/* Bison parser for Rust expressions, for GDB.
Copyright (C) 2016-2017 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 3 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, see . */
/* Removing the last conflict seems difficult. */
%expect 1
%{
#include "defs.h"
#include "block.h"
#include "charset.h"
#include "cp-support.h"
#include "gdb_obstack.h"
#include "gdb_regex.h"
#include "rust-lang.h"
#include "parser-defs.h"
#include "selftest.h"
#include "value.h"
#include "vec.h"
#define GDB_YY_REMAP_PREFIX rust
#include "yy-remap.h"
#define RUSTSTYPE YYSTYPE
extern initialize_file_ftype _initialize_rust_exp;
struct rust_op;
typedef const struct rust_op *rust_op_ptr;
DEF_VEC_P (rust_op_ptr);
/* A typed integer constant. */
struct typed_val_int
{
LONGEST val;
struct type *type;
};
/* A typed floating point constant. */
struct typed_val_float
{
DOUBLEST dval;
struct type *type;
};
/* An identifier and an expression. This is used to represent one
element of a struct initializer. */
struct set_field
{
struct stoken name;
const struct rust_op *init;
};
typedef struct set_field set_field;
DEF_VEC_O (set_field);
static int rustyylex (void);
static void rust_push_back (char c);
static const char *rust_copy_name (const char *, int);
static struct stoken rust_concat3 (const char *, const char *, const char *);
static struct stoken make_stoken (const char *);
static struct block_symbol rust_lookup_symbol (const char *name,
const struct block *block,
const domain_enum domain);
static struct type *rust_lookup_type (const char *name,
const struct block *block);
static struct type *rust_type (const char *name);
static const struct rust_op *crate_name (const struct rust_op *name);
static const struct rust_op *super_name (const struct rust_op *name,
unsigned int n_supers);
static const struct rust_op *ast_operation (enum exp_opcode opcode,
const struct rust_op *left,
const struct rust_op *right);
static const struct rust_op *ast_compound_assignment
(enum exp_opcode opcode, const struct rust_op *left,
const struct rust_op *rust_op);
static const struct rust_op *ast_literal (struct typed_val_int val);
static const struct rust_op *ast_dliteral (struct typed_val_float val);
static const struct rust_op *ast_structop (const struct rust_op *left,
const char *name,
int completing);
static const struct rust_op *ast_structop_anonymous
(const struct rust_op *left, struct typed_val_int number);
static const struct rust_op *ast_unary (enum exp_opcode opcode,
const struct rust_op *expr);
static const struct rust_op *ast_cast (const struct rust_op *expr,
const struct rust_op *type);
static const struct rust_op *ast_call_ish (enum exp_opcode opcode,
const struct rust_op *expr,
VEC (rust_op_ptr) **params);
static const struct rust_op *ast_path (struct stoken name,
VEC (rust_op_ptr) **params);
static const struct rust_op *ast_string (struct stoken str);
static const struct rust_op *ast_struct (const struct rust_op *name,
VEC (set_field) **fields);
static const struct rust_op *ast_range (const struct rust_op *lhs,
const struct rust_op *rhs);
static const struct rust_op *ast_array_type (const struct rust_op *lhs,
struct typed_val_int val);
static const struct rust_op *ast_slice_type (const struct rust_op *type);
static const struct rust_op *ast_reference_type (const struct rust_op *type);
static const struct rust_op *ast_pointer_type (const struct rust_op *type,
int is_mut);
static const struct rust_op *ast_function_type (const struct rust_op *result,
VEC (rust_op_ptr) **params);
static const struct rust_op *ast_tuple_type (VEC (rust_op_ptr) **params);
/* The state of the parser, used internally when we are parsing the
expression. */
static struct parser_state *pstate = NULL;
/* A regular expression for matching Rust numbers. This is split up
since it is very long and this gives us a way to comment the
sections. */
static const char *number_regex_text =
/* subexpression 1: allows use of alternation, otherwise uninteresting */
"^("
/* First comes floating point. */
/* Recognize number after the decimal point, with optional
exponent and optional type suffix.
subexpression 2: allows "?", otherwise uninteresting
subexpression 3: if present, type suffix
*/
"[0-9][0-9_]*\\.[0-9][0-9_]*([eE][-+]?[0-9][0-9_]*)?(f32|f64)?"
#define FLOAT_TYPE1 3
"|"
/* Recognize exponent without decimal point, with optional type
suffix.
subexpression 4: if present, type suffix
*/
#define FLOAT_TYPE2 4
"[0-9][0-9_]*[eE][-+]?[0-9][0-9_]*(f32|f64)?"
"|"
/* "23." is a valid floating point number, but "23.e5" and
"23.f32" are not. So, handle the trailing-. case
separately. */
"[0-9][0-9_]*\\."
"|"
/* Finally come integers.
subexpression 5: text of integer
subexpression 6: if present, type suffix
subexpression 7: allows use of alternation, otherwise uninteresting
*/
#define INT_TEXT 5
#define INT_TYPE 6
"(0x[a-fA-F0-9_]+|0o[0-7_]+|0b[01_]+|[0-9][0-9_]*)"
"([iu](size|8|16|32|64))?"
")";
/* The number of subexpressions to allocate space for, including the
"0th" whole match subexpression. */
#define NUM_SUBEXPRESSIONS 8
/* The compiled number-matching regex. */
static regex_t number_regex;
/* True if we're running unit tests. */
static int unit_testing;
/* Obstack for data temporarily allocated during parsing. */
static auto_obstack work_obstack;
/* Result of parsing. Points into work_obstack. */
static const struct rust_op *rust_ast;
%}
%union
{
/* A typed integer constant. */
struct typed_val_int typed_val_int;
/* A typed floating point constant. */
struct typed_val_float typed_val_float;
/* An identifier or string. */
struct stoken sval;
/* A token representing an opcode, like "==". */
enum exp_opcode opcode;
/* A list of expressions; for example, the arguments to a function
call. */
VEC (rust_op_ptr) **params;
/* A list of field initializers. */
VEC (set_field) **field_inits;
/* A single field initializer. */
struct set_field one_field_init;
/* An expression. */
const struct rust_op *op;
/* A plain integer, for example used to count the number of
"super::" prefixes on a path. */
unsigned int depth;
}
%{
/* Rust AST operations. We build a tree of these; then lower them
to gdb expressions when parsing has completed. */
struct rust_op
{
/* The opcode. */
enum exp_opcode opcode;
/* If OPCODE is OP_TYPE, then this holds information about what type
is described by this node. */
enum type_code typecode;
/* Indicates whether OPCODE actually represents a compound
assignment. For example, if OPCODE is GTGT and this is false,
then this rust_op represents an ordinary ">>"; but if this is
true, then this rust_op represents ">>=". Unused in other
cases. */
unsigned int compound_assignment : 1;
/* Only used by a field expression; if set, indicates that the field
name occurred at the end of the expression and is eligible for
completion. */
unsigned int completing : 1;
/* Operands of expression. Which one is used and how depends on the
particular opcode. */
RUSTSTYPE left;
RUSTSTYPE right;
};
%}
%token GDBVAR
%token IDENT
%token COMPLETE
%token INTEGER
%token DECIMAL_INTEGER
%token STRING
%token BYTESTRING
%token FLOAT
%token COMPOUND_ASSIGN
/* Keyword tokens. */
%token KW_AS
%token KW_IF
%token KW_TRUE
%token KW_FALSE
%token KW_SUPER
%token KW_SELF
%token KW_MUT
%token KW_EXTERN
%token KW_CONST
%token KW_FN
%token KW_SIZEOF
/* Operator tokens. */
%token DOTDOT
%token OROR
%token ANDAND
%token EQEQ
%token NOTEQ
%token LTEQ
%token GTEQ
%token LSH RSH
%token COLONCOLON
%token ARROW
%type type
%type path_for_expr
%type identifier_path_for_expr
%type path_for_type
%type identifier_path_for_type
%type just_identifiers_for_type
%type maybe_type_list
