[patch] Re: GFortran testsuite problems with "dg-do compile"

[Tobias Schlüter]

Tobias Burnus <burnus@net-b.de> wrote on Tue, 19 Dec 2006:
>>> link:  char_result_11.f90
>>> assemble: same_name_1.f90
>>>
>>> I converted the first one to a "run" and the second one to a "link" to
>>> increase testing.
>>
>> I'm not sure I understand the reason for doing that -- if the intent
>> is to not have "link" or "assemble" in the gfortran testsuite, why did
>> you change the second one to "link"?  And, if that isn't the intent,
>> what's the purpose of changing these at all, other than that they're
>> there? (Even if that is the intent: Why is it useful?)
> Well, for the first one: To make sure that the conversion actually works;
>
> for the second one: To make sure that the two "u" don't clash, which is
> detected at link time
> 000000000040067e T __m__u
> 0000000000400678 T __n__u
>
> Especially for the first one, I'm not sure whether we could not manage
> to break it.
> The second one is fairly unlikely, but doing not only compiling but also
> linking does not cost much time.

I'm not sure I understand why char_result_11.f90 uses dg-do link, but  
same_name_1.f90 uses dg-do assemble, because the original error was an  
error given by the assembler -- instead of __n__u and __m__u, both  
symbols were given the same name, which the assembler rightfully  
comlained about.  The linker wouldn't have complainedd if both symbols  
were in different source files, as the assembly symbols have file  
scope, so linking would mean no additional testing for this bug.

- Tobi

----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.