+// OBSOLETE
+// OBSOLETE if (digits == 0)
+// OBSOLETE {
+// OBSOLETE return (0);
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE {
+// OBSOLETE *tokptrptr = tokptr;
+// OBSOLETE return (1);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE static int
+// OBSOLETE decode_integer_literal (LONGEST *valptr, char **tokptrptr)
+// OBSOLETE {
+// OBSOLETE char *tokptr = *tokptrptr;
+// OBSOLETE int base = 0;
+// OBSOLETE LONGEST ival = 0;
+// OBSOLETE int explicit_base = 0;
+// OBSOLETE
+// OBSOLETE /* Look for an explicit base specifier, which is optional. */
+// OBSOLETE
+// OBSOLETE switch (*tokptr)
+// OBSOLETE {
+// OBSOLETE case 'd':
+// OBSOLETE case 'D':
+// OBSOLETE explicit_base++;
+// OBSOLETE base = 10;
+// OBSOLETE tokptr++;
+// OBSOLETE break;
+// OBSOLETE case 'b':
+// OBSOLETE case 'B':
+// OBSOLETE explicit_base++;
+// OBSOLETE base = 2;
+// OBSOLETE tokptr++;
+// OBSOLETE break;
+// OBSOLETE case 'h':
+// OBSOLETE case 'H':
+// OBSOLETE explicit_base++;
+// OBSOLETE base = 16;
+// OBSOLETE tokptr++;
+// OBSOLETE break;
+// OBSOLETE case 'o':
+// OBSOLETE case 'O':
+// OBSOLETE explicit_base++;
+// OBSOLETE base = 8;
+// OBSOLETE tokptr++;
+// OBSOLETE break;
+// OBSOLETE default:
+// OBSOLETE base = 10;
+// OBSOLETE break;
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* If we found an explicit base ensure that the character after the
+// OBSOLETE explicit base is a single quote. */
+// OBSOLETE
+// OBSOLETE if (explicit_base && (*tokptr++ != '\''))
+// OBSOLETE {
+// OBSOLETE return (0);
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Attempt to decode whatever follows as an integer value in the
+// OBSOLETE indicated base, updating the token pointer in the process and
+// OBSOLETE computing the value into ival. Also, if we have an explicit
+// OBSOLETE base, then the next character must not be a single quote, or we
+// OBSOLETE have a bitstring literal, so reject the entire token in this case.
+// OBSOLETE Otherwise, update the lexical scan pointer, and return non-zero
+// OBSOLETE for success. */
+// OBSOLETE
+// OBSOLETE if (!decode_integer_value (base, &tokptr, &ival))
+// OBSOLETE {
+// OBSOLETE return (0);
+// OBSOLETE }
+// OBSOLETE else if (explicit_base && (*tokptr == '\''))
+// OBSOLETE {
+// OBSOLETE return (0);
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE {
+// OBSOLETE *valptr = ival;
+// OBSOLETE *tokptrptr = tokptr;
+// OBSOLETE return (1);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* If it wasn't for the fact that floating point values can contain '_'
+// OBSOLETE characters, we could just let strtod do all the hard work by letting it
+// OBSOLETE try to consume as much of the current token buffer as possible and
+// OBSOLETE find a legal conversion. Unfortunately we need to filter out the '_'
+// OBSOLETE characters before calling strtod, which we do by copying the other
+// OBSOLETE legal chars to a local buffer to be converted. However since we also
+// OBSOLETE need to keep track of where the last unconsumed character in the input
+// OBSOLETE buffer is, we have transfer only as many characters as may compose a
+// OBSOLETE legal floating point value. */
+// OBSOLETE
+// OBSOLETE static enum ch_terminal
+// OBSOLETE match_float_literal (void)
+// OBSOLETE {
+// OBSOLETE char *tokptr = lexptr;
+// OBSOLETE char *buf;
+// OBSOLETE char *copy;
+// OBSOLETE double dval;
+// OBSOLETE extern double strtod ();
+// OBSOLETE
+// OBSOLETE /* Make local buffer in which to build the string to convert. This is
+// OBSOLETE required because underscores are valid in chill floating point numbers