%type type_list
%type super_path
%type literal
%type expr
%type field_expr
%type idx_expr
%type unop_expr
%type binop_expr
%type binop_expr_expr
%type type_cast_expr
%type assignment_expr
%type compound_assignment_expr
%type paren_expr
%type call_expr
%type path_expr
%type tuple_expr
%type unit_expr
%type struct_expr
%type array_expr
%type range_expr
%type expr_list
%type maybe_expr_list
%type paren_expr_list
%type struct_expr_list
%type struct_expr_tail
/* Precedence. */
%nonassoc DOTDOT
%right '=' COMPOUND_ASSIGN
%left OROR
%left ANDAND
%nonassoc EQEQ NOTEQ '<' '>' LTEQ GTEQ
%left '|'
%left '^'
%left '&'
%left LSH RSH
%left '@'
%left '+' '-'
%left '*' '/' '%'
/* These could be %precedence in Bison, but that isn't a yacc
feature. */
%left KW_AS
%left UNARY
%left '[' '.' '('
%%
start:
expr
{
/* If we are completing and see a valid parse,
rust_ast will already have been set. */
if (rust_ast == NULL)
rust_ast = $1;
}
;
/* Note that the Rust grammar includes a method_call_expr, but we
handle this differently, to avoid a shift/reduce conflict with
call_expr. */
expr:
literal
| path_expr
| tuple_expr
| unit_expr
| struct_expr
| field_expr
| array_expr
| idx_expr
| range_expr
| unop_expr /* Must precede call_expr because of ambiguity with sizeof. */
| binop_expr
| paren_expr
| call_expr
;
tuple_expr:
'(' expr ',' maybe_expr_list ')'
{
VEC_safe_insert (rust_op_ptr, *$4, 0, $2);
error (_("Tuple expressions not supported yet"));
}
;
unit_expr:
'(' ')'
{
struct typed_val_int val;
val.type
= language_lookup_primitive_type (parse_language (pstate),
parse_gdbarch (pstate),
"()");
val.val = 0;
$$ = ast_literal (val);
}
;
/* To avoid a shift/reduce conflict with call_expr, we don't handle
tuple struct expressions here, but instead when examining the
AST. */
struct_expr:
path_for_expr '{' struct_expr_list '}'
{ $$ = ast_struct ($1, $3); }
;
struct_expr_tail:
DOTDOT expr
{
struct set_field sf;
sf.name.ptr = NULL;
sf.name.length = 0;
sf.init = $2;
$$ = sf;
}
| IDENT ':' expr
{
struct set_field sf;
sf.name = $1;
sf.init = $3;
$$ = sf;
}
;
struct_expr_list:
/* %empty */
{
VEC (set_field) **result
= OBSTACK_ZALLOC (&work_obstack, VEC (set_field) *);
$$ = result;
}
| struct_expr_tail
{
VEC (set_field) **result
= OBSTACK_ZALLOC (&work_obstack, VEC (set_field) *);
make_cleanup (VEC_cleanup (set_field), result);
VEC_safe_push (set_field, *result, &$1);
$$ = result;
}
| IDENT ':' expr ',' struct_expr_list
{
struct set_field sf;
sf.name = $1;
sf.init = $3;
VEC_safe_push (set_field, *$5, &sf);
$$ = $5;
}
;
array_expr:
'[' KW_MUT expr_list ']'
{ $$ = ast_call_ish (OP_ARRAY, NULL, $3); }
| '[' expr_list ']'
{ $$ = ast_call_ish (OP_ARRAY, NULL, $2); }
| '[' KW_MUT expr ';' expr ']'
{ $$ = ast_operation (OP_RUST_ARRAY, $3, $5); }
| '[' expr ';' expr ']'
{ $$ = ast_operation (OP_RUST_ARRAY, $2, $4); }
;
range_expr:
expr DOTDOT
{ $$ = ast_range ($1, NULL); }
| expr DOTDOT expr
{ $$ = ast_range ($1, $3); }
| DOTDOT expr
{ $$ = ast_range (NULL, $2); }
| DOTDOT
{ $$ = ast_range (NULL, NULL); }
;
literal:
INTEGER
{ $$ = ast_literal ($1); }
| DECIMAL_INTEGER
{ $$ = ast_literal ($1); }
| FLOAT
{ $$ = ast_dliteral ($1); }
| STRING
{
const struct rust_op *str = ast_string ($1);
VEC (set_field) **fields;
struct set_field field;
struct typed_val_int val;
struct stoken token;
fields = OBSTACK_ZALLOC (&work_obstack, VEC (set_field) *);
make_cleanup (VEC_cleanup (set_field), fields);
/* Wrap the raw string in the &str struct. */
field.name.ptr = "data_ptr";
field.name.length = strlen (field.name.ptr);
field.init = ast_unary (UNOP_ADDR, ast_string ($1));
VEC_safe_push (set_field, *fields, &field);
val.type = rust_type ("usize");
val.val = $1.length;
field.name.ptr = "length";
field.name.length = strlen (field.name.ptr);
field.init = ast_literal (val);
VEC_safe_push (set_field, *fields, &field);
token.ptr = "&str";
token.length = strlen (token.ptr);
$$ = ast_struct (ast_path (token, NULL), fields);
}
| BYTESTRING
{ $$ = ast_string ($1); }
| KW_TRUE
{
struct typed_val_int val;
val.type = language_bool_type (parse_language (pstate),
parse_gdbarch (pstate));
val.val = 1;
$$ = ast_literal (val);
}
| KW_FALSE
{
struct typed_val_int val;
val.type = language_bool_type (parse_language (pstate),
parse_gdbarch (pstate));
val.val = 0;
$$ = ast_literal (val);
}
;
field_expr:
expr '.' IDENT
{ $$ = ast_structop ($1, $3.ptr, 0); }
| expr '.' COMPLETE
{
$$ = ast_structop ($1, $3.ptr, 1);
rust_ast = $$;
}
| expr '.' DECIMAL_INTEGER
{ $$ = ast_structop_anonymous ($1, $3); }
;
idx_expr:
expr '[' expr ']'
{ $$ = ast_operation (BINOP_SUBSCRIPT, $1, $3); }
;
unop_expr:
'+' expr %prec UNARY
{ $$ = ast_unary (UNOP_PLUS, $2); }
| '-' expr %prec UNARY
{ $$ = ast_unary (UNOP_NEG, $2); }
| '!' expr %prec UNARY
{
/* Note that we provide a Rust-specific evaluator
override for UNOP_COMPLEMENT, so it can do the
right thing for both bool and integral
values. */
$$ = ast_unary (UNOP_COMPLEMENT, $2);
}
| '*' expr %prec UNARY
{ $$ = ast_unary (UNOP_IND, $2); }
| '&' expr %prec UNARY
{ $$ = ast_unary (UNOP_ADDR, $2); }
| '&' KW_MUT expr %prec UNARY
{ $$ = ast_unary (UNOP_ADDR, $3); }
| KW_SIZEOF '(' expr ')' %prec UNARY
{ $$ = ast_unary (UNOP_SIZEOF, $3); }
;
binop_expr:
binop_expr_expr
| type_cast_expr
| assignment_expr
| compound_assignment_expr
;
binop_expr_expr:
expr '*' expr
{ $$ = ast_operation (BINOP_MUL, $1, $3); }
| expr '@' expr
{ $$ = ast_operation (BINOP_REPEAT, $1, $3); }
| expr '/' expr
{ $$ = ast_operation (BINOP_DIV, $1, $3); }
| expr '%' expr
{ $$ = ast_operation (BINOP_REM, $1, $3); }
| expr '<' expr
{ $$ = ast_operation (BINOP_LESS, $1, $3); }
| expr '>' expr
{ $$ = ast_operation (BINOP_GTR, $1, $3); }
| expr '&' expr
{ $$ = ast_operation (BINOP_BITWISE_AND, $1, $3); }
| expr '|' expr
{ $$ = ast_operation (BINOP_BITWISE_IOR, $1, $3); }
| expr '^' expr
{ $$ = ast_operation (BINOP_BITWISE_XOR, $1, $3); }
| expr '+' expr
{ $$ = ast_operation (BINOP_ADD, $1, $3); }
| expr '-' expr
{ $$ = ast_operation (BINOP_SUB, $1, $3); }
| expr OROR expr
{ $$ = ast_operation (BINOP_LOGICAL_OR, $1, $3); }
| expr ANDAND expr
{ $$ = ast_operation (BINOP_LOGICAL_AND, $1, $3); }
| expr EQEQ expr
{ $$ = ast_operation (BINOP_EQUAL, $1, $3); }
| expr NOTEQ expr
{ $$ = ast_operation (BINOP_NOTEQUAL, $1, $3); }
| expr LTEQ expr
{ $$ = ast_operation (BINOP_LEQ, $1, $3); }
| expr GTEQ expr
{ $$ = ast_operation (BINOP_GEQ, $1, $3); }
| expr LSH expr
{ $$ = ast_operation (BINOP_LSH, $1, $3); }
| expr RSH expr
{ $$ = ast_operation (BINOP_RSH, $1, $3); }
;
type_cast_expr:
expr KW_AS type
{ $$ = ast_cast ($1, $3); }
;
assignment_expr:
expr '=' expr
{ $$ = ast_operation (BINOP_ASSIGN, $1, $3); }
;
compound_assignment_expr:
expr COMPOUND_ASSIGN expr
{ $$ = ast_compound_assignment ($2, $1, $3); }
;
paren_expr:
'(' expr ')'
{ $$ = $2; }
;
expr_list:
expr
{
$$ = OBSTACK_ZALLOC (&work_obstack, VEC (rust_op_ptr) *);
make_cleanup (VEC_cleanup (rust_op_ptr), $$);
VEC_safe_push (rust_op_ptr, *$$, $1);
}
| expr_list ',' expr
{
VEC_safe_push (rust_op_ptr, *$1, $3);
$$ = $1;
}
;
maybe_expr_list:
/* %empty */
{
/* The result can't be NULL. */
$$ = OBSTACK_ZALLOC (&work_obstack, VEC (rust_op_ptr) *);
make_cleanup (VEC_cleanup (rust_op_ptr), $$);
}
| expr_list
{ $$ = $1; }
;
paren_expr_list:
'('
maybe_expr_list
')'
{ $$ = $2; }
;
call_expr:
expr paren_expr_list
{ $$ = ast_call_ish (OP_FUNCALL, $1, $2); }
;
maybe_self_path:
/* %empty */
| KW_SELF COLONCOLON
;
super_path:
KW_SUPER COLONCOLON
{ $$ = 1; }
| super_path KW_SUPER COLONCOLON
{ $$ = $1 + 1; }
;
path_expr:
path_for_expr
{ $$ = $1; }
| GDBVAR
{ $$ = ast_path ($1, NULL); }
| KW_SELF
{ $$ = ast_path (make_stoken ("self"), NULL); }
;
path_for_expr:
