Re: [Patch] libffi testsuite: Use correct type specifiers.

[David Daney <ddaney at avtrex dot com>]

Andrew Haley wrote:
> David Daney writes:
>  > As noted in my previous message:
>  > http://gcc.gnu.org/ml/java-patches/2007-q3/msg00126.html
>  > 
>  > The libffi testsuite has many places where the wrong type specifier is 
>  > used.  This patch corrects all the erroneous specifiers that I could 
>  > find.  Also there was a definition in ffitest.h for ffi_type_mylong that 
>  > exactly mirrored a standard type definition found in ffi.h, so I removed 
>  > it and used the correct standard specifier instead.
>  > 
>  > Tested on x86_64-pc-linux-gnu (both native and -m32 ABIs) and 
>  > mips64-linux with no failures.
>  > 
>  > OK to commit?
>
> Some of these don't look right.  For example, cls_align_sint32.c is
> supposed to check structure alignment of sint32, but you changed it to
> int.
No.  Although its name would imply otherwise, that test checks the 
alignment of *int* fields within structures.  I didn't change that 
fact.  All I did was to make the libffi type specifiers accurately 
reflect the types of the parameters and return values.

>   It would make more sense to change "signed int" to sint32_t,
> wouldn't it?  Then the test would actually be doing what it says it's
> supposed to do.
>   
My understanding is that valid test cases should never be changed.  If 
one wants to test for something different, a new test case should be 
written.  The intent of my patch was to fix internal inconsistencies in 
the test cases with out changing what they were testing for.  I agree 
that it would make sense to add tests that check passing of fixed width 
integers, but that is not what this patch does.

David Daney