+// OBSOLETE but not in the string passed to strtod to convert. The string will be
+// OBSOLETE no longer than our input string. */
+// OBSOLETE
+// OBSOLETE copy = buf = (char *) alloca (strlen (tokptr) + 1);
+// OBSOLETE
+// OBSOLETE /* Transfer all leading digits to the conversion buffer, discarding any
+// OBSOLETE underscores. */
+// OBSOLETE
+// OBSOLETE while (isdigit (*tokptr) || *tokptr == '_')
+// OBSOLETE {
+// OBSOLETE if (*tokptr != '_')
+// OBSOLETE {
+// OBSOLETE *copy++ = *tokptr;
+// OBSOLETE }
+// OBSOLETE tokptr++;
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Now accept either a '.', or one of [eEdD]. Dot is legal regardless
+// OBSOLETE of whether we found any leading digits, and we simply accept it and
+// OBSOLETE continue on to look for the fractional part and/or exponent. One of
+// OBSOLETE [eEdD] is legal only if we have seen digits, and means that there
+// OBSOLETE is no fractional part. If we find neither of these, then this is
+// OBSOLETE not a floating point number, so return failure. */
+// OBSOLETE
+// OBSOLETE switch (*tokptr++)
+// OBSOLETE {
+// OBSOLETE case '.':
+// OBSOLETE /* Accept and then look for fractional part and/or exponent. */
+// OBSOLETE *copy++ = '.';
+// OBSOLETE break;
+// OBSOLETE
+// OBSOLETE case 'e':
+// OBSOLETE case 'E':
+// OBSOLETE case 'd':
+// OBSOLETE case 'D':
+// OBSOLETE if (copy == buf)
+// OBSOLETE {
+// OBSOLETE return (0);
+// OBSOLETE }
+// OBSOLETE *copy++ = 'e';
+// OBSOLETE goto collect_exponent;
+// OBSOLETE break;
+// OBSOLETE
+// OBSOLETE default:
+// OBSOLETE return (0);
+// OBSOLETE break;
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* We found a '.', copy any fractional digits to the conversion buffer, up
+// OBSOLETE to the first nondigit, non-underscore character. */
+// OBSOLETE
+// OBSOLETE while (isdigit (*tokptr) || *tokptr == '_')
+// OBSOLETE {
+// OBSOLETE if (*tokptr != '_')
+// OBSOLETE {
+// OBSOLETE *copy++ = *tokptr;
+// OBSOLETE }
+// OBSOLETE tokptr++;
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Look for an exponent, which must start with one of [eEdD]. If none
+// OBSOLETE is found, jump directly to trying to convert what we have collected
+// OBSOLETE so far. */
+// OBSOLETE
+// OBSOLETE switch (*tokptr)
+// OBSOLETE {
+// OBSOLETE case 'e':
+// OBSOLETE case 'E':
+// OBSOLETE case 'd':
+// OBSOLETE case 'D':
+// OBSOLETE *copy++ = 'e';
+// OBSOLETE tokptr++;
+// OBSOLETE break;
+// OBSOLETE default:
+// OBSOLETE goto convert_float;
+// OBSOLETE break;
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Accept an optional '-' or '+' following one of [eEdD]. */
+// OBSOLETE
+// OBSOLETE collect_exponent:
+// OBSOLETE if (*tokptr == '+' || *tokptr == '-')
+// OBSOLETE {
+// OBSOLETE *copy++ = *tokptr++;
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Now copy an exponent into the conversion buffer. Note that at the
+// OBSOLETE moment underscores are *not* allowed in exponents. */
+// OBSOLETE
+// OBSOLETE while (isdigit (*tokptr))
+// OBSOLETE {
+// OBSOLETE *copy++ = *tokptr++;
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* If we transfered any chars to the conversion buffer, try to interpret its
+// OBSOLETE contents as a floating point value. If any characters remain, then we
+// OBSOLETE must not have a valid floating point string. */
+// OBSOLETE
+// OBSOLETE convert_float:
+// OBSOLETE *copy = '\0';
+// OBSOLETE if (copy != buf)
+// OBSOLETE {
+// OBSOLETE dval = strtod (buf, &copy);
+// OBSOLETE if (*copy == '\0')
+// OBSOLETE {
+// OBSOLETE yylval.dval = dval;
+// OBSOLETE lexptr = tokptr;
+// OBSOLETE return (FLOAT_LITERAL);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE return (0);