identifier_path_for_expr
| KW_SELF COLONCOLON identifier_path_for_expr
{ $$ = super_name ($3, 0); }
| maybe_self_path super_path identifier_path_for_expr
{ $$ = super_name ($3, $2); }
| COLONCOLON identifier_path_for_expr
{ $$ = crate_name ($2); }
| KW_EXTERN identifier_path_for_expr
{
/* This is a gdb extension to make it possible to
refer to items in other crates. It just bypasses
adding the current crate to the front of the
name. */
$$ = ast_path (rust_concat3 ("::", $2->left.sval.ptr, NULL),
$2->right.params);
}
;
identifier_path_for_expr:
IDENT
{ $$ = ast_path ($1, NULL); }
| identifier_path_for_expr COLONCOLON IDENT
{
$$ = ast_path (rust_concat3 ($1->left.sval.ptr, "::",
$3.ptr),
NULL);
}
| identifier_path_for_expr COLONCOLON '<' type_list '>'
{ $$ = ast_path ($1->left.sval, $4); }
| identifier_path_for_expr COLONCOLON '<' type_list RSH
{
$$ = ast_path ($1->left.sval, $4);
rust_push_back ('>');
}
;
path_for_type:
identifier_path_for_type
| KW_SELF COLONCOLON identifier_path_for_type
{ $$ = super_name ($3, 0); }
| maybe_self_path super_path identifier_path_for_type
{ $$ = super_name ($3, $2); }
| COLONCOLON identifier_path_for_type
{ $$ = crate_name ($2); }
| KW_EXTERN identifier_path_for_type
{
/* This is a gdb extension to make it possible to
refer to items in other crates. It just bypasses
adding the current crate to the front of the
name. */
$$ = ast_path (rust_concat3 ("::", $2->left.sval.ptr, NULL),
$2->right.params);
}
;
just_identifiers_for_type:
IDENT
{ $$ = ast_path ($1, NULL); }
| just_identifiers_for_type COLONCOLON IDENT
{
$$ = ast_path (rust_concat3 ($1->left.sval.ptr, "::",
$3.ptr),
NULL);
}
;
identifier_path_for_type:
just_identifiers_for_type
| just_identifiers_for_type '<' type_list '>'
{ $$ = ast_path ($1->left.sval, $3); }
| just_identifiers_for_type '<' type_list RSH
{
$$ = ast_path ($1->left.sval, $3);
rust_push_back ('>');
}
;
type:
path_for_type
| '[' type ';' INTEGER ']'
{ $$ = ast_array_type ($2, $4); }
| '[' type ';' DECIMAL_INTEGER ']'
{ $$ = ast_array_type ($2, $4); }
| '&' '[' type ']'
{ $$ = ast_slice_type ($3); }
| '&' type
{ $$ = ast_reference_type ($2); }
| '*' KW_MUT type
{ $$ = ast_pointer_type ($3, 1); }
| '*' KW_CONST type
{ $$ = ast_pointer_type ($3, 0); }
| KW_FN '(' maybe_type_list ')' ARROW type
{ $$ = ast_function_type ($6, $3); }
| '(' maybe_type_list ')'
{ $$ = ast_tuple_type ($2); }
;
maybe_type_list:
/* %empty */
{ $$ = NULL; }
| type_list
{ $$ = $1; }
;
type_list:
type
{
VEC (rust_op_ptr) **result
= OBSTACK_ZALLOC (&work_obstack, VEC (rust_op_ptr) *);
make_cleanup (VEC_cleanup (rust_op_ptr), result);
VEC_safe_push (rust_op_ptr, *result, $1);
$$ = result;
}
| type_list ',' type
{
VEC_safe_push (rust_op_ptr, *$1, $3);
$$ = $1;
}
;
%%
/* A struct of this type is used to describe a token. */
struct token_info
{
const char *name;
int value;
enum exp_opcode opcode;
};
/* Identifier tokens. */
static const struct token_info identifier_tokens[] =
{
{ "as", KW_AS, OP_NULL },
{ "false", KW_FALSE, OP_NULL },
{ "if", 0, OP_NULL },
{ "mut", KW_MUT, OP_NULL },
{ "const", KW_CONST, OP_NULL },
{ "self", KW_SELF, OP_NULL },
{ "super", KW_SUPER, OP_NULL },
{ "true", KW_TRUE, OP_NULL },
{ "extern", KW_EXTERN, OP_NULL },
{ "fn", KW_FN, OP_NULL },
{ "sizeof", KW_SIZEOF, OP_NULL },
};
/* Operator tokens, sorted longest first. */
static const struct token_info operator_tokens[] =
{
{ ">>=", COMPOUND_ASSIGN, BINOP_RSH },
{ "<<=", COMPOUND_ASSIGN, BINOP_LSH },
{ "<<", LSH, OP_NULL },
{ ">>", RSH, OP_NULL },
{ "&&", ANDAND, OP_NULL },
{ "||", OROR, OP_NULL },
{ "==", EQEQ, OP_NULL },
{ "!=", NOTEQ, OP_NULL },
{ "<=", LTEQ, OP_NULL },
{ ">=", GTEQ, OP_NULL },
{ "+=", COMPOUND_ASSIGN, BINOP_ADD },
{ "-=", COMPOUND_ASSIGN, BINOP_SUB },
{ "*=", COMPOUND_ASSIGN, BINOP_MUL },
{ "/=", COMPOUND_ASSIGN, BINOP_DIV },
{ "%=", COMPOUND_ASSIGN, BINOP_REM },
{ "&=", COMPOUND_ASSIGN, BINOP_BITWISE_AND },
{ "|=", COMPOUND_ASSIGN, BINOP_BITWISE_IOR },
{ "^=", COMPOUND_ASSIGN, BINOP_BITWISE_XOR },
{ "::", COLONCOLON, OP_NULL },
{ "..", DOTDOT, OP_NULL },
{ "->", ARROW, OP_NULL }
};
/* Helper function to copy to the name obstack. */
static const char *
rust_copy_name (const char *name, int len)
{
return (const char *) obstack_copy0 (&work_obstack, name, len);
}
/* Helper function to make an stoken from a C string. */
static struct stoken
make_stoken (const char *p)
{
struct stoken result;
result.ptr = p;
result.length = strlen (result.ptr);
return result;
}
/* Helper function to concatenate three strings on the name
obstack. */
static struct stoken
rust_concat3 (const char *s1, const char *s2, const char *s3)
{
return make_stoken (obconcat (&work_obstack, s1, s2, s3, (char *) NULL));
}
/* Return an AST node referring to NAME, but relative to the crate's
name. */
static const struct rust_op *
crate_name (const struct rust_op *name)
{
std::string crate = rust_crate_for_block (expression_context_block);
struct stoken result;
gdb_assert (name->opcode == OP_VAR_VALUE);
if (crate.empty ())
error (_("Could not find crate for current location"));
result = make_stoken (obconcat (&work_obstack, "::", crate.c_str (), "::",
name->left.sval.ptr, (char *) NULL));
return ast_path (result, name->right.params);
}
/* Create an AST node referring to a "super::" qualified name. IDENT
is the base name and N_SUPERS is how many "super::"s were
provided. N_SUPERS can be zero. */
static const struct rust_op *
super_name (const struct rust_op *ident, unsigned int n_supers)
{
const char *scope = block_scope (expression_context_block);
int offset;
gdb_assert (ident->opcode == OP_VAR_VALUE);
if (scope[0] == '\0')
error (_("Couldn't find namespace scope for self::"));
if (n_supers > 0)
{
int i;
int len;
std::vector offsets;
unsigned int current_len;
current_len = cp_find_first_component (scope);
while (scope[current_len] != '\0')
{
offsets.push_back (current_len);
gdb_assert (scope[current_len] == ':');
/* The "::". */
current_len += 2;
current_len += cp_find_first_component (scope
+ current_len);
}
len = offsets.size ();
if (n_supers >= len)
error (_("Too many super:: uses from '%s'"), scope);
offset = offsets[len - n_supers];
}
else
offset = strlen (scope);
obstack_grow (&work_obstack, "::", 2);
obstack_grow (&work_obstack, scope, offset);
obstack_grow (&work_obstack, "::", 2);
obstack_grow0 (&work_obstack, ident->left.sval.ptr, ident->left.sval.length);
return ast_path (make_stoken ((const char *) obstack_finish (&work_obstack)),
ident->right.params);
}
/* A helper that updates innermost_block as appropriate. */
static void
update_innermost_block (struct block_symbol sym)
{
if (symbol_read_needs_frame (sym.symbol)
&& (innermost_block == NULL
|| contained_in (sym.block, innermost_block)))
innermost_block = sym.block;
}
/* A helper to look up a Rust type, or fail. This only works for
types defined by rust_language_arch_info. */
static struct type *
rust_type (const char *name)
{
struct type *type;
/* When unit testing, we don't bother checking the types, so avoid a
possibly-failing lookup here. */
if (unit_testing)
return NULL;
type = language_lookup_primitive_type (parse_language (pstate),
parse_gdbarch (pstate),
name);
if (type == NULL)
error (_("Could not find Rust type %s"), name);
return type;
}
/* Lex a hex number with at least MIN digits and at most MAX
digits. */
static uint32_t
lex_hex (int min, int max)
{
uint32_t result = 0;
int len = 0;
/* We only want to stop at MAX if we're lexing a byte escape. */
int check_max = min == max;
while ((check_max ? len <= max : 1)
&& ((lexptr[0] >= 'a' && lexptr[0] <= 'f')
|| (lexptr[0] >= 'A' && lexptr[0] <= 'F')
|| (lexptr[0] >= '0' && lexptr[0] <= '9')))
