RE: GDB/Fortran: Support for Assumed Rank Zero.

[Andrew Burgess via Gdb-patches <gdb-patches@sourceware.org>]

"Potharla, Rupesh via Gdb-patches" <gdb-patches@sourceware.org> writes:

> [AMD Official Use Only]
>
>
>>On Mon, 18 Apr 2022 08:33:28 -0700
>>Kevin Buettner via Gdb-patches <gdb-patches@sourceware.org> wrote:
>>
>>> I'm also (still) wondering how to run the testsuite to use flang
>>> instead of gfortran since it'd be nice to test to see if the
>>> setup_xfail actually works.  I spent just enough time looking at it
>>> last week to convince myself that GDB's testing infrastructure
>>> currently lacks the mechanisms to do this.  However, I could be wrong,
>>> so if someone knows how it's done, I'd like to learn about it...
>>
>>It seems that the answer is to run the test as follows:
>>
>>make check TESTS="gdb.fortran/assumedrank.exp"
>>RUNTESTFLAGS="F90_FOR_TARGET=flang"
>>
>>(I thought that I had tried this last week.)
>>
>>When I run the above command on my Fedora 35 machine, I see:
>>
>># of expected passes            500
>># of unexpected failures        2503
>># of unresolved testcases       500
>>
> But the command should only run assumedrank.exp testcase, looks like it ran all the tests under gdb.fortran.
>
> gdb summary after running assumedrank.exp with flang:
>
>                 === gdb Summary ===
>
> # of expected passes            25
> # of expected failures          1
>
>>So it seems that this test doesn't work very well with flang at the moment.
>>
>>That being the case, while I'm not against adding setup_xfail for clang, it
>>doesn't seem all that useful due to the dismal results when testing against
>>flang.
>>
>  
>
>         if {$test_count == 1 && [test_compiler_info {clang-*}]} {
>            setup_xfail "*-*-*"
>            fail "compiler does not support rank 0"
>            continue
>         }

It is possible to call 'xfail' directly, so you could just do:

  xfail "compiler does not support rank 0"

But I don't think that's correct really, xfail should be used when a
specific test is actually being run, so we'd expect to do something
like:

  xfail $test_name

See gdb.ada/call_pn.exp for examples.

I wonder if really 'unsupported' would be a better choice here - the
description in the Dejagnu manual says:

  UNSUPPORTED
       There is no support for the tested case.  This may mean that a
       conditional feature of an operating system, or of a compiler, is
       not implemented....

Which seems to describe this case perfectly, so:

    if {$test_count == 1 && [test_compiler_info {clang-*}]} {
      unsupported "compiler does not support rank 0"
      continue
    }

Thoughts?

Thanks,
Andrew


>
> I think failing the testcase at the iteration entry and continuing to test next case is the best way to handle these dismal results.  
> Please let me know your suggestions/comments on this.    
>
> Regards,
> Rupesh P