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Recognize a string literal. A string literal is a sequence
+// OBSOLETE of characters enclosed in matching single or double quotes, except that
+// OBSOLETE a single character inside single quotes is a character literal, which
+// OBSOLETE we reject as a string literal. To embed the terminator character inside
+// OBSOLETE a string, it is simply doubled (I.E. "this""is""one""string") */
+// OBSOLETE
+// OBSOLETE static enum ch_terminal
+// OBSOLETE match_string_literal (void)
+// OBSOLETE {
+// OBSOLETE char *tokptr = lexptr;
+// OBSOLETE int in_ctrlseq = 0;
+// OBSOLETE LONGEST ival;
+// OBSOLETE
+// OBSOLETE for (tempbufindex = 0, tokptr++; *tokptr != '\0'; tokptr++)
+// OBSOLETE {
+// OBSOLETE CHECKBUF (1);
+// OBSOLETE tryagain:;
+// OBSOLETE if (in_ctrlseq)
+// OBSOLETE {
+// OBSOLETE /* skip possible whitespaces */
+// OBSOLETE while ((*tokptr == ' ' || *tokptr == '\t') && *tokptr)
+// OBSOLETE tokptr++;
+// OBSOLETE if (*tokptr == ')')
+// OBSOLETE {
+// OBSOLETE in_ctrlseq = 0;
+// OBSOLETE tokptr++;
+// OBSOLETE goto tryagain;
+// OBSOLETE }
+// OBSOLETE else if (*tokptr != ',')
+// OBSOLETE error ("Invalid control sequence");
+// OBSOLETE tokptr++;
+// OBSOLETE /* skip possible whitespaces */
+// OBSOLETE while ((*tokptr == ' ' || *tokptr == '\t') && *tokptr)
+// OBSOLETE tokptr++;
+// OBSOLETE if (!decode_integer_literal (&ival, &tokptr))
+// OBSOLETE error ("Invalid control sequence");
+// OBSOLETE tokptr--;
+// OBSOLETE }
+// OBSOLETE else if (*tokptr == *lexptr)
+// OBSOLETE {
+// OBSOLETE if (*(tokptr + 1) == *lexptr)
+// OBSOLETE {
+// OBSOLETE ival = *tokptr++;
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE {
+// OBSOLETE break;
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE else if (*tokptr == '^')
+// OBSOLETE {
+// OBSOLETE if (*(tokptr + 1) == '(')
+// OBSOLETE {
+// OBSOLETE in_ctrlseq = 1;
+// OBSOLETE tokptr += 2;
+// OBSOLETE if (!decode_integer_literal (&ival, &tokptr))
+// OBSOLETE error ("Invalid control sequence");
+// OBSOLETE tokptr--;
+// OBSOLETE }
+// OBSOLETE else if (*(tokptr + 1) == '^')
+// OBSOLETE ival = *tokptr++;
+// OBSOLETE else
+// OBSOLETE error ("Invalid control sequence");
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE ival = *tokptr;
+// OBSOLETE tempbuf[tempbufindex++] = ival;
+// OBSOLETE }
+// OBSOLETE if (in_ctrlseq)
+// OBSOLETE error ("Invalid control sequence");
+// OBSOLETE
+// OBSOLETE if (*tokptr == '\0' /* no terminator */
+// OBSOLETE || (tempbufindex == 1 && *tokptr == '\'')) /* char literal */
+// OBSOLETE {
+// OBSOLETE return (0);
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE {
+// OBSOLETE tempbuf[tempbufindex] = '\0';
+// OBSOLETE yylval.sval.ptr = tempbuf;
+// OBSOLETE yylval.sval.length = tempbufindex;
+// OBSOLETE lexptr = ++tokptr;
+// OBSOLETE return (CHARACTER_STRING_LITERAL);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Recognize a character literal. A character literal is single character
+// OBSOLETE or a control sequence, enclosed in single quotes. A control sequence
+// OBSOLETE is a comma separated list of one or more integer literals, enclosed
+// OBSOLETE in parenthesis and introduced with a circumflex character.
+// OBSOLETE
+// OBSOLETE EX: 'a' '^(7)' '^(7,8)'
+// OBSOLETE
+// OBSOLETE As a GNU chill extension, the syntax C'xx' is also recognized as a
+// OBSOLETE character literal, where xx is a hex value for the character.
+// OBSOLETE
+// OBSOLETE Note that more than a single character, enclosed in single quotes, is
+// OBSOLETE a string literal.
+// OBSOLETE
+// OBSOLETE Returns CHARACTER_LITERAL if a match is found.
