Daniel Stenberg <daniel_at_haxx.se> wrote:

> On Wed, 21 May 2014, Fabian Keil wrote:
>
> > I'm still doing that and would like to get the required runtests.pl changes
> > upstream after making them more generic.
>
> Thanks for keeping it up Fabian. I'm in favor of making sure that the curl
> test suite can be re-used for more purposes as I believe it can help improving
> more code and it will help the test suite and thus it also improves curl
> itself.

Great. I'll look into making the patches more generic in the near feature.

> > I attached some of the patches I'm currently using as examples.
>
> I'm fine with all of them really. My only gripe is that they add functionality
> and features not really exposed or exercised by curl itself so there's a big
> risk that we ruin this by mistake as we go along.
>
> It would be good to add a test for curl that uses the feature for that reason
> alone.

Agreed.

> > Patch 0003 isn't absolutely required and the chosen syntax turned out to be
> > flawed
>
> ... but you're still proposing we merge it?

Yes, but not in its current form. As there don't seem to be any objections
against the idea in general, I propose I change the patch to use a more
convenient syntax. I think it would be sufficient to allow to specify
multiple test number on the same line. For example I currently need:

test:0030: Expected to fail with Privoxy. In case of missing server headers Privoxy generates an error message the test doesn't expect.
test:0037: Expected to fail with Privoxy. In case of missing server headers Privoxy generates an error message the test doesn't expect.
test:0066: Expected to fail with Privoxy. In case of missing server headers Privoxy generates an error message the test doesn't expect.
test:1079: Expected to fail with Privoxy. In case of missing server headers Privoxy generates an error message the test doesn't expect.

Which could be expressed as:
test:30, 37, 66, 1079: Expected to fail with Privoxy. In case of missing server headers Privoxy generates an error message the test doesn't expect.

BTW, this could be partially tested by converting tests/data/DISABLED
to the "manifest" syntax which would have the advantage that runtests.pl
could show the reason why the test is disabled instead of merely listing
them as "configured as DISABLED".

Fabian

-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette: http://curl.haxx.se/mail/etiquette.html