diff options
author | David Malcolm <dmalcolm@redhat.com> | 2023-02-10 18:10:21 -0500 |
---|---|---|
committer | David Malcolm <dmalcolm@redhat.com> | 2023-02-10 18:10:21 -0500 |
commit | aa601e30758581837c9ca7b738ec2810a18350f5 (patch) | |
tree | 5bc184795ea6a8607334f310a0f4bcb8ce053c17 /gcc/analyzer/sm-malloc.cc | |
parent | 305037ee3ed49641cc1db2a0e92e3eeb9a7ec2b5 (diff) | |
download | gcc-aa601e30758581837c9ca7b738ec2810a18350f5.zip gcc-aa601e30758581837c9ca7b738ec2810a18350f5.tar.gz gcc-aa601e30758581837c9ca7b738ec2810a18350f5.tar.bz2 |
analyzer: don't warn for deref-before-check for checks in macros [PR108745]
Integration testing shows this patch fixes all 9 known false positives
from -Wanalyzer-deref-before-check within ImageMagick-7.1.0-57, and
eliminates 34 further as-yet unassessed such diagnostics, without
eliminating the 1 known true positive.
This improves the rate of true positives for the warning from
1.56% to 4.76% of the total:
-Wanalyzer-deref-before-check: 1.56% -> 4.76% (GOOD: 1 BAD: 63->20)
TRUE: 1
FALSE: 15 -> 6 (-9)
ImageMagick-7.1.0-57: 9 -> 0 (-9)
TODO: 48 -> 14 (-34)
ImageMagick-7.1.0-57: 21 -> 1 (-20)
qemu-7.2.0: 25 -> 11 (-14)
gcc/analyzer/ChangeLog:
PR analyzer/108745
* sm-malloc.cc (deref_before_check::emit): Reject the warning if
the check occurs within a macro defintion.
gcc/testsuite/ChangeLog:
PR analyzer/108745
* gcc.dg/analyzer/deref-before-check-macro-pr108745.c: New test.
* gcc.dg/analyzer/deref-before-check-macro.c: New test.
Signed-off-by: David Malcolm <dmalcolm@redhat.com>
Diffstat (limited to 'gcc/analyzer/sm-malloc.cc')
-rw-r--r-- | gcc/analyzer/sm-malloc.cc | 37 |
1 files changed, 37 insertions, 0 deletions
diff --git a/gcc/analyzer/sm-malloc.cc b/gcc/analyzer/sm-malloc.cc index 9aee810..c24fe73 100644 --- a/gcc/analyzer/sm-malloc.cc +++ b/gcc/analyzer/sm-malloc.cc @@ -1519,6 +1519,43 @@ public: != &m_check_enode->get_point ().get_call_string ()) return false; + /* Reject the warning if the check occurs within a macro defintion. + This avoids false positives for such code as: + + #define throw_error \ + do { \ + if (p) \ + cleanup (p); \ + return; \ + } while (0) + + if (p->idx >= n) + throw_error (); + + where the usage of "throw_error" implicitly adds a check + on 'p'. + + We do warn when the check is in a macro expansion if we can get + at the location of the condition and it is't part of the + definition, so that we warn for checks such as: + if (words[0][0] == '@') + return; + g_assert(words[0] != NULL); <--- here + Unfortunately we don't have locations for individual gimple + arguments, so in: + g_assert (ptr); + we merely have a gimple_cond + if (p_2(D) == 0B) + with no way of getting at the location of the condition separately + from that of the gimple_cond (where the "if" is within the macro + definition). We reject the warning for such cases. + + We do warn when the *deref* occurs in a macro, since this can be + a source of real bugs; see e.g. PR 77425. */ + location_t check_loc = m_check_enode->get_point ().get_location (); + if (linemap_location_from_macro_definition_p (line_table, check_loc)) + return false; + /* Reject the warning if the deref's BB doesn't dominate that of the check, so that we don't warn e.g. for shared cleanup code that checks a pointer for NULL, when that code is sometimes |