Re: Bring C++ ICE handling in line with other front ends

[Zack Weinberg <zack at wolery dot cumb dot org>]

On Thu, Jul 20, 2000 at 03:04:55PM -0700, Mark Mitchell wrote:
> 
> Zack --
> 
> I agree that the C++ front-end should handle errors similarly to the
> other front-ends.  I don't entirely agree with you about who should be
> changing.
> 
> Turning signals into ICEs, or doing something similar, is good.  Users
> don't care with the compiler crashed because of a SEGV or because of
> an abort; they just need to know that the compiler crashed.  Giving
> file and line information is good; it gives users a hint what
> constructs likely caused the problem, helping them to work around
> bugs. 

That's fair.  What would you think of pulling that logic up to
toplev.c so everyone got it?  I'm after consistency here.

> I don't feel strongly about turning ICEs into fatal errors.  We should
> have better error recovery, and you're right that this stuff can mask
> bugs.  That portion of the patch is approved, i.e., everything but the
> signal-handler stuff.
> 
> +      tree_check_failed (__t, VAR_DECL, __FILE__, __LINE__, __FUNCTION__); \
>      __t; })
> 
> Leave this on two lines so it's tidy, in both places.
> 
> Make sure to change the C++ testsuite so that the new crashes are
> XFAILs.

Okay.  I'll have to bootstrap again, then I'll post a revised patch.

zw