+// OBSOLETE */
+// OBSOLETE
+// OBSOLETE static enum ch_terminal
+// OBSOLETE match_character_literal (void)
+// OBSOLETE {
+// OBSOLETE char *tokptr = lexptr;
+// OBSOLETE LONGEST ival = 0;
+// OBSOLETE
+// OBSOLETE if ((*tokptr == 'c' || *tokptr == 'C') && (*(tokptr + 1) == '\''))
+// OBSOLETE {
+// OBSOLETE /* We have a GNU chill extension form, so skip the leading "C'",
+// OBSOLETE decode the hex value, and then ensure that we have a trailing
+// OBSOLETE single quote character. */
+// OBSOLETE tokptr += 2;
+// OBSOLETE if (!decode_integer_value (16, &tokptr, &ival) || (*tokptr != '\''))
+// OBSOLETE {
+// OBSOLETE return (0);
+// OBSOLETE }
+// OBSOLETE tokptr++;
+// OBSOLETE }
+// OBSOLETE else if (*tokptr == '\'')
+// OBSOLETE {
+// OBSOLETE tokptr++;
+// OBSOLETE
+// OBSOLETE /* Determine which form we have, either a control sequence or the
+// OBSOLETE single character form. */
+// OBSOLETE
+// OBSOLETE if (*tokptr == '^')
+// OBSOLETE {
+// OBSOLETE if (*(tokptr + 1) == '(')
+// OBSOLETE {
+// OBSOLETE /* Match and decode a control sequence. Return zero if we don't
+// OBSOLETE find a valid integer literal, or if the next unconsumed character
+// OBSOLETE after the integer literal is not the trailing ')'. */
+// OBSOLETE tokptr += 2;
+// OBSOLETE if (!decode_integer_literal (&ival, &tokptr) || (*tokptr++ != ')'))
+// OBSOLETE {
+// OBSOLETE return (0);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE else if (*(tokptr + 1) == '^')
+// OBSOLETE {
+// OBSOLETE ival = *tokptr;
+// OBSOLETE tokptr += 2;
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE /* fail */
+// OBSOLETE error ("Invalid control sequence");
+// OBSOLETE }
+// OBSOLETE else if (*tokptr == '\'')
+// OBSOLETE {
+// OBSOLETE /* this must be duplicated */
+// OBSOLETE ival = *tokptr;
+// OBSOLETE tokptr += 2;
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE {
+// OBSOLETE ival = *tokptr++;
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* The trailing quote has not yet been consumed. If we don't find
+// OBSOLETE it, then we have no match. */
+// OBSOLETE
+// OBSOLETE if (*tokptr++ != '\'')
+// OBSOLETE {
+// OBSOLETE return (0);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE {
+// OBSOLETE /* Not a character literal. */
+// OBSOLETE return (0);
+// OBSOLETE }
+// OBSOLETE yylval.typed_val.val = ival;
+// OBSOLETE yylval.typed_val.type = builtin_type_chill_char;
+// OBSOLETE lexptr = tokptr;
+// OBSOLETE return (CHARACTER_LITERAL);
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Recognize an integer literal, as specified in Z.200 sec 5.2.4.2.
+// OBSOLETE Note that according to 5.2.4.2, a single "_" is also a valid integer
+// OBSOLETE literal, however GNU-chill requires there to be at least one "digit"
+// OBSOLETE in any integer literal. */
+// OBSOLETE
+// OBSOLETE static enum ch_terminal
+// OBSOLETE match_integer_literal (void)
+// OBSOLETE {
+// OBSOLETE char *tokptr = lexptr;
+// OBSOLETE LONGEST ival;
+// OBSOLETE
+// OBSOLETE if (!decode_integer_literal (&ival, &tokptr))
+// OBSOLETE {
+// OBSOLETE return (0);
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE {
+// OBSOLETE yylval.typed_val.val = ival;
+// OBSOLETE #if defined(CC_HAS_LONG_LONG)
+// OBSOLETE if (ival > (LONGEST) 2147483647U || ival < -(LONGEST) 2147483648U)
+// OBSOLETE yylval.typed_val.type = builtin_type_long_long;
+// OBSOLETE else
+// OBSOLETE #endif
+// OBSOLETE yylval.typed_val.type = builtin_type_int;
+// OBSOLETE lexptr = tokptr;
