https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116219
--- Comment #16 from Richard Biener <rguenth at gcc dot gnu.org> --- (In reply to Jakub Jelinek from comment #14) > DECL_NAMELESS VAR_DECL names don't leak into debug info. The > -fcompare-debug failures > I believe are solely because these fancy names appear in the > -fdump-final-insns= > dumps (e.g. inside of MEM_EXPRs or REG_EXPRs). > And, they wouldn't appear there if it was just the VAR_DECLs created by SRA > with DECL_NAMELESS set, we have that > /* For -fcompare-debug don't dump DECL_NAMELESS names at all, > -g might have created more fancy names and their indexes > could get out of sync. Usually those should be DECL_IGNORED_P > too, SRA can create even non-DECL_IGNORED_P DECL_NAMELESS fancy > names, let's hope those never get out of sync after doing the > dump_fancy_name sanitization. */ > else if ((flags & TDF_COMPARE_DEBUG) > && DECL_NAMELESS (node) > && DECL_IGNORED_P (node)) > name = NULL_TREE; > /* For DECL_NAMELESS names look for embedded uids in the > names and sanitize them for TDF_NOUID. */ > else if ((flags & TDF_NOUID) && DECL_NAMELESS (node)) > dump_fancy_name (pp, name); > else > pp_tree_identifier (pp, name); > where dump_fancy_name will try to look for the Dnnnnn$ inside of it, so > offset$D94316$_M_impl$D93629$_M_start_588 > would be either not dumped as just _588 or > offset$Dxxxx$_M_impl$Dxxxx$_M_start_588 Note create_tmp_var also sets DECL_IGNORED_P. So the D12345 to Dxxxx is only done when there's a $ after the number? Why not also do this when there's a _ as clean_symbol_name does?
