Hi Michael,

On Sun, 18 Aug 2002, Michael Widenius wrote:

[...]

MW>>>Melvyn> The test-suite, really needs a brush up though, so I'm unable to
report MW>>>Melvyn> any specifics in that area (which may or may not reveal that the
different MW>>>Melvyn> semaphore implementations are resulting in a bug). MW>>> MW>>>In the long run we plan to rewrite mysql-test-run to C to avoid MW>>>problems with a lot of different shells and commands.

Good choice.

MW>>>But for now we have just to patch mysql-test-run to work reasonable MW>>>well with all OS.

OK.

MW>>>Melvyn> Some things I've encountered: MW>>> MW>>>Melvyn> -- GNU time required, but no test-mechanism and time is MW>>>Melvyn> called as 'time', so even if GNU time is prefered in the PATH, the
shell MW>>>Melvyn> built-in takes precedence (BSD time gives the same information,
just on 1 MW>>>Melvyn> line, rather than 3) MW>>> MW>>>I have done a temporary fix for this by searching after 'time' in the PATH.

By default, /usr/bin/time takes precedence in a path. I think it make it easier on everybody if existence of variable GNUTIME=/path/to/time is available in the environment.

MW>>> MW>>>Melvyn> -- no skip-mechanism when latin1_de is not installed and it's only MW>>>Melvyn> installed with --with-charsets=all (additionally the available
charsets MW>>>Melvyn> are missing in the configure --help output) MW>>> MW>>>This will be fixed in MySQL 4.1. For MySQL 4.0 this is a bit hard to MW>>>fix :(

Yeah, I see the problem. Would be nice if something like this would work: CREATE TABLE vars SHOW VARIABLES LIKE 'character_sets'; SELECT COUNT(1) FROM vars WHERE Value LIKE '% latin1_de %';

Or: SHOW VARIABLES LIKE 'character_sets' VALUE LIKE '% latin1_de %';

The latter prefered, not only for this test, but in general, to automate some scripting.

MW>>>I tried to fix configure --help but was not able to get a variable MW>>>printed in the help text :( MW>>>(Need to learn more about autoconf to be able to do this).

Hmm, tried this and I agree - it's not that easy, because it's expecting text.

MW>>>Melvyn> -- rm uses trailing slashes (will give errors on BSD's) MW>>> MW>>>Found one case where this was used (fixed)

I got 2 warnings, but indeed, one spot.

MW>>>Melvyn> -- when installed, assumption of mysqld being in ../bin/ is wrong
and file MW>>>Melvyn> r/have_bdb.require is missing MW>>> MW>>>I have now fixed the mysqld is first searched for in libexec.

OK.

MW>>>In which distribution is have_bdb.require missing ? MW>>>(I just checked and it's included in the MySQL 4.0.3 source distribution)

4.0.3 isn't released, so it's 4.0.2, but the file is there, just not after make install, which is fixed by the following patch, to be applied in $(top_source_dir)/mysql-test and should be self-explanatory:

--- Makefile.am.dist Mon Jul 1 13:16:52 2002 +++ Makefile.am Sun Aug 18 16:36:24 2002 @@ -42,6 +42,7 @@ $(INSTALL_DATA) $(srcdir)/t/*.test $(DESTDIR)$(testdir)/t $(INSTALL_DATA) $(srcdir)/t/*.opt $(DESTDIR)$(testdir)/t $(INSTALL_DATA) $(srcdir)/r/*.result $(DESTDIR)$(testdir)/r + $(INSTALL_DATA) $(srcdir)/r/*.require $(DESTDIR)$(testdir)/r $(INSTALL_DATA) $(srcdir)/include/*.inc $(DESTDIR)$(testdir)/include $(INSTALL_DATA) $(srcdir)/std_data/*.dat $(DESTDIR)$(testdir)/std_data

--- Makefile.in.dist Mon Jul 1 14:16:41 2002 +++ Makefile.in Sun Aug 18 16:37:49 2002 @@ -341,6 +341,7 @@ $(INSTALL_DATA) $(srcdir)/t/*.test $(DESTDIR)$(testdir)/t $(INSTALL_DATA) $(srcdir)/t/*.opt $(DESTDIR)$(testdir)/t $(INSTALL_DATA) $(srcdir)/r/*.result $(DESTDIR)$(testdir)/r + $(INSTALL_DATA) $(srcdir)/r/*.require $(DESTDIR)$(testdir)/r $(INSTALL_DATA) $(srcdir)/include/*.inc $(DESTDIR)$(testdir)/include $(INSTALL_DATA) $(srcdir)/std_data/*.dat $(DESTDIR)$(testdir)/std_data

MW>>>Melvyn> Is this work in progress for the 4.0 release cycle, or would you MW>>>Melvyn> appreciate any compatibility patches now? MW>>> MW>>>Yes, the above are bugs that should be fixed in the 4.0 tree. MW>>>Any help you can give is welcome...

With respect to the latin1_de test, I think it should either be removed as it will interfere with any setup not specifying --with-extra-charsets=all, or latin1_de should be installed, but to do this, just to make the test suite work, is a bit overkill.

Ideally, the test should be kept at an alternate place, and configure should figure out, whether it should be copied (and removed by make clean) when latin1_de is in the configured charsets.

Or - the mysql-test-run script itself should accept the option --skip-test=name[[,name],...] which would be the long-term fix, even when moving to C this is usefull.