tests summary (was: AIX tests failures)

[Ralf Wildenhues]

[ sorry for replying to the wrong article, I'm still not subscribed ]

Jim Meyering writes:
> 
> If people like you exercise new releases and provide feedback about them
> (even if just to say "it builds and passes make check on system X.Y.Z")
> on older systems, we'll be more likely to spend time integrating patches
> to support those systems.

I'm not Albert, but: coreutils-5.96 builds and passes make check for me
on:
  AIX 4.3.3, 5.1, 5.2, 5.3
  Darwin 8.2.0
  FreeBSD 5.4
  HP-UX 10.20, 11.00, 11.11, 11.23 (hppa and ia64)
  IRIX 6.5
  OpenBSD 3.8
  Solaris 2.6, 7, 8, 9, 10
  Tru64 UNIX 4.0D, 5.1

except for the bugs that I have already reported, and the following
notes:

I've chosen a third-party perl other than /bin/perl or /usr/bin/perl on
some of the systems, so that the tests all worked (found all their perl
modules and so on).  Which needed the patch below, by the way.

I've used GNU make on a number of those systems, as some native makes
are not smart enough to detect that, say, dircolors.h is up to date in
the source tree (I always do VPATH builds).  That said, I've tried
native make mostly, and reported the other resulting bugs.  HP-UX
11.00 make seemed to hang at one point (I did not research this
further), and I stopped using Solaris make after the bug report.

I've always used the native compiler, hid most GNU software away, esp.
sed, awk and grep in order to expose more potential Autoconf issues.
And plain `make check', without being root, no expensive tests,
generally not on a local file system, and generally I am member of only
one group on test accounts, and tests have no controlling terminal.
(So you get an idea about the test coverage.)

Oh, forgot one thing: `grep -w' isn't working on (at least) IRIX, HP-UX,
Solaris (used in doc/Makefile.am), but that's known, and the test is not
interesting per-system anyway.  (The consequence isn't uniform though,
as some grep implementations error out, and some cause the test suite to
continue; so for user ease you may want to conditionalize this.)

On IRIX, during the tty-eof test, I get a bunch of
| Use of uninitialized value in ioctl at 
/opt/perl/lib/5.8.2/IP30-irix-thread-multi/IO/Pty.pm line 119.
| warning: TIOCSCTTY failed, slave might not be set as controlling terminal: 
Invalid argument at /opt/perl/lib/5.8.2/IP30-irix-thread-multi/IO/Pty.pm line 
120.
| Use of uninitialized value in numeric eq (==) at - line 41.
| Use of uninitialized value in ioctl at 
/opt/perl/lib/5.8.2/IP30-irix-thread-multi/IO/Pty.pm line 119.

and similar on HP-UX:
| Use of uninitialized value in ioctl at 
/opt/perl/lib/5.8.2/PA-RISC1.1-thread-multi/IO/Pty.pm line 90.
[...]
| Use of uninitialized value in ioctl at 
/opt/perl/lib/5.8.2/PA-RISC1.1-thread-multi/IO/Pty.pm line 90.

and Solaris:
| Use of uninitialized value in ioctl at 
/opt/perl/lib/5.8.2/sun4-solaris-thread-multi/IO/Pty.pm line 119.
| warning: TIOCSCTTY failed, slave might not be set as controlling terminal: 
Invalid argument at /opt/perl/lib/5.8.2/sun4-solaris-thread-multi/IO/Pty.pm 
line 120.

I have no idea whether they are a perl bug or something else.

Cheers,
Ralf

        * tests/du/Makefile.am (TESTS_ENVIRONMENT): Pass $(PERL), for
        files0-from test.

Index: tests/du/Makefile.am
===================================================================
RCS file: /sources/coreutils/coreutils/tests/du/Makefile.am,v
retrieving revision 1.21
diff -u -r1.21 Makefile.am
--- tests/du/Makefile.am        18 Jan 2006 06:29:14 -0000      1.21
+++ tests/du/Makefile.am        31 May 2006 16:12:52 -0000
@@ -14,4 +14,5 @@
   exclude no-x no-deref trailing-slash deref two-args slink
 EXTRA_DIST = $(TESTS)
 TESTS_ENVIRONMENT = \
+  PERL="$(PERL)" \
   PATH="`pwd`/../../src$(PATH_SEPARATOR)$$PATH"