Re: Reported location of test-suite.log

[Ralf Wildenhues]

* Bob Friesenhahn wrote on Sun, Feb 21, 2010 at 05:22:45PM CET:
> On Sun, 21 Feb 2010, Ralf Wildenhues wrote:
> >
> >Is $(subdir) wrong in your Makefile, or are you using a package
> >hierarchy with sub-configure scripts?  In the latter case, autotools so
> >far never produce relative paths wrt. the topmost package.
> 
> $(subdir) is always '.' for my non-recursive project.

Well, but TEST_SUITE_LOG may actually be in a subdirectory, and then
even if $(subdir) is '.' it would differentiate between different test
suites within one package (except Automake parallel-tests doesn't
support that mode of operation yet).

>  Totally
> useless info.  Even if it contained './foo/test-suite.log', it would
> be not be much more helpful.

> To clarify, I run test suites for many different builds on many
> different systems and at the end of the test run, there is just this
> little summary message in a terminal window at the end of the run
> which gives no clue as to what directory the build was done in.

Thanks for providing this additional hint, without which the bug report
was not understandable.

A general comment: If you look at the output of a typical build from an
autotooled package, then there is very little that varies for different
builds.  So this problem actually applies not only to this case, but to
many different cases as well.

There is one crucial exception: `make' prints notices when it enters and
leaves a directory.  These notices contain absolute directories.  They
typically come right after a failure such as the one you posted.

>  In
> some cases, my build/test scheme uses a command script which causes
> the build to be done in some other directory so that when the build
> completes, the original directory is restored.  Without some other
> way to glean what I did, when I come back to a terminal session, it
> is not always easy to find the relevant test-suite.log file.
> 
> If an absolute path is provided instead, then there is no confusion.

Yes, sure, and I'm really undecided about this particular case, but I'd
prefer a clear logic of which set of messages all would need to provide
an absolute path and which don't need to, so that we can fix all
instances and not keep iterating.

Thanks,
Ralf