Re: [Patch, Fortran, committed] PR 78592: [7 Regression] ICE in gfc_find_specific_dtio_proc, at fortran/interface.c:4939

[Janus Weil <janus at gcc dot gnu dot org>]

2016-12-18 11:18 GMT+01:00 Mikael Morin <morin-mikael@orange.fr>:
> Le 17/12/2016 à 22:49, Janus Weil a écrit :
>>
>> Hi Mikael,
>>
>>>> I have just committed a completely obvious patch for this PR. All it
>>>> does is rearrange some expressions to avoid an ICE (see attachment):
>>>>
>>> I have made a late review of it, and I think it’s not as innocent as it
>>> seems.
>>> With it, if the first element’s formal is not properly set, the rest of
>>> the
>>> generic linked list is ignored.
>>>
>>> Here is a variant of the testcase committed.
>>> It shows no error if the module procedure line is commented, and two
>>> errors
>>> if it’s uncommented, one error saying that the write of z2 should use
>>> DTIO.
>>> The latter error should not appear.
>>
>>
>> thanks for the comment, and sorry that I didn't notice this problem.
>>
>> The attached patch should fix it. What do you think about it?
>>
> Yes, it looks good.

Ok, will commit this to trunk together with your test case.


>> Btw, with trunk r243776 I get an ICE on your test case, when the
>> module procedure is commented out. You don't see this?
>>
> My trunk is from the 8 december (r243465), and I don’t see it.
> So it should be a quite recent regression.

That confirms my suspicion that it was introduced by r243609.

Thanks again for finding both of these problems ;)

Cheers,
Janus