+// OBSOLETE return (INTEGER_LITERAL);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Recognize a bit-string literal, as specified in Z.200 sec 5.2.4.8
+// OBSOLETE Note that according to 5.2.4.8, a single "_" is also a valid bit-string
+// OBSOLETE literal, however GNU-chill requires there to be at least one "digit"
+// OBSOLETE in any bit-string literal. */
+// OBSOLETE
+// OBSOLETE static enum ch_terminal
+// OBSOLETE match_bitstring_literal (void)
+// OBSOLETE {
+// OBSOLETE register char *tokptr = lexptr;
+// OBSOLETE int bitoffset = 0;
+// OBSOLETE int bitcount = 0;
+// OBSOLETE int bits_per_char;
+// OBSOLETE int digit;
+// OBSOLETE
+// OBSOLETE tempbufindex = 0;
+// OBSOLETE CHECKBUF (1);
+// OBSOLETE tempbuf[0] = 0;
+// OBSOLETE
+// OBSOLETE /* Look for the required explicit base specifier. */
+// OBSOLETE
+// OBSOLETE switch (*tokptr++)
+// OBSOLETE {
+// OBSOLETE case 'b':
+// OBSOLETE case 'B':
+// OBSOLETE bits_per_char = 1;
+// OBSOLETE break;
+// OBSOLETE case 'o':
+// OBSOLETE case 'O':
+// OBSOLETE bits_per_char = 3;
+// OBSOLETE break;
+// OBSOLETE case 'h':
+// OBSOLETE case 'H':
+// OBSOLETE bits_per_char = 4;
+// OBSOLETE break;
+// OBSOLETE default:
+// OBSOLETE return (0);
+// OBSOLETE break;
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Ensure that the character after the explicit base is a single quote. */
+// OBSOLETE
+// OBSOLETE if (*tokptr++ != '\'')
+// OBSOLETE {
+// OBSOLETE return (0);
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE while (*tokptr != '\0' && *tokptr != '\'')
+// OBSOLETE {
+// OBSOLETE digit = *tokptr;
+// OBSOLETE if (isupper (digit))
+// OBSOLETE digit = tolower (digit);
+// OBSOLETE tokptr++;
+// OBSOLETE switch (digit)
+// OBSOLETE {
+// OBSOLETE case '_':
+// OBSOLETE continue;
+// OBSOLETE case '0':
+// OBSOLETE case '1':
+// OBSOLETE case '2':
+// OBSOLETE case '3':
+// OBSOLETE case '4':
+// OBSOLETE case '5':
+// OBSOLETE case '6':
+// OBSOLETE case '7':
+// OBSOLETE case '8':
+// OBSOLETE case '9':
+// OBSOLETE digit -= '0';
+// OBSOLETE break;
+// OBSOLETE case 'a':
+// OBSOLETE case 'b':
+// OBSOLETE case 'c':
+// OBSOLETE case 'd':
+// OBSOLETE case 'e':
+// OBSOLETE case 'f':
+// OBSOLETE digit -= 'a';
+// OBSOLETE digit += 10;
+// OBSOLETE break;
+// OBSOLETE default:
+// OBSOLETE /* this is not a bitstring literal, probably an integer */
+// OBSOLETE return 0;
+// OBSOLETE }
+// OBSOLETE if (digit >= 1 << bits_per_char)
+// OBSOLETE {
+// OBSOLETE /* Found something not in domain for current base. */
+// OBSOLETE error ("Too-large digit in bitstring or integer.");
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE {
+// OBSOLETE /* Extract bits from digit, packing them into the bitstring byte. */
+// OBSOLETE int k = TARGET_BYTE_ORDER == BFD_ENDIAN_BIG ? bits_per_char - 1 : 0;
+// OBSOLETE for (; TARGET_BYTE_ORDER == BFD_ENDIAN_BIG ? k >= 0 : k < bits_per_char;
+// OBSOLETE TARGET_BYTE_ORDER == BFD_ENDIAN_BIG ? k-- : k++)
+// OBSOLETE {
+// OBSOLETE bitcount++;
+// OBSOLETE if (digit & (1 << k))
+// OBSOLETE {
+// OBSOLETE tempbuf[tempbufindex] |=
+// OBSOLETE (TARGET_BYTE_ORDER == BFD_ENDIAN_BIG)
+// OBSOLETE ? (1 << (HOST_CHAR_BIT - 1 - bitoffset))
+// OBSOLETE : (1 << bitoffset);
+// OBSOLETE }
+// OBSOLETE bitoffset++;
+// OBSOLETE if (bitoffset == HOST_CHAR_BIT)
+// OBSOLETE {
+// OBSOLETE bitoffset = 0;
+// OBSOLETE tempbufindex++;
+// OBSOLETE CHECKBUF (1);
+// OBSOLETE tempbuf[tempbufindex] = 0;
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Verify that we consumed everything up to the trailing single quote,
+// OBSOLETE and that we found some bits (IE not just underbars). */
