[Bug c++/20201] requesting -Wfatal-errors=n

["manu at gcc dot gnu dot org" <gcc-bugzilla at gcc dot gnu dot org>]

------- Comment #6 from manu at gcc dot gnu dot org  2007-02-16 15:43 -------
(In reply to comment #5)
> I agree with Manuel. One error should be one error, regardless of the number of
> lines it takes to print it.
> 
> Two errors should be two errors, etc etc etc.
> 
> Seems like a pretty simple rule to me.
> 
> I find myself wishing for this rule more and more as metaprogramming becomes
> more and more established in the C++ community.
> 

I think -Wfatal-errors=n is the wrong fix to a broader problem. I have read the
thread started by Jim Wilson and I concluded two things:

1) People are annoyed by GCC spilling lots of error messages for a single
error. So they want to use -Wfatal-errors.
2) Sometimes a single error message does not contain enough information to
identify the source of the problem (see PR 15766).

I think both issues should be reported as bugs always.[*] I have seen a lot of
messages in forums and blogs complaining that GCC error messages are confusing
or redundant or simply too many errors for a single issue. However, there are
less than 20 PRs open for cases like these. I hope it is not because the PRs
have been being closed systematically since the error messages are *true*. The
messages can be completely *true* and conforming to the standards but that
doesn't mean that they are *useful* (see PR 29062). Perhaps it sounds strange
but a compiler can have usability issues. And GCC haves them.

[*] Also, they are normally the kind of low-hanging fruit that could attract
occassional contributors that don't have much free time to hack on a big
project but are willing to dedicate two or three hours of a raining weekend to
fix a PR like that. That is, like me ;-)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20201