Re: patch #1 - towards a generic backend datastore

[Mike M. Volokhov]

On Thu, 24 Feb 2005 00:38:03 -0800
Mel Hatzis <address@hidden> wrote:

> Mike M. Volokhov wrote:
> > On Wed, 23 Feb 2005 22:50:16 -0800
> > Mel Hatzis <address@hidden> wrote:
> > 
> > [snip]
> > 
> >>OK, I'll go ahead and commit to the mainline tomorrow morning (PST).
> >>Note that I ran the changes through a fairly comprehensive set of
> >>regression tests (which I put together) and they appear to be sound.
> > 
> > 
> > Wow. I'd love to see such tests in tree so as when huge changes come,
> > it's sometimes bit difficult to found underwater stones. Many thanks!
> 
> If there's general interest in incorporating the test suite I have,
> I'd be happy to try and set something up.
> 
> There's a perl component in it though...for setting up a test database
> against which the regression tests are run. I'm not sure how people
> feel about adding a perl dependency to the mix.

These checks are needed to developers or beta-users only, when other
users ususally will not ran any regression tests at all. Thus IMHO it
can be implemented in any covenient way.

> My original idea was to have the tests run via a 'make test, but this
> was not possible since GNATS needs to be (partly) installed in order
> to run...the binaries all have a hard-wired reference to the 'databases'
> file. We could fairly easily overcome this by modifying the binaries
> to allow the databases file to be supplied...either by environment
> variable or command line option (or both).

Hmm... We also may create a database named, say "testdb", and encourage
testers add the following line to databases file:

        testdb:GNATS Test Database:/path/to/testdb

But we must use a new executables, which can use any path to databases,
specified at compilation time. 

I'm pesonally have my own copy of the GNATS installation and test
database by specifying --prefix=/usr/home/mishka/gnats at configure
stage. For regression tests I believe there are no needs of running
gnatsd via network.

> The perl requirement could perhaps, be resolved, by simply committing
> the generated database to the mix.

Well, why not commit both (if database is not huge)? The database
creation may be another important part of test itself. Moreover, testers
can download reference database as separate tarball.

--
Mishka.