+// OBSOLETE
+// OBSOLETE if (*tokptr++ != '\'')
+// OBSOLETE {
+// OBSOLETE return (0);
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE {
+// OBSOLETE yylval.sval.ptr = tempbuf;
+// OBSOLETE yylval.sval.length = bitcount;
+// OBSOLETE lexptr = tokptr;
+// OBSOLETE return (BIT_STRING_LITERAL);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE struct token
+// OBSOLETE {
+// OBSOLETE char *operator;
+// OBSOLETE int token;
+// OBSOLETE };
+// OBSOLETE
+// OBSOLETE static const struct token idtokentab[] =
+// OBSOLETE {
+// OBSOLETE {"array", ARRAY},
+// OBSOLETE {"length", LENGTH},
+// OBSOLETE {"lower", LOWER},
+// OBSOLETE {"upper", UPPER},
+// OBSOLETE {"andif", ANDIF},
+// OBSOLETE {"pred", PRED},
+// OBSOLETE {"succ", SUCC},
+// OBSOLETE {"card", CARD},
+// OBSOLETE {"size", SIZE},
+// OBSOLETE {"orif", ORIF},
+// OBSOLETE {"num", NUM},
+// OBSOLETE {"abs", ABS},
+// OBSOLETE {"max", MAX_TOKEN},
+// OBSOLETE {"min", MIN_TOKEN},
+// OBSOLETE {"mod", MOD},
+// OBSOLETE {"rem", REM},
+// OBSOLETE {"not", NOT},
+// OBSOLETE {"xor", LOGXOR},
+// OBSOLETE {"and", LOGAND},
+// OBSOLETE {"in", IN},
+// OBSOLETE {"or", LOGIOR},
+// OBSOLETE {"up", UP},
+// OBSOLETE {"addr", ADDR_TOKEN},
+// OBSOLETE {"null", EMPTINESS_LITERAL}
+// OBSOLETE };
+// OBSOLETE
+// OBSOLETE static const struct token tokentab2[] =
+// OBSOLETE {
+// OBSOLETE {":=", GDB_ASSIGNMENT},
+// OBSOLETE {"//", SLASH_SLASH},
+// OBSOLETE {"->", POINTER},
+// OBSOLETE {"/=", NOTEQUAL},
+// OBSOLETE {"<=", LEQ},
+// OBSOLETE {">=", GEQ}
+// OBSOLETE };
+// OBSOLETE
+// OBSOLETE /* Read one token, getting characters through lexptr. */
+// OBSOLETE /* This is where we will check to make sure that the language and the
+// OBSOLETE operators used are compatible. */
+// OBSOLETE
+// OBSOLETE static enum ch_terminal
+// OBSOLETE ch_lex (void)
+// OBSOLETE {
+// OBSOLETE unsigned int i;
+// OBSOLETE enum ch_terminal token;
+// OBSOLETE char *inputname;
+// OBSOLETE struct symbol *sym;
+// OBSOLETE
+// OBSOLETE /* Skip over any leading whitespace. */
+// OBSOLETE while (isspace (*lexptr))
+// OBSOLETE {
+// OBSOLETE lexptr++;
+// OBSOLETE }
+// OBSOLETE /* Look for special single character cases which can't be the first
+// OBSOLETE character of some other multicharacter token. */
+// OBSOLETE switch (*lexptr)
+// OBSOLETE {
+// OBSOLETE case '\0':
+// OBSOLETE return END_TOKEN;
+// OBSOLETE case ',':
+// OBSOLETE case '=':
+// OBSOLETE case ';':
+// OBSOLETE case '!':
+// OBSOLETE case '+':
+// OBSOLETE case '*':
+// OBSOLETE case '(':
+// OBSOLETE case ')':
+// OBSOLETE case '[':
+// OBSOLETE case ']':
+// OBSOLETE return (*lexptr++);
+// OBSOLETE }
+// OBSOLETE /* Look for characters which start a particular kind of multicharacter
+// OBSOLETE token, such as a character literal, register name, convenience
+// OBSOLETE variable name, string literal, etc. */
+// OBSOLETE switch (*lexptr)
+// OBSOLETE {
+// OBSOLETE case '\'':
+// OBSOLETE case '\"':
+// OBSOLETE /* First try to match a string literal, which is any
+// OBSOLETE sequence of characters enclosed in matching single or double
+// OBSOLETE quotes, except that a single character inside single quotes
+// OBSOLETE is a character literal, so we have to catch that case also. */
+// OBSOLETE token = match_string_literal ();
+// OBSOLETE if (token != 0)
+// OBSOLETE {
+// OBSOLETE return (token);
+// OBSOLETE }
+// OBSOLETE if (*lexptr == '\'')
+// OBSOLETE {
+// OBSOLETE token = match_character_literal ();
+// OBSOLETE if (token != 0)
+// OBSOLETE {
