------- Comment #1 from jamborm at gcc dot gnu dot org 2009-09-07 16:28 -------
We have discussed this in the mailing list thread that eventually lead
to http://gcc.gnu.org/ml/gcc-patches/2009-09/msg00374.html
The bottom line is that we two options:
1) Do not set DECL_VALUE_EXPR to callee copied parameters in
gimplify_parameters() and represent the convention explicitely. I
have tested this preliminary by running the c testsuite (no
bootstrap or some such thing) and so far found no ICEs or
miscompilations.
The drawback is that this approach will most probably mess up debug
info and the link in between the original and actual parameter will
get lost. The benefit is that we would uphold the stated (but
currently violated) rule that declarations with the DECL_VALUE_EXPR
do not occur in the function body.
2) The second approach is to teach gimplifier not to do the
substitutions which can be bogus. Once the function body has been
gimplifed, we might set some flag in the gimplification context or,
alternatively, I'm about to try to simply disable the substitutions
in gimplify_var_or_parm_decl() for PARM_DECLs.
This would mean we would need to drop the rule not allowing such
decls in the function body. I don't know whether it needs to be
replaced by some weaker restriction.
--
jamborm at gcc dot gnu dot org changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jamborm at gcc dot gnu dot
| |org
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41250
