------- Comment #6 from rguenther at suse dot de 2009-09-10 08:42 ------- Subject: Re: [4.5 Regression] ICE: expand_expr_real_1, at expr.c:8416
On Wed, 9 Sep 2009, matz at gcc dot gnu dot org wrote: > ------- Comment #5 from matz at gcc dot gnu dot org 2009-09-09 15:46 ------- > Oh, maybe an explanation: what is expected is that such decls (externs) either > have no context set (global) or have the current function as context. For > inlining (cloning/versioning all the same) this needs to be done by remapping > the decls. No idea why this isn't called on the _switch_mode_errorstr > in the inliner. As this works with normal functions I assume that somehow > the free_lang_data work changed internal structures in a way that this decl > isn't visible to the inliner anymore (in a way). It changes most DECL_CONTEXTs to NULL anyway. Cloning is running way before free_lang_data, so if it is working without fld then it really must be a fld issue... Richard. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41275