+// OBSOLETE return (token);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE break;
+// OBSOLETE case 'C':
+// OBSOLETE case 'c':
+// OBSOLETE token = match_character_literal ();
+// OBSOLETE if (token != 0)
+// OBSOLETE {
+// OBSOLETE return (token);
+// OBSOLETE }
+// OBSOLETE break;
+// OBSOLETE case '$':
+// OBSOLETE yylval.sval.ptr = lexptr;
+// OBSOLETE do
+// OBSOLETE {
+// OBSOLETE lexptr++;
+// OBSOLETE }
+// OBSOLETE while (isalnum (*lexptr) || *lexptr == '_' || *lexptr == '$');
+// OBSOLETE yylval.sval.length = lexptr - yylval.sval.ptr;
+// OBSOLETE write_dollar_variable (yylval.sval);
+// OBSOLETE return GDB_VARIABLE;
+// OBSOLETE break;
+// OBSOLETE }
+// OBSOLETE /* See if it is a special token of length 2. */
+// OBSOLETE for (i = 0; i < sizeof (tokentab2) / sizeof (tokentab2[0]); i++)
+// OBSOLETE {
+// OBSOLETE if (STREQN (lexptr, tokentab2[i].operator, 2))
+// OBSOLETE {
+// OBSOLETE lexptr += 2;
+// OBSOLETE return (tokentab2[i].token);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE /* Look for single character cases which which could be the first
+// OBSOLETE character of some other multicharacter token, but aren't, or we
+// OBSOLETE would already have found it. */
+// OBSOLETE switch (*lexptr)
+// OBSOLETE {
+// OBSOLETE case '-':
+// OBSOLETE case ':':
+// OBSOLETE case '/':
+// OBSOLETE case '<':
+// OBSOLETE case '>':
+// OBSOLETE return (*lexptr++);
+// OBSOLETE }
+// OBSOLETE /* Look for a float literal before looking for an integer literal, so
+// OBSOLETE we match as much of the input stream as possible. */
+// OBSOLETE token = match_float_literal ();
+// OBSOLETE if (token != 0)
+// OBSOLETE {
+// OBSOLETE return (token);
+// OBSOLETE }
+// OBSOLETE token = match_bitstring_literal ();
+// OBSOLETE if (token != 0)
+// OBSOLETE {
+// OBSOLETE return (token);
+// OBSOLETE }
+// OBSOLETE token = match_integer_literal ();
+// OBSOLETE if (token != 0)
+// OBSOLETE {
+// OBSOLETE return (token);
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Try to match a simple name string, and if a match is found, then
+// OBSOLETE further classify what sort of name it is and return an appropriate
+// OBSOLETE token. Note that attempting to match a simple name string consumes
+// OBSOLETE the token from lexptr, so we can't back out if we later find that
+// OBSOLETE we can't classify what sort of name it is. */
+// OBSOLETE
+// OBSOLETE inputname = match_simple_name_string ();
+// OBSOLETE
+// OBSOLETE if (inputname != NULL)
+// OBSOLETE {
+// OBSOLETE char *simplename = (char *) alloca (strlen (inputname) + 1);
+// OBSOLETE
+// OBSOLETE char *dptr = simplename, *sptr = inputname;
+// OBSOLETE for (; *sptr; sptr++)
+// OBSOLETE *dptr++ = isupper (*sptr) ? tolower (*sptr) : *sptr;
+// OBSOLETE *dptr = '\0';
+// OBSOLETE
+// OBSOLETE /* See if it is a reserved identifier. */
+// OBSOLETE for (i = 0; i < sizeof (idtokentab) / sizeof (idtokentab[0]); i++)
+// OBSOLETE {
+// OBSOLETE if (STREQ (simplename, idtokentab[i].operator))
+// OBSOLETE {
+// OBSOLETE return (idtokentab[i].token);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Look for other special tokens. */
+// OBSOLETE if (STREQ (simplename, "true"))
+// OBSOLETE {
+// OBSOLETE yylval.ulval = 1;
+// OBSOLETE return (BOOLEAN_LITERAL);
+// OBSOLETE }
+// OBSOLETE if (STREQ (simplename, "false"))
+// OBSOLETE {
+// OBSOLETE yylval.ulval = 0;
+// OBSOLETE return (BOOLEAN_LITERAL);
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE sym = lookup_symbol (inputname, expression_context_block,
+// OBSOLETE VAR_NAMESPACE, (int *) NULL,
