[PATCH v2] libsframe: testsuite: Fix testsuite build on Solaris

[Rainer Orth]

Hi Alan,

>> >> Thanks for taking care of this.
>> >>
>> >> sframe-test.h in toplevel include/ looks odd IMO.  Adding to
>> >> libsframe/testsuite/ folder along with the -I${top_srcdir}/testsuite
>> >> addition in libsframe/testsuite/libsframe.decode/local.mk,
>> >> libsframe/testsuite/libsframe.encode/local.mk,
>> >> libsframe/testsuite/libsframe.find/local.mk may be a better thing to do.
>> >
>> > I thought about this, but this seems like lots of duplication to me.
>> > Besides, these files are extremly repetetive, doing the same over and
>> > over again.  I wonder if this cannot be simplified, avoiding the massive
>> > duplication to improve maintainablitly?  Let's see what Alan thinks
>> > about this.
>> 
>> any word on how to handle this.  The current duplication worries me,
>> especially for a patch that tries to reduce duplication in one place
>> increasing duplication even more in a different place.
>
> Did you try putting the new include file in libsframe/?  I agree with
> Indu that sframe-test.h does not belong in the top level include/.

my initial plan was to put it into libsframe/testsuite where it belongs.
Only when I saw the duplication required to do so did I decide to put it
into toplevel include which is already in the include path.

> Or even fix the makefile fragment duplication like this, and put
> sframe-test.h in libsframe/testsuite.  I've taken out -Wall from
> testsuite_CPPFLAGS and -I$(srcdir) from AM_CPPFLAGS because they don't
> seem to be needed.

That works just fine indeed, thanks.  I'll post a v3 patch soon.
Btw. how should I acknowledge your contribution to the patch?  In
ChangeLog's it's easy to see, but without them you cannot.  A
Coauthored-By: tag or some such?

> The testsuite object file naming is somewhat annoying too.  I don't
> know how to wrestle automake into dropping testsuite_libsframe_decode_
> prefix here:
>   CC       testsuite/libsframe.decode/testsuite_libsframe_decode_be_flipping-be-flipping.o

True, that's ugly.  However, I believe there's an underlying issue that
would fix that as a side effect: currently the testcases are built by
make -k check.  When some or fall of them fail to compile, as in the
present case, the error is easily lost in the wad of output produced by
make -k check.

I believe that the tests should be built by runtest instead.  This way,
every failure will be prominently visible in libsframe.sum, instead of
being hidden in the make output.

I usually use gcc's contrib/test_summary -t (used in toplevel make
mail-report.log) to collect all those summaries into a single file which
easily can be check for failures and regressions.  libsframe build
failures would show up there, too.

libsframe is weird in another way: it doesn't use ChangeLog's unlike the
rest of binutils.  Such inconsistencies are always confusing.

	Rainer

-- 
-----------------------------------------------------------------------------
Rainer Orth, Center for Biotechnology, Bielefeld University