{
result *= 16;
if (lexptr[0] >= 'a' && lexptr[0] <= 'f')
result = result + 10 + lexptr[0] - 'a';
else if (lexptr[0] >= 'A' && lexptr[0] <= 'F')
result = result + 10 + lexptr[0] - 'A';
else
result = result + lexptr[0] - '0';
++lexptr;
++len;
}
if (len < min)
error (_("Not enough hex digits seen"));
if (len > max)
{
gdb_assert (min != max);
error (_("Overlong hex escape"));
}
return result;
}
/* Lex an escape. IS_BYTE is true if we're lexing a byte escape;
otherwise we're lexing a character escape. */
static uint32_t
lex_escape (int is_byte)
{
uint32_t result;
gdb_assert (lexptr[0] == '\\');
++lexptr;
switch (lexptr[0])
{
case 'x':
++lexptr;
result = lex_hex (2, 2);
break;
case 'u':
if (is_byte)
error (_("Unicode escape in byte literal"));
++lexptr;
if (lexptr[0] != '{')
error (_("Missing '{' in Unicode escape"));
++lexptr;
result = lex_hex (1, 6);
/* Could do range checks here. */
if (lexptr[0] != '}')
error (_("Missing '}' in Unicode escape"));
++lexptr;
break;
case 'n':
result = '\n';
++lexptr;
break;
case 'r':
result = '\r';
++lexptr;
break;
case 't':
result = '\t';
++lexptr;
break;
case '\\':
result = '\\';
++lexptr;
break;
case '0':
result = '\0';
++lexptr;
break;
case '\'':
result = '\'';
++lexptr;
break;
case '"':
result = '"';
++lexptr;
break;
default:
error (_("Invalid escape \\%c in literal"), lexptr[0]);
}
return result;
}
/* Lex a character constant. */
static int
lex_character (void)
{
int is_byte = 0;
uint32_t value;
if (lexptr[0] == 'b')
{
is_byte = 1;
++lexptr;
}
gdb_assert (lexptr[0] == '\'');
++lexptr;
/* This should handle UTF-8 here. */
if (lexptr[0] == '\\')
value = lex_escape (is_byte);
else
{
value = lexptr[0] & 0xff;
++lexptr;
}
if (lexptr[0] != '\'')
error (_("Unterminated character literal"));
++lexptr;
rustyylval.typed_val_int.val = value;
rustyylval.typed_val_int.type = rust_type (is_byte ? "u8" : "char");
return INTEGER;
}
/* Return the offset of the double quote if STR looks like the start
of a raw string, or 0 if STR does not start a raw string. */
static int
starts_raw_string (const char *str)
{
const char *save = str;
if (str[0] != 'r')
return 0;
++str;
while (str[0] == '#')
++str;
if (str[0] == '"')
return str - save;
return 0;
}
/* Return true if STR looks like the end of a raw string that had N
hashes at the start. */
static bool
ends_raw_string (const char *str, int n)
{
int i;
gdb_assert (str[0] == '"');
for (i = 0; i < n; ++i)
if (str[i + 1] != '#')
return false;
return true;
}
/* Lex a string constant. */
static int
lex_string (void)
{
int is_byte = lexptr[0] == 'b';
int raw_length;
int len_in_chars = 0;
if (is_byte)
++lexptr;
raw_length = starts_raw_string (lexptr);
lexptr += raw_length;
gdb_assert (lexptr[0] == '"');
++lexptr;
while (1)
{
uint32_t value;
if (raw_length > 0)
{
if (lexptr[0] == '"' && ends_raw_string (lexptr, raw_length - 1))
{
/* Exit with lexptr pointing after the final "#". */
lexptr += raw_length;
break;
}
else if (lexptr[0] == '\0')
error (_("Unexpected EOF in string"));
value = lexptr[0] & 0xff;
if (is_byte && value > 127)
error (_("Non-ASCII value in raw byte string"));
obstack_1grow (&work_obstack, value);
++lexptr;
}
else if (lexptr[0] == '"')
{
/* Make sure to skip the quote. */
++lexptr;
break;
}
else if (lexptr[0] == '\\')
{
value = lex_escape (is_byte);
if (is_byte)
obstack_1grow (&work_obstack, value);
else
convert_between_encodings ("UTF-32", "UTF-8", (gdb_byte *) &value,
sizeof (value), sizeof (value),
&work_obstack, translit_none);
}
else if (lexptr[0] == '\0')
error (_("Unexpected EOF in string"));
else
{
value = lexptr[0] & 0xff;
if (is_byte && value > 127)
error (_("Non-ASCII value in byte string"));
obstack_1grow (&work_obstack, value);
++lexptr;
}
}
rustyylval.sval.length = obstack_object_size (&work_obstack);
rustyylval.sval.ptr = (const char *) obstack_finish (&work_obstack);
return is_byte ? BYTESTRING : STRING;
}
/* Return true if STRING starts with whitespace followed by a digit. */
static bool
space_then_number (const char *string)
{
const char *p = string;
while (p[0] == ' ' || p[0] == '\t')
++p;
if (p == string)
return false;
return *p >= '0' && *p <= '9';
}
/* Return true if C can start an identifier. */
static bool
rust_identifier_start_p (char c)
{
return ((c >= 'a' && c <= 'z')
|| (c >= 'A' && c <= 'Z')
|| c == '_'
|| c == '$');
}
/* Lex an identifier. */
static int
lex_identifier (void)
{
const char *start = lexptr;
unsigned int length;
const struct token_info *token;
int i;
int is_gdb_var = lexptr[0] == '$';
gdb_assert (rust_identifier_start_p (lexptr[0]));
++lexptr;
/* For the time being this doesn't handle Unicode rules. Non-ASCII
identifiers are gated anyway. */
while ((lexptr[0] >= 'a' && lexptr[0] <= 'z')
|| (lexptr[0] >= 'A' && lexptr[0] <= 'Z')
|| lexptr[0] == '_'
|| (is_gdb_var && lexptr[0] == '$')
|| (lexptr[0] >= '0' && lexptr[0] <= '9'))
++lexptr;
length = lexptr - start;
token = NULL;
for (i = 0; i < ARRAY_SIZE (identifier_tokens); ++i)
{
if (length == strlen (identifier_tokens[i].name)
&& strncmp (identifier_tokens[i].name, start, length) == 0)
{
token = &identifier_tokens[i];
break;
}
}
if (token != NULL)
{
if (token->value == 0)
{
/* Leave the terminating token alone. */
lexptr = start;
return 0;
}
}
else if (token == NULL
&& (strncmp (start, "thread", length) == 0
|| strncmp (start, "task", length) == 0)
&& space_then_number (lexptr))
{
/* "task" or "thread" followed by a number terminates the
parse, per gdb rules. */
lexptr = start;
return 0;
}
if (token == NULL || (parse_completion && lexptr[0] == '\0'))
rustyylval.sval = make_stoken (rust_copy_name (start, length));
if (parse_completion && lexptr[0] == '\0')
{
/* Prevent rustyylex from returning two COMPLETE tokens. */
prev_lexptr = lexptr;
return COMPLETE;
}
if (token != NULL)
return token->value;
if (is_gdb_var)
return GDBVAR;
return IDENT;
}
/* Lex an operator. */
static int
lex_operator (void)
{
const struct token_info *token = NULL;
int i;
for (i = 0; i < ARRAY_SIZE (operator_tokens); ++i)
{
if (strncmp (operator_tokens[i].name, lexptr,
strlen (operator_tokens[i].name)) == 0)
{
lexptr += strlen (operator_tokens[i].name);
token = &operator_tokens[i];
break;
}
}
if (token != NULL)
{
rustyylval.opcode = token->opcode;
return token->value;
}
return *lexptr++;
}
/* Lex a number. */
static int
lex_number (void)
{
regmatch_t subexps[NUM_SUBEXPRESSIONS];
int match;
int is_integer = 0;
int could_be_decimal = 1;
int implicit_i32 = 0;
const char *type_name = NULL;
struct type *type;
int end_index;
int type_index = -1;
int i;
match = regexec (&number_regex, lexptr, ARRAY_SIZE (subexps), subexps, 0);
/* Failure means the regexp is broken. */
gdb_assert (match == 0);
if (subexps[INT_TEXT].rm_so != -1)
{
/* Integer part matched. */
is_integer = 1;
end_index = subexps[INT_TEXT].rm_eo;
if (subexps[INT_TYPE].rm_so == -1)
{
type_name = "i32";
implicit_i32 = 1;
}
else
{
type_index = INT_TYPE;
could_be_decimal = 0;
}
}
else if (subexps[FLOAT_TYPE1].rm_so != -1)
{
/* Found floating point type suffix. */
end_index = subexps[FLOAT_TYPE1].rm_so;
type_index = FLOAT_TYPE1;
}
else if (subexps[FLOAT_TYPE2].rm_so != -1)
{
/* Found floating point type suffix. */
end_index = subexps[FLOAT_TYPE2].rm_so;
type_index = FLOAT_TYPE2;
}
else
{
/* Any other floating point match. */
end_index = subexps[0].rm_eo;
type_name = "f64";
}
/* We need a special case if the final character is ".". In this
case we might need to parse an integer. For example, "23.f()" is
a request for a trait method call, not a syntax error involving
the floating point number "23.". */
gdb_assert (subexps[0].rm_eo > 0);
if (lexptr[subexps[0].rm_eo - 1] == '.')
