[patch, fortran] PR18923 segfault after subroutine name confusion

[Jerry DeLisle]

Bernhard Fischer wrote:
> On Tue, May 22, 2007 at 08:31:28PM -0700, Jerry DeLisle wrote:
>> Tobias Burnus wrote:
>>> Jerry DeLisle wrote:
>>>> I would like to add this one additional change.  We need to give it up
>>>> at some point and one of these test cases triggers this.  It seems to
>>>> make sense to me.  Let me know if this is OK and I will include it. 
>>>> Regression tested of course and it obviously passes.
>>>> Index: symbol.c
>>>> -     gfc_internal_error ("gfc_get_default_type(): Bad symbol");
>>>> +     gfc_fatal_error ("Symbol name confusion at %C");
>>> Hmm, I like that it shows the position, but it only shows "Fatal Error:"
>>> instead of "Internal Compiler Error:". Internal errors should never
>>> happen (they are by definition a bug in the compiler) and should be
>>> reported.
>>>
>>> "Fatal Error:" on the other hand sound more like user errors which
>>> should not be reported.
>>>
>>> I think invalid user code should and are rejected earlier, namely in
>>> gfc_match ("Invalid character in name at %C"). Therefore, I would prefer
>>> an gfc_internal_error. Couldn't you use something like:
>>>
>>>     gfc_internal_error ("gfc_get_default_type(): Bad symbol at %C");
>>>
>>> this gives more debugging information but still implies the ICE. If I
>>> read error.c correctly, this should work.
>>>
>>> Tobias
>>>
>> This is one of those invalids where we ought to quit after the syntax error 
>> is found.  Instead we move on into resolve and get additional nonsense 
>> messages before finally hitting the internal error.  I have an upcoming 
>> patch, still being developed and tested to attempt to intercept these and 
>> bail out earlier.
> 
> Would calling gfc_check_error() there be sufficient to bail right at
> that spot?
>> I will commit the segfault fixes, then work on this new patch to catch the 
>> syntax and quit.  After that I will get the test cases in place.
> 
I am not sure, I will see what it does.  I still think stopping earlier would be 
better if possible.