Much fallout from PR 27668

Volker Reichelt
Mon Aug 7 19:46:00 GMT 2006


the patch for PR c++/27668 - using error_mark_node instead of
void_type_node to represent invalid template parameters - has caused
quite some fallout:

PR 28637, 28638, 28639, 28640, 28641, and 28594

(There's a patch available for the last regression.)

The regressions are all of the "ice-on-invalid-code, error-recovey" kind,
(as the original bug) but they are still regressions. And maybe there's
even more fallout. Nevertheless, IMHO the patch is still in the right

I just wanted to raise the question whether the change is appropriate
for stage 3 of mainline or even the 4.1 branch. Maybe we should back
it out from the branch until the issues have been resolved on mainline.

Any thoughts?

Btw. I added the fallout to the "blocks" field of PR 27668 for


More information about the Gcc-patches mailing list