Re: bizarre failures on NetBSD 9.3 x86_64

[Dennis Clarke]

On 5/10/23 16:58, Bruno Haible wrote:
> Dennis Clarke wrote:
> > bash-5.1$ echo $CONFIG_SHELL
> > /usr/pkg/bin/bash
> > bash-5.1$
> >
> > Guess what ?
> >
> > ============================================================================
> > Testsuite summary for GNU patch 2.7.6
> > ============================================================================
> > # TOTAL: 44
> > # PASS:  42
> > # SKIP:  0
> > # XFAIL: 2
> > # FAIL:  0
> > # XPASS: 0
> > # ERROR: 0
> > ============================================================================
> >
> > Gotta love the need for GNU bash for this to work. Makes no sense at all
> > for "patch".
>
> Interesting. I came up with a workaround, by modifying test-lib.sh, and it
> is in fact already committed in git:
> https://git.savannah.gnu.org/gitweb/?p=patch.git;a=commitdiff;h=7623b2dc0d1837ecfd58f32efc78e35834deeb38
>
> But for me, then I get a test failure, see attached log file.
>
> The error messages "shift: can't shift that many" and
> "test: ==: unexpected operator" indicate another shell problem.

I see that and have no idea why that happens. On NetBSD 9.3 AMD64 the
solution seems to be that we need CONFIG_SHELL set to the bash shell.
That is not portable nor does it make sense but it works in that it
allows the testsuite to run cleanly. Which prooves to me that the binary
that was just compiled works fine. The testsuite is not the final
product for me ... I just want that gpatch binary!  So maybe we need to
test all this on a really strict POSIX type machine which will stomp all
over anything not strictly portable. May be a waste of time. I have a
friend with an IBM MainFrame in his life and those things are a
nightmare for anything that is not strictly in compliance. Might be
worth a look.


--
Dennis Clarke
RISC-V/SPARC/PPC/ARM/CISC
Four decades in production systems.