Re: GCC compat testsuite

[Janis Johnson <janis187 at us dot ibm dot com>]

On Mon, Jan 05, 2004 at 06:57:19PM +0100, Eric Botcazou wrote:
> Hello Janis,
> 
> For various reasons it might be useful to run the compat testsuite against a 
> non-GCC compiler (for example, the vendor compiler might help to correctly 
> implement an awkward ABI like the SPARC64's).  Now some testcases use 
> GCC-specific idioms, so they don't compile with non-GCC compilers, although 
> they contain individual tests that would.
> 
> I've attached a logfile for a GCC vs Sun Studio 8 CC run on SPARC64:
> - _Complex means the keyword is used with non-fp types,
> - packed means the 'packed' attribute is used,
> - __INT_MAX__ means '__INT_MAX__' is tested,
> - array[0] means an array is declared with a null size,
> - ?? means the cause is unknown (most likely a deviation from the ABI).
> 
> What about clearly separating the testcases that use only Standard C99 
> features from the GCC-specific ones? The latter could be prefixed by "gcc_".

That sounds like a good idea.  I made the tests big to reduce the time
to compile them, but if there are good reasons then the functionality
can be split into multiple tests.

> I'm ok to do the work

Cool.  Feel free to contact me privately to discuss changes first, if
you'd like.

> (btw, is it really necessary to test zero-sized arrays?).

Do zero-sized arrays have ABI implications?  I think that's why they're
tested.  They can be in separate tests, though.

> I've also attached a tiny patch that fixes the FAIL marked as 'abort()'.
> 
> 
> 2004-01-05  Eric Botcazou  <ebotcazou@libertysurf.fr>
> 
> 	* gcc.dg/compat/sdata-section.h: Declare 'abort'.

This is OK, thanks.

Janis