{
const char *next = skip_spaces_const (&lexptr[subexps[0].rm_eo]);
if (rust_identifier_start_p (*next) || *next == '.')
{
--subexps[0].rm_eo;
is_integer = 1;
end_index = subexps[0].rm_eo;
type_name = "i32";
could_be_decimal = 1;
implicit_i32 = 1;
}
}
/* Compute the type name if we haven't already. */
std::string type_name_holder;
if (type_name == NULL)
{
gdb_assert (type_index != -1);
type_name_holder = std::string (lexptr + subexps[type_index].rm_so,
(subexps[type_index].rm_eo
- subexps[type_index].rm_so));
type_name = type_name_holder.c_str ();
}
/* Look up the type. */
type = rust_type (type_name);
/* Copy the text of the number and remove the "_"s. */
std::string number;
for (i = 0; i < end_index && lexptr[i]; ++i)
{
if (lexptr[i] == '_')
could_be_decimal = 0;
else
number.push_back (lexptr[i]);
}
/* Advance past the match. */
lexptr += subexps[0].rm_eo;
/* Parse the number. */
if (is_integer)
{
uint64_t value;
int radix = 10;
int offset = 0;
if (number[0] == '0')
{
if (number[1] == 'x')
radix = 16;
else if (number[1] == 'o')
radix = 8;
else if (number[1] == 'b')
radix = 2;
if (radix != 10)
{
offset = 2;
could_be_decimal = 0;
}
}
value = strtoul (number.c_str () + offset, NULL, radix);
if (implicit_i32 && value >= ((uint64_t) 1) << 31)
type = rust_type ("i64");
rustyylval.typed_val_int.val = value;
rustyylval.typed_val_int.type = type;
}
else
{
rustyylval.typed_val_float.dval = strtod (number.c_str (), NULL);
rustyylval.typed_val_float.type = type;
}
return is_integer ? (could_be_decimal ? DECIMAL_INTEGER : INTEGER) : FLOAT;
}
/* The lexer. */
static int
rustyylex (void)
{
/* Skip all leading whitespace. */
while (lexptr[0] == ' ' || lexptr[0] == '\t' || lexptr[0] == '\r'
|| lexptr[0] == '\n')
++lexptr;
/* If we hit EOF and we're completing, then return COMPLETE -- maybe
we're completing an empty string at the end of a field_expr.
But, we don't want to return two COMPLETE tokens in a row. */
if (lexptr[0] == '\0' && lexptr == prev_lexptr)
return 0;
prev_lexptr = lexptr;
if (lexptr[0] == '\0')
{
if (parse_completion)
{
rustyylval.sval = make_stoken ("");
return COMPLETE;
}
return 0;
}
if (lexptr[0] >= '0' && lexptr[0] <= '9')
return lex_number ();
else if (lexptr[0] == 'b' && lexptr[1] == '\'')
return lex_character ();
else if (lexptr[0] == 'b' && lexptr[1] == '"')
return lex_string ();
else if (lexptr[0] == 'b' && starts_raw_string (lexptr + 1))
return lex_string ();
else if (starts_raw_string (lexptr))
return lex_string ();
else if (rust_identifier_start_p (lexptr[0]))
return lex_identifier ();
else if (lexptr[0] == '"')
return lex_string ();
else if (lexptr[0] == '\'')
return lex_character ();
else if (lexptr[0] == '}' || lexptr[0] == ']')
{
/* Falls through to lex_operator. */
--paren_depth;
}
else if (lexptr[0] == '(' || lexptr[0] == '{')
{
/* Falls through to lex_operator. */
++paren_depth;
}
else if (lexptr[0] == ',' && comma_terminates && paren_depth == 0)
return 0;
return lex_operator ();
}
/* Push back a single character to be re-lexed. */
static void
rust_push_back (char c)
{
/* Can't be called before any lexing. */
gdb_assert (prev_lexptr != NULL);
--lexptr;
gdb_assert (*lexptr == c);
}
/* Make an arbitrary operation and fill in the fields. */
static const struct rust_op *
ast_operation (enum exp_opcode opcode, const struct rust_op *left,
const struct rust_op *right)
{
struct rust_op *result = OBSTACK_ZALLOC (&work_obstack, struct rust_op);
result->opcode = opcode;
result->left.op = left;
result->right.op = right;
return result;
}
/* Make a compound assignment operation. */
static const struct rust_op *
ast_compound_assignment (enum exp_opcode opcode, const struct rust_op *left,
const struct rust_op *right)
{
struct rust_op *result = OBSTACK_ZALLOC (&work_obstack, struct rust_op);
result->opcode = opcode;
result->compound_assignment = 1;
result->left.op = left;
result->right.op = right;
return result;
}
/* Make a typed integer literal operation. */
static const struct rust_op *
ast_literal (struct typed_val_int val)
{
struct rust_op *result = OBSTACK_ZALLOC (&work_obstack, struct rust_op);
result->opcode = OP_LONG;
result->left.typed_val_int = val;
return result;
}
/* Make a typed floating point literal operation. */
static const struct rust_op *
ast_dliteral (struct typed_val_float val)
{
struct rust_op *result = OBSTACK_ZALLOC (&work_obstack, struct rust_op);
result->opcode = OP_DOUBLE;
result->left.typed_val_float = val;
return result;
}
/* Make a unary operation. */
static const struct rust_op *
ast_unary (enum exp_opcode opcode, const struct rust_op *expr)
{
return ast_operation (opcode, expr, NULL);
}
/* Make a cast operation. */
static const struct rust_op *
ast_cast (const struct rust_op *expr, const struct rust_op *type)
{
struct rust_op *result = OBSTACK_ZALLOC (&work_obstack, struct rust_op);
result->opcode = UNOP_CAST;
result->left.op = expr;
result->right.op = type;
return result;
}
/* Make a call-like operation. This is nominally a function call, but
when lowering we may discover that it actually represents the
creation of a tuple struct. */
static const struct rust_op *
ast_call_ish (enum exp_opcode opcode, const struct rust_op *expr,
VEC (rust_op_ptr) **params)
{
struct rust_op *result = OBSTACK_ZALLOC (&work_obstack, struct rust_op);
result->opcode = opcode;
result->left.op = expr;
result->right.params = params;
return result;
}
/* Make a structure creation operation. */
static const struct rust_op *
ast_struct (const struct rust_op *name, VEC (set_field) **fields)
{
struct rust_op *result = OBSTACK_ZALLOC (&work_obstack, struct rust_op);
result->opcode = OP_AGGREGATE;
result->left.op = name;
result->right.field_inits = fields;
return result;
}
/* Make an identifier path. */
static const struct rust_op *
ast_path (struct stoken path, VEC (rust_op_ptr) **params)
{
struct rust_op *result = OBSTACK_ZALLOC (&work_obstack, struct rust_op);
result->opcode = OP_VAR_VALUE;
result->left.sval = path;
result->right.params = params;
return result;
}
/* Make a string constant operation. */
static const struct rust_op *
ast_string (struct stoken str)
{
struct rust_op *result = OBSTACK_ZALLOC (&work_obstack, struct rust_op);
result->opcode = OP_STRING;
result->left.sval = str;
return result;
}
/* Make a field expression. */
static const struct rust_op *
ast_structop (const struct rust_op *left, const char *name, int completing)
{
struct rust_op *result = OBSTACK_ZALLOC (&work_obstack, struct rust_op);
result->opcode = STRUCTOP_STRUCT;
result->completing = completing;
result->left.op = left;
result->right.sval = make_stoken (name);
return result;
}
/* Make an anonymous struct operation, like 'x.0'. */
static const struct rust_op *
ast_structop_anonymous (const struct rust_op *left,
struct typed_val_int number)
{
struct rust_op *result = OBSTACK_ZALLOC (&work_obstack, struct rust_op);
result->opcode = STRUCTOP_ANONYMOUS;
result->left.op = left;
result->right.typed_val_int = number;
return result;
}
/* Make a range operation. */
static const struct rust_op *
ast_range (const struct rust_op *lhs, const struct rust_op *rhs)
{
struct rust_op *result = OBSTACK_ZALLOC (&work_obstack, struct rust_op);
result->opcode = OP_RANGE;
result->left.op = lhs;
result->right.op = rhs;
return result;
}
/* A helper function to make a type-related AST node. */
static struct rust_op *
ast_basic_type (enum type_code typecode)
{
struct rust_op *result = OBSTACK_ZALLOC (&work_obstack, struct rust_op);
