[C++ Patch] PR 37087 (take 2)

[Manuel López-Ibáñez]

2008/8/12 Paolo Carlini <paolo.carlini@oracle.com>:
> Hi Manuel,
>>
>> Would you mind using in testcases as much as the error output as
>> possible to match? Otherwise we may issue a wrong message and not
>> notice the regression. Moreover, automatic changes are easier if you
>> can grep for the error messages.
>>
>> Also, in C++ testcases, you MUST use "error:", "warning:" or "note:"
>> in the match because the g++ testsuite cannot distinguish those.
>> That is PR 25241.
>>
>
> Ok, no problem, I didn't know about PR 25241 and well, I suspect a few
> others contributors are "guilty" of the same crime... ;)
>

In fact, the g++ testsuite has never been able to tell the difference
between warning, errors and notes. That combined with the fact that
the testsuite (none of them) cannot detect duplicated error/warning
messages (PR30612)... well, you can imagine that anything may make the
test pass.

The reason I did not fix PR 25241 for the C++ testsuite when I fixed
it for the C one is that there are far too many tests broken already!
That is, the test is testing for dg-warning but we actually give an
error and similar situations. One would have to go through every
failing test, and check whether it is a bug or the dg-* is incorrect
or whatever is going on and update the test or file a PR. If you know
someone willing to do this, directly him/her to me. It is very easy
but kind of tedious.

If GCC diagnostics are going to improve we have to be more aware of
unintended regressions. So please, spread the word and take a look at
other peoples patches.

Thanks,

Manuel.