[C++ Patch] PR 31446

[Paolo Carlini]

Hi,

a bit of story to understand this simple patch: the fix for PR 27668 led
unfortunately to a few new ice-on-invalid (28736, etc). Then Lee took
care of the latter by storing the error_mark_node corresponding to an
invalid template parameter in the TREE_VALUE field of the given
TREE_LIST element. The latter changes also included a robustification to
current_template_args. Unfortunately, however, the latter are at root of
the present problem: when error_mark_node is found in a TREE_VALUE,
instead of leaving the list element completely alone, it's changed by
saving error_mark_node in the list element itself, instead of its
TREE_VALUE, against the basic idea of the fix for 28736 & co. Adjusting
this small glitch solves the present problem.

Tested x86_64-linux. Of for mainline?

Paolo.

////////////////////
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: CL_31446
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20070926/18d2d352/attachment.ksh>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: patch_31446
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20070926/18d2d352/attachment-0001.ksh>