+// OBSOLETE (struct symtab **) NULL);
+// OBSOLETE if (sym == NULL && strcmp (inputname, simplename) != 0)
+// OBSOLETE {
+// OBSOLETE sym = lookup_symbol (simplename, expression_context_block,
+// OBSOLETE VAR_NAMESPACE, (int *) NULL,
+// OBSOLETE (struct symtab **) NULL);
+// OBSOLETE }
+// OBSOLETE if (sym != NULL)
+// OBSOLETE {
+// OBSOLETE yylval.ssym.stoken.ptr = NULL;
+// OBSOLETE yylval.ssym.stoken.length = 0;
+// OBSOLETE yylval.ssym.sym = sym;
+// OBSOLETE yylval.ssym.is_a_field_of_this = 0; /* FIXME, C++'ism */
+// OBSOLETE switch (SYMBOL_CLASS (sym))
+// OBSOLETE {
+// OBSOLETE case LOC_BLOCK:
+// OBSOLETE /* Found a procedure name. */
+// OBSOLETE return (GENERAL_PROCEDURE_NAME);
+// OBSOLETE case LOC_STATIC:
+// OBSOLETE /* Found a global or local static variable. */
+// OBSOLETE return (LOCATION_NAME);
+// OBSOLETE case LOC_REGISTER:
+// OBSOLETE case LOC_ARG:
+// OBSOLETE case LOC_REF_ARG:
+// OBSOLETE case LOC_REGPARM:
+// OBSOLETE case LOC_REGPARM_ADDR:
+// OBSOLETE case LOC_LOCAL:
+// OBSOLETE case LOC_LOCAL_ARG:
+// OBSOLETE case LOC_BASEREG:
+// OBSOLETE case LOC_BASEREG_ARG:
+// OBSOLETE if (innermost_block == NULL
+// OBSOLETE || contained_in (block_found, innermost_block))
+// OBSOLETE {
+// OBSOLETE innermost_block = block_found;
+// OBSOLETE }
+// OBSOLETE return (LOCATION_NAME);
+// OBSOLETE break;
+// OBSOLETE case LOC_CONST:
+// OBSOLETE case LOC_LABEL:
+// OBSOLETE return (LOCATION_NAME);
+// OBSOLETE break;
+// OBSOLETE case LOC_TYPEDEF:
+// OBSOLETE yylval.tsym.type = SYMBOL_TYPE (sym);
+// OBSOLETE return TYPENAME;
+// OBSOLETE case LOC_UNDEF:
+// OBSOLETE case LOC_CONST_BYTES:
+// OBSOLETE case LOC_OPTIMIZED_OUT:
+// OBSOLETE error ("Symbol \"%s\" names no location.", inputname);
+// OBSOLETE break;
+// OBSOLETE default:
+// OBSOLETE internal_error (__FILE__, __LINE__,
+// OBSOLETE "unhandled SYMBOL_CLASS in ch_lex()");
+// OBSOLETE break;
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE else if (!have_full_symbols () && !have_partial_symbols ())
+// OBSOLETE {
+// OBSOLETE error ("No symbol table is loaded. Use the \"file\" command.");
+// OBSOLETE }
+// OBSOLETE else
+// OBSOLETE {
+// OBSOLETE error ("No symbol \"%s\" in current context.", inputname);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE /* Catch single character tokens which are not part of some
+// OBSOLETE longer token. */
+// OBSOLETE
+// OBSOLETE switch (*lexptr)
+// OBSOLETE {
+// OBSOLETE case '.': /* Not float for example. */
+// OBSOLETE lexptr++;
+// OBSOLETE while (isspace (*lexptr))
+// OBSOLETE lexptr++;
+// OBSOLETE inputname = match_simple_name_string ();
+// OBSOLETE if (!inputname)
+// OBSOLETE return '.';
+// OBSOLETE return DOT_FIELD_NAME;
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE return (ILLEGAL_TOKEN);
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE static void
+// OBSOLETE write_lower_upper_value (enum exp_opcode opcode, /* Either UNOP_LOWER or UNOP_UPPER */
+// OBSOLETE struct type *type)
+// OBSOLETE {
+// OBSOLETE if (type == NULL)
+// OBSOLETE write_exp_elt_opcode (opcode);
+// OBSOLETE else
+// OBSOLETE {
+// OBSOLETE struct type *result_type;
+// OBSOLETE LONGEST val = type_lower_upper (opcode, type, &result_type);
+// OBSOLETE write_exp_elt_opcode (OP_LONG);
+// OBSOLETE write_exp_elt_type (result_type);
+// OBSOLETE write_exp_elt_longcst (val);
+// OBSOLETE write_exp_elt_opcode (OP_LONG);
+// OBSOLETE }
+// OBSOLETE }
+// OBSOLETE
+// OBSOLETE void
+// OBSOLETE chill_error (char *msg)
+// OBSOLETE {
+// OBSOLETE /* Never used. */
+// OBSOLETE }