result->opcode = OP_TYPE;
result->typecode = typecode;
return result;
}
/* Create an AST node describing an array type. */
static const struct rust_op *
ast_array_type (const struct rust_op *lhs, struct typed_val_int val)
{
struct rust_op *result = ast_basic_type (TYPE_CODE_ARRAY);
result->left.op = lhs;
result->right.typed_val_int = val;
return result;
}
/* Create an AST node describing a reference type. */
static const struct rust_op *
ast_slice_type (const struct rust_op *type)
{
/* Use TYPE_CODE_COMPLEX just because it is handy. */
struct rust_op *result = ast_basic_type (TYPE_CODE_COMPLEX);
result->left.op = type;
return result;
}
/* Create an AST node describing a reference type. */
static const struct rust_op *
ast_reference_type (const struct rust_op *type)
{
struct rust_op *result = ast_basic_type (TYPE_CODE_REF);
result->left.op = type;
return result;
}
/* Create an AST node describing a pointer type. */
static const struct rust_op *
ast_pointer_type (const struct rust_op *type, int is_mut)
{
struct rust_op *result = ast_basic_type (TYPE_CODE_PTR);
result->left.op = type;
/* For the time being we ignore is_mut. */
return result;
}
/* Create an AST node describing a function type. */
static const struct rust_op *
ast_function_type (const struct rust_op *rtype, VEC (rust_op_ptr) **params)
{
struct rust_op *result = ast_basic_type (TYPE_CODE_FUNC);
result->left.op = rtype;
result->right.params = params;
return result;
}
/* Create an AST node describing a tuple type. */
static const struct rust_op *
ast_tuple_type (VEC (rust_op_ptr) **params)
{
struct rust_op *result = ast_basic_type (TYPE_CODE_STRUCT);
result->left.params = params;
return result;
}
/* A helper to appropriately munge NAME and BLOCK depending on the
presence of a leading "::". */
static void
munge_name_and_block (const char **name, const struct block **block)
{
/* If it is a global reference, skip the current block in favor of
the static block. */
if (strncmp (*name, "::", 2) == 0)
{
*name += 2;
*block = block_static_block (*block);
}
}
/* Like lookup_symbol, but handles Rust namespace conventions, and
doesn't require field_of_this_result. */
static struct block_symbol
rust_lookup_symbol (const char *name, const struct block *block,
const domain_enum domain)
{
struct block_symbol result;
munge_name_and_block (&name, &block);
result = lookup_symbol (name, block, domain, NULL);
if (result.symbol != NULL)
update_innermost_block (result);
return result;
}
/* Look up a type, following Rust namespace conventions. */
static struct type *
rust_lookup_type (const char *name, const struct block *block)
{
struct block_symbol result;
struct type *type;
munge_name_and_block (&name, &block);
result = lookup_symbol (name, block, STRUCT_DOMAIN, NULL);
if (result.symbol != NULL)
{
update_innermost_block (result);
return SYMBOL_TYPE (result.symbol);
}
type = lookup_typename (parse_language (pstate), parse_gdbarch (pstate),
name, NULL, 1);
if (type != NULL)
return type;
/* Last chance, try a built-in type. */
return language_lookup_primitive_type (parse_language (pstate),
parse_gdbarch (pstate),
name);
}
static struct type *convert_ast_to_type (struct parser_state *state,
const struct rust_op *operation);
static const char *convert_name (struct parser_state *state,
const struct rust_op *operation);
/* Convert a vector of rust_ops representing types to a vector of
types. */
static std::vector
convert_params_to_types (struct parser_state *state, VEC (rust_op_ptr) *params)
{
int i;
const struct rust_op *op;
std::vector result;
for (i = 0; VEC_iterate (rust_op_ptr, params, i, op); ++i)
result.push_back (convert_ast_to_type (state, op));
return result;
}
/* Convert a rust_op representing a type to a struct type *. */
static struct type *
convert_ast_to_type (struct parser_state *state,
const struct rust_op *operation)
{
struct type *type, *result = NULL;
if (operation->opcode == OP_VAR_VALUE)
{
const char *varname = convert_name (state, operation);
result = rust_lookup_type (varname, expression_context_block);
if (result == NULL)
error (_("No typed name '%s' in current context"), varname);
return result;
}
gdb_assert (operation->opcode == OP_TYPE);
switch (operation->typecode)
{
case TYPE_CODE_ARRAY:
type = convert_ast_to_type (state, operation->left.op);
if (operation->right.typed_val_int.val < 0)
error (_("Negative array length"));
result = lookup_array_range_type (type, 0,
operation->right.typed_val_int.val - 1);
break;
case TYPE_CODE_COMPLEX:
{
struct type *usize = rust_type ("usize");
type = convert_ast_to_type (state, operation->left.op);
result = rust_slice_type ("&[*gdb*]", type, usize);
}
break;
case TYPE_CODE_REF:
case TYPE_CODE_PTR:
/* For now we treat &x and *x identically. */
type = convert_ast_to_type (state, operation->left.op);
result = lookup_pointer_type (type);
break;
case TYPE_CODE_FUNC:
{
std::vector args
(convert_params_to_types (state, *operation->right.params));
struct type **argtypes = NULL;
type = convert_ast_to_type (state, operation->left.op);
if (!args.empty ())
argtypes = args.data ();
result
= lookup_function_type_with_arguments (type, args.size (),
argtypes);
result = lookup_pointer_type (result);
}
break;
case TYPE_CODE_STRUCT:
{
std::vector args
(convert_params_to_types (state, *operation->left.params));
int i;
struct type *type;
const char *name;
obstack_1grow (&work_obstack, '(');
for (i = 0; i < args.size (); ++i)
{
std::string type_name = type_to_string (args[i]);
if (i > 0)
obstack_1grow (&work_obstack, ',');
obstack_grow_str (&work_obstack, type_name.c_str ());
}
obstack_grow_str0 (&work_obstack, ")");
name = (const char *) obstack_finish (&work_obstack);
/* We don't allow creating new tuple types (yet), but we do
allow looking up existing tuple types. */
result = rust_lookup_type (name, expression_context_block);
if (result == NULL)
error (_("could not find tuple type '%s'"), name);
}
break;
default:
gdb_assert_not_reached ("unhandled opcode in convert_ast_to_type");
}
gdb_assert (result != NULL);
return result;
}
/* A helper function to turn a rust_op representing a name into a full
name. This applies generic arguments as needed. The returned name
is allocated on the work obstack. */
static const char *
convert_name (struct parser_state *state, const struct rust_op *operation)
{
int i;
gdb_assert (operation->opcode == OP_VAR_VALUE);
if (operation->right.params == NULL)
return operation->left.sval.ptr;
std::vector types
(convert_params_to_types (state, *operation->right.params));
obstack_grow_str (&work_obstack, operation->left.sval.ptr);
obstack_1grow (&work_obstack, '<');
for (i = 0; i < types.size (); ++i)
{
std::string type_name = type_to_string (types[i]);
if (i > 0)
obstack_1grow (&work_obstack, ',');
obstack_grow_str (&work_obstack, type_name.c_str ());
}
obstack_grow_str0 (&work_obstack, ">");
return (const char *) obstack_finish (&work_obstack);
}
static void convert_ast_to_expression (struct parser_state *state,
const struct rust_op *operation,
const struct rust_op *top);
/* A helper function that converts a vec of rust_ops to a gdb
expression. */
static void
convert_params_to_expression (struct parser_state *state,
VEC (rust_op_ptr) *params,
const struct rust_op *top)
{
int i;
rust_op_ptr elem;
for (i = 0; VEC_iterate (rust_op_ptr, params, i, elem); ++i)
convert_ast_to_expression (state, elem, top);
}
/* Lower a rust_op to a gdb expression. STATE is the parser state.
OPERATION is the operation to lower. TOP is a pointer to the
top-most operation; it is used to handle the special case where the
top-most expression is an identifier and can be optionally lowered
to OP_TYPE. */
static void
convert_ast_to_expression (struct parser_state *state,
const struct rust_op *operation,
const struct rust_op *top)
{
switch (operation->opcode)
{
case OP_LONG:
write_exp_elt_opcode (state, OP_LONG);
write_exp_elt_type (state, operation->left.typed_val_int.type);
write_exp_elt_longcst (state, operation->left.typed_val_int.val);
write_exp_elt_opcode (state, OP_LONG);
break;
case OP_DOUBLE:
write_exp_elt_opcode (state, OP_DOUBLE);
write_exp_elt_type (state, operation->left.typed_val_float.type);
write_exp_elt_dblcst (state, operation->left.typed_val_float.dval);
write_exp_elt_opcode (state, OP_DOUBLE);
break;
case STRUCTOP_STRUCT:
{
convert_ast_to_expression (state, operation->left.op, top);
if (operation->completing)
mark_struct_expression (state);
write_exp_elt_opcode (state, STRUCTOP_STRUCT);
write_exp_string (state, operation->right.sval);
write_exp_elt_opcode (state, STRUCTOP_STRUCT);
}
break;
case STRUCTOP_ANONYMOUS:
{
convert_ast_to_expression (state, operation->left.op, top);
write_exp_elt_opcode (state, STRUCTOP_ANONYMOUS);
write_exp_elt_longcst (state, operation->right.typed_val_int.val);
write_exp_elt_opcode (state, STRUCTOP_ANONYMOUS);
}
break;
case UNOP_PLUS:
case UNOP_NEG:
case UNOP_COMPLEMENT:
case UNOP_IND:
case UNOP_ADDR:
case UNOP_SIZEOF:
convert_ast_to_expression (state, operation->left.op, top);
write_exp_elt_opcode (state, operation->opcode);
break;
case BINOP_SUBSCRIPT:
case BINOP_MUL:
case BINOP_REPEAT:
case BINOP_DIV:
case BINOP_REM:
case BINOP_LESS:
case BINOP_GTR:
case BINOP_BITWISE_AND:
case BINOP_BITWISE_IOR:
case BINOP_BITWISE_XOR:
case BINOP_ADD:
case BINOP_SUB:
case BINOP_LOGICAL_OR:
case BINOP_LOGICAL_AND:
case BINOP_EQUAL:
case BINOP_NOTEQUAL:
case BINOP_LEQ:
case BINOP_GEQ:
case BINOP_LSH:
case BINOP_RSH:
case BINOP_ASSIGN:
case OP_RUST_ARRAY:
convert_ast_to_expression (state, operation->left.op, top);
convert_ast_to_expression (state, operation->right.op, top);
if (operation->compound_assignment)
{
write_exp_elt_opcode (state, BINOP_ASSIGN_MODIFY);
write_exp_elt_opcode (state, operation->opcode);
write_exp_elt_opcode (state, BINOP_ASSIGN_MODIFY);
}
else
write_exp_elt_opcode (state, operation->opcode);
if (operation->compound_assignment
|| operation->opcode == BINOP_ASSIGN)
{
struct type *type;
type = language_lookup_primitive_type (parse_language (state),
parse_gdbarch (state),
"()");
write_exp_elt_opcode (state, OP_LONG);
write_exp_elt_type (state, type);
write_exp_elt_longcst (state, 0);
write_exp_elt_opcode (state, OP_LONG);
write_exp_elt_opcode (state, BINOP_COMMA);
}
break;
case UNOP_CAST:
{
struct type *type = convert_ast_to_type (state, operation->right.op);
convert_ast_to_expression (state, operation->left.op, top);
write_exp_elt_opcode (state, UNOP_CAST);
write_exp_elt_type (state, type);
write_exp_elt_opcode (state, UNOP_CAST);
}
break;
case OP_FUNCALL:
{
if (operation->left.op->opcode == OP_VAR_VALUE)
{
struct type *type;
const char *varname = convert_name (state, operation->left.op);
type = rust_lookup_type (varname, expression_context_block);
if (type != NULL)
{
/* This is actually a tuple struct expression, not a
call expression. */
rust_op_ptr elem;
int i;
VEC (rust_op_ptr) *params = *operation->right.params;
if (TYPE_CODE (type) != TYPE_CODE_NAMESPACE)
{
if (!rust_tuple_struct_type_p (type))
error (_("Type %s is not a tuple struct"), varname);
for (i = 0;
VEC_iterate (rust_op_ptr, params, i, elem);
++i)
{
char *cell = get_print_cell ();
xsnprintf (cell, PRINT_CELL_SIZE, "__%d", i);
write_exp_elt_opcode (state, OP_NAME);
write_exp_string (state, make_stoken (cell));
write_exp_elt_opcode (state, OP_NAME);
convert_ast_to_expression (state, elem, top);
}
write_exp_elt_opcode (state, OP_AGGREGATE);
write_exp_elt_type (state, type);
write_exp_elt_longcst (state,
2 * VEC_length (rust_op_ptr,
params));
write_exp_elt_opcode (state, OP_AGGREGATE);
break;
}
}
}
convert_ast_to_expression (state, operation->left.op, top);
convert_params_to_expression (state, *operation->right.params, top);
write_exp_elt_opcode (state, OP_FUNCALL);
write_exp_elt_longcst (state, VEC_length (rust_op_ptr,
*operation->right.params));
write_exp_elt_longcst (state, OP_FUNCALL);
}
break;
case OP_ARRAY:
gdb_assert (operation->left.op == NULL);
convert_params_to_expression (state, *operation->right.params, top);
write_exp_elt_opcode (state, OP_ARRAY);
write_exp_elt_longcst (state, 0);
write_exp_elt_longcst (state, VEC_length (rust_op_ptr,
*operation->right.params) - 1);
write_exp_elt_longcst (state, OP_ARRAY);
break;
case OP_VAR_VALUE:
{
struct block_symbol sym;
const char *varname;
if (operation->left.sval.ptr[0] == '$')
{
write_dollar_variable (state, operation->left.sval);
break;
}
varname = convert_name (state, operation);
sym = rust_lookup_symbol (varname, expression_context_block,
VAR_DOMAIN);
if (sym.symbol != NULL)
{
write_exp_elt_opcode (state, OP_VAR_VALUE);
write_exp_elt_block (state, sym.block);
write_exp_elt_sym (state, sym.symbol);
write_exp_elt_opcode (state, OP_VAR_VALUE);
}
else
{
struct type *type;
type = rust_lookup_type (varname, expression_context_block);
if (type == NULL)
error (_("No symbol '%s' in current context"), varname);
if (TYPE_CODE (type) == TYPE_CODE_STRUCT
&& TYPE_NFIELDS (type) == 0)
{
/* A unit-like struct. */
write_exp_elt_opcode (state, OP_AGGREGATE);
write_exp_elt_type (state, type);
write_exp_elt_longcst (state, 0);
write_exp_elt_opcode (state, OP_AGGREGATE);
}
else if (operation == top)
{
write_exp_elt_opcode (state, OP_TYPE);
write_exp_elt_type (state, type);
write_exp_elt_opcode (state, OP_TYPE);
break;
}
}
}
break;
case OP_AGGREGATE:
{
int i;
int length;
struct set_field *init;
VEC (set_field) *fields = *operation->right.field_inits;
struct type *type;
const char *name;
length = 0;
for (i = 0; VEC_iterate (set_field, fields, i, init); ++i)
{
if (init->name.ptr != NULL)
{
write_exp_elt_opcode (state, OP_NAME);
write_exp_string (state, init->name);
write_exp_elt_opcode (state, OP_NAME);
++length;
}
convert_ast_to_expression (state, init->init, top);
++length;
if (init->name.ptr == NULL)
{
/* This is handled differently from Ada in our
evaluator. */
write_exp_elt_opcode (state, OP_OTHERS);
}
}
name = convert_name (state, operation->left.op);
type = rust_lookup_type (name, expression_context_block);
if (type == NULL)
error (_("Could not find type '%s'"), operation->left.sval.ptr);
if (TYPE_CODE (type) != TYPE_CODE_STRUCT
|| rust_tuple_type_p (type)
|| rust_tuple_struct_type_p (type))
error (_("Struct expression applied to non-struct type"));
write_exp_elt_opcode (state, OP_AGGREGATE);
write_exp_elt_type (state, type);
write_exp_elt_longcst (state, length);
write_exp_elt_opcode (state, OP_AGGREGATE);
}
break;
case OP_STRING:
{
write_exp_elt_opcode (state, OP_STRING);
write_exp_string (state, operation->left.sval);
write_exp_elt_opcode (state, OP_STRING);
}
break;
case OP_RANGE:
{
enum range_type kind = BOTH_BOUND_DEFAULT;
if (operation->left.op != NULL)
{
convert_ast_to_expression (state, operation->left.op, top);
kind = HIGH_BOUND_DEFAULT;
}
if (operation->right.op != NULL)
{
convert_ast_to_expression (state, operation->right.op, top);
if (kind == BOTH_BOUND_DEFAULT)
kind = LOW_BOUND_DEFAULT;
else
{
gdb_assert (kind == HIGH_BOUND_DEFAULT);
kind = NONE_BOUND_DEFAULT;
}
}
write_exp_elt_opcode (state, OP_RANGE);
write_exp_elt_longcst (state, kind);
write_exp_elt_opcode (state, OP_RANGE);
}
break;
default:
gdb_assert_not_reached ("unhandled opcode in convert_ast_to_expression");
}
}
/* The parser as exposed to gdb. */
int
rust_parse (struct parser_state *state)
{
int result;
work_obstack.clear ();
