Logging in make test

[simo]

On Tue, 2011-08-09 at 06:35 +1000, Andrew Bartlett wrote:
> On Fri, 2011-08-05 at 15:37 +0000, simo wrote:
> > On Thu, 2011-06-02 at 08:46 +1000, Andrew Bartlett wrote:
> > > On Wed, 2011-06-01 at 10:04 +0200, Andreas Schneider wrote:
> > > > Hello,
> > > > 
> > > > I would like to revert commit ea95e0967b33d75acc960b5308a667e956ba4f29.
> > > > 
> > > > If we log to stdout we loose a lot of information in Samba3 like filename, 
> > > > line of code, function name and especially the *time and date*. The logs are 
> > > > more or less useless. If you have multiple daemons it is more or less not 
> > > > possible to read the log correctly.
> > > > 
> > > > http://git.samba.org/?p=asn/samba.git;a=commitdiff;h=886ca981b27324157e492925e598b5f4b9c09574
> > > 
> > > The reason this change was made is that Samba otherwise logs
> > > to /usr/local/samba (or whatever the prefix is) during startup of make
> > > test, and that's not a good idea.
> > 
> > I t looks like the right way here is to add an option to the smbd binary
> > to change the default logging directory then.
> > 
> > > If our stdout logging isn't useful for a developer, then it probably is
> > > equally useless for our end users and then we need to fix that.
> > 
> > Yes feel free to fix that, meanwhile though logs are useless given
> > samba3 doesn't use a single process model.
> > 
> > > Also, using a predicable filename also makes it possible for the
> > > selftest system to capture the server-side logs of a failing test.  (See
> > > getlog_env() in selftest/target/Samba4.pm and
> > > selftest/target/Samba3.pm).  This wasn't done properly in the past. 
> > 
> > And it is useless in the present with samba3 as you'll find everything
> > mixed together.
> > 
> > As it stands the only way to get reasonable logging when running make
> > test in samba3 is to revert your patch explicitly.
> > 
> > I think I'd be ok if you want to make it an option for make test, but
> > unless you make it configurable I would really like you to revert it,
> > please.
> 
> Simo,
> 
> Clearly the logging environment in make test isn't ideal for your needs,
> or else you would not have reason to complain.  I'm sorry about that.
> 
> 'make test' has many purposes, both as a regression suite and a
> development environment.  It certainly isn't perfect, but I would argue
> that it's pretty darn good given the alternatives, and supports us
> pretty well.  Having logging output appear correctly via the 'make test'
> infrastructure (selftest.pl et al) is key to ensuring the output of
> autobuild and the build farm captures server-side errors, which is in
> turn key to ensuring that issues in these automated environments can be
> debugged.
> 
> I'm more than happy to review changes you may propose to make this more
> useful to developers for interactive use, and it would seem reasonable
> to have the debug header in the stdout logging, but simply reverting
> this patch isn't the answer. 

I have no experience with the test framework and since this looks like a
regression I am asking you to fix it by at least providing an option. I
am sure that you can do that much more easily than I can given you know
the framework and I do not, it would also seem that given you broke make
test for developers I would see as your duty to fix it too.

I guess I'll keep carrying the revert locally until it will work in the
meanwhile.

Simo.

-- 
Simo Sorce
Samba Team GPL Compliance Officer <simo at samba.org>
Principal Software Engineer at Red Hat, Inc. <simo at redhat.com>