Re: the "separate, but integrated" website proposal

[Carl Sorensen]

On 8/4/09 4:00 AM, "John Mandereau" <address@hidden> wrote:

> Le mardi 04 août 2009 à 01:12 -0700, Graham Percival a écrit :
> 
> 
> Yes, even power users could run lilypond and output-distance on entire
> input file directories and detect changes in formatting.  This is a goal
> for 2.14, or more reasonnably after 2.14.0.
> 
> 
>> Our word "stable" refers to the input syntax, so it still has
>> meaning.  I certainly admit that we want to avoid regressions in a
>> stable branch, but we already have the regtests (which nobody
>> looks at) for this purpose!
> 
> Regtests comparisons should be enough to check the regtests, I don't
> know if they still work.
> 
> 
>> Yes.  I mean, if somebody asked me "hey, want to break this now?",
>> I'd say no, of course.  But a regression occurs, I'm not going to
>> cry.  And more to the point, I'm not going to expend a lot of
>> energy trying to ensure that no regressions occur.  Given the lack
>> of *other* people willing expend this effort, I don't think I'm
>> unique in this regard.
> 
> If you are willing to allow development on stable branch with accepting
> regressions, then please put a big fat warning on the download page to
> warn about this, otherwise we'