https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89850
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org,
| |mpolacek at gcc dot gnu.org
--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Guess that is because
/* Compare the hashes. */
if (h0 == h1
&& operand_equal_p (thenb, elseb, OEP_LEXICOGRAPHIC)
/* Don't warn if any of the branches or their subexpressions comes
from a macro. */
&& !walk_tree_without_duplicates (&thenb, expr_from_macro_expansion_r,
NULL)
&& !walk_tree_without_duplicates (&elseb, expr_from_macro_expansion_r,
NULL))
warning_at (EXPR_LOCATION (expr), OPT_Wduplicated_branches,
"this condition has identical branches");
calls operand_equal_p with OEP_LEXICOGRAPHIC and that doesn't handle SAVE_EXPRs
occuring more than once and nested inside of each other very well.
case SAVE_EXPR:
if (flags & OEP_LEXICOGRAPHIC)
return OP_SAME (0);
return 0;
There is also a problem already before, when computing the hash, as
inchash::add_expr doesn't special-case SAVE_EXPR at all.
Now, to fix this properly we'd need to create some hash_map from the SAVE_EXPRs
to their hash values and merge in that hash value each time we see it, instead
of computing it again. And similarly for operand_equal_p, noting we've
compared this SAVE_EXPR with that SAVE_EXPR successfully already once with
OEP_LEXICOGRAPHIC and so that comparison pair is ok even in all following cases
(at least with the same flags). But unfortunately there is no
inchash::add_expr or operand_equal_p context where we could stick such a
hash_map, unless it would be a global variable and we'd say required that if
OEP_LEXICOGRAPHIC is used, then the caller must set the hash_map global var and
release it at the end or something similar.