rust_ast = NULL;
pstate = state;
/* Note that parsing (within rustyyparse) freely installs cleanups
assuming they're run here (below). */
struct cleanup *cleanup = make_cleanup (null_cleanup, NULL);
result = rustyyparse ();
if (!result || (parse_completion && rust_ast != NULL))
{
const struct rust_op *ast = rust_ast;
rust_ast = NULL;
gdb_assert (ast != NULL);
convert_ast_to_expression (state, ast, ast);
}
do_cleanups (cleanup);
return result;
}
/* The parser error handler. */
void
rustyyerror (const char *msg)
{
const char *where = prev_lexptr ? prev_lexptr : lexptr;
error (_("%s in expression, near `%s'."), (msg ? msg : "Error"), where);
}
#if GDB_SELF_TEST
/* Initialize the lexer for testing. */
static void
rust_lex_test_init (const char *input)
{
prev_lexptr = NULL;
lexptr = input;
paren_depth = 0;
}
/* A test helper that lexes a string, expecting a single token. It
returns the lexer data for this token. */
static RUSTSTYPE
rust_lex_test_one (const char *input, int expected)
{
int token;
RUSTSTYPE result;
rust_lex_test_init (input);
token = rustyylex ();
SELF_CHECK (token == expected);
result = rustyylval;
if (token)
{
token = rustyylex ();
SELF_CHECK (token == 0);
}
return result;
}
/* Test that INPUT lexes as the integer VALUE. */
static void
rust_lex_int_test (const char *input, int value, int kind)
{
RUSTSTYPE result = rust_lex_test_one (input, kind);
SELF_CHECK (result.typed_val_int.val == value);
}
/* Test that INPUT throws an exception with text ERR. */
static void
rust_lex_exception_test (const char *input, const char *err)
{
TRY
{
/* The "kind" doesn't matter. */
rust_lex_test_one (input, DECIMAL_INTEGER);
SELF_CHECK (0);
}
CATCH (except, RETURN_MASK_ERROR)
{
SELF_CHECK (strcmp (except.message, err) == 0);
}
END_CATCH
}
/* Test that INPUT lexes as the identifier, string, or byte-string
VALUE. KIND holds the expected token kind. */
static void
rust_lex_stringish_test (const char *input, const char *value, int kind)
{
RUSTSTYPE result = rust_lex_test_one (input, kind);
SELF_CHECK (result.sval.length == strlen (value));
SELF_CHECK (strncmp (result.sval.ptr, value, result.sval.length) == 0);
}
/* Helper to test that a string parses as a given token sequence. */
static void
rust_lex_test_sequence (const char *input, int len, const int expected[])
{
int i;
lexptr = input;
paren_depth = 0;
for (i = 0; i < len; ++i)
{
int token = rustyylex ();
SELF_CHECK (token == expected[i]);
}
}
/* Tests for an integer-parsing corner case. */
static void
rust_lex_test_trailing_dot (void)
{
const int expected1[] = { DECIMAL_INTEGER, '.', IDENT, '(', ')', 0 };
const int expected2[] = { INTEGER, '.', IDENT, '(', ')', 0 };
const int expected3[] = { FLOAT, EQEQ, '(', ')', 0 };
const int expected4[] = { DECIMAL_INTEGER, DOTDOT, DECIMAL_INTEGER, 0 };
rust_lex_test_sequence ("23.g()", ARRAY_SIZE (expected1), expected1);
rust_lex_test_sequence ("23_0.g()", ARRAY_SIZE (expected2), expected2);
rust_lex_test_sequence ("23.==()", ARRAY_SIZE (expected3), expected3);
rust_lex_test_sequence ("23..25", ARRAY_SIZE (expected4), expected4);
}
/* Tests of completion. */
static void
rust_lex_test_completion (void)
{
const int expected[] = { IDENT, '.', COMPLETE, 0 };
parse_completion = 1;
rust_lex_test_sequence ("something.wha", ARRAY_SIZE (expected), expected);
rust_lex_test_sequence ("something.", ARRAY_SIZE (expected), expected);
parse_completion = 0;
}
/* Test pushback. */
static void
rust_lex_test_push_back (void)
{
int token;
rust_lex_test_init (">>=");
token = rustyylex ();
SELF_CHECK (token == COMPOUND_ASSIGN);
SELF_CHECK (rustyylval.opcode == BINOP_RSH);
rust_push_back ('=');
token = rustyylex ();
SELF_CHECK (token == '=');
token = rustyylex ();
SELF_CHECK (token == 0);
}
/* Unit test the lexer. */
static void
rust_lex_tests (void)
{
int i;
work_obstack.clear ();
unit_testing = 1;
rust_lex_test_one ("", 0);
rust_lex_test_one (" \t \n \r ", 0);
rust_lex_test_one ("thread 23", 0);
rust_lex_test_one ("task 23", 0);
rust_lex_test_one ("th 104", 0);
rust_lex_test_one ("ta 97", 0);
rust_lex_int_test ("'z'", 'z', INTEGER);
rust_lex_int_test ("'\\xff'", 0xff, INTEGER);
rust_lex_int_test ("'\\u{1016f}'", 0x1016f, INTEGER);
rust_lex_int_test ("b'z'", 'z', INTEGER);
rust_lex_int_test ("b'\\xfe'", 0xfe, INTEGER);
rust_lex_int_test ("b'\\xFE'", 0xfe, INTEGER);
rust_lex_int_test ("b'\\xfE'", 0xfe, INTEGER);
/* Test all escapes in both modes. */
rust_lex_int_test ("'\\n'", '\n', INTEGER);
rust_lex_int_test ("'\\r'", '\r', INTEGER);
rust_lex_int_test ("'\\t'", '\t', INTEGER);
rust_lex_int_test ("'\\\\'", '\\', INTEGER);
rust_lex_int_test ("'\\0'", '\0', INTEGER);
rust_lex_int_test ("'\\''", '\'', INTEGER);
rust_lex_int_test ("'\\\"'", '"', INTEGER);
rust_lex_int_test ("b'\\n'", '\n', INTEGER);
rust_lex_int_test ("b'\\r'", '\r', INTEGER);
rust_lex_int_test ("b'\\t'", '\t', INTEGER);
rust_lex_int_test ("b'\\\\'", '\\', INTEGER);
rust_lex_int_test ("b'\\0'", '\0', INTEGER);
rust_lex_int_test ("b'\\''", '\'', INTEGER);
rust_lex_int_test ("b'\\\"'", '"', INTEGER);
rust_lex_exception_test ("'z", "Unterminated character literal");
rust_lex_exception_test ("b'\\x0'", "Not enough hex digits seen");
rust_lex_exception_test ("b'\\u{0}'", "Unicode escape in byte literal");
rust_lex_exception_test ("'\\x0'", "Not enough hex digits seen");
rust_lex_exception_test ("'\\u0'", "Missing '{' in Unicode escape");
rust_lex_exception_test ("'\\u{0", "Missing '}' in Unicode escape");
rust_lex_exception_test ("'\\u{0000007}", "Overlong hex escape");
rust_lex_exception_test ("'\\u{}", "Not enough hex digits seen");
rust_lex_exception_test ("'\\Q'", "Invalid escape \\Q in literal");
rust_lex_exception_test ("b'\\Q'", "Invalid escape \\Q in literal");
rust_lex_int_test ("23", 23, DECIMAL_INTEGER);
rust_lex_int_test ("2_344__29", 234429, INTEGER);
rust_lex_int_test ("0x1f", 0x1f, INTEGER);
rust_lex_int_test ("23usize", 23, INTEGER);
rust_lex_int_test ("23i32", 23, INTEGER);
rust_lex_int_test ("0x1_f", 0x1f, INTEGER);
rust_lex_int_test ("0b1_101011__", 0x6b, INTEGER);
rust_lex_int_test ("0o001177i64", 639, INTEGER);
rust_lex_test_trailing_dot ();
rust_lex_test_one ("23.", FLOAT);
rust_lex_test_one ("23.99f32", FLOAT);
rust_lex_test_one ("23e7", FLOAT);
rust_lex_test_one ("23E-7", FLOAT);
rust_lex_test_one ("23e+7", FLOAT);
rust_lex_test_one ("23.99e+7f64", FLOAT);
rust_lex_test_one ("23.82f32", FLOAT);
rust_lex_stringish_test ("hibob", "hibob", IDENT);
rust_lex_stringish_test ("hibob__93", "hibob__93", IDENT);
rust_lex_stringish_test ("thread", "thread", IDENT);
rust_lex_stringish_test ("\"string\"", "string", STRING);
rust_lex_stringish_test ("\"str\\ting\"", "str\ting", STRING);
rust_lex_stringish_test ("\"str\\\"ing\"", "str\"ing", STRING);
rust_lex_stringish_test ("r\"str\\ing\"", "str\\ing", STRING);
rust_lex_stringish_test ("r#\"str\\ting\"#", "str\\ting", STRING);
rust_lex_stringish_test ("r###\"str\\\"ing\"###", "str\\\"ing", STRING);
rust_lex_stringish_test ("b\"string\"", "string", BYTESTRING);
rust_lex_stringish_test ("b\"\x73tring\"", "string", BYTESTRING);
rust_lex_stringish_test ("b\"str\\\"ing\"", "str\"ing", BYTESTRING);
rust_lex_stringish_test ("br####\"\\x73tring\"####", "\\x73tring",
BYTESTRING);
for (i = 0; i < ARRAY_SIZE (identifier_tokens); ++i)
rust_lex_test_one (identifier_tokens[i].name, identifier_tokens[i].value);
for (i = 0; i < ARRAY_SIZE (operator_tokens); ++i)
rust_lex_test_one (operator_tokens[i].name, operator_tokens[i].value);
rust_lex_test_completion ();
rust_lex_test_push_back ();
unit_testing = 0;
}
#endif /* GDB_SELF_TEST */
void
_initialize_rust_exp (void)
{
int code = regcomp (&number_regex, number_regex_text, REG_EXTENDED);
/* If the regular expression was incorrect, it was a programming
error. */
gdb_assert (code == 0);
#if GDB_SELF_TEST
register_self_test (rust_lex_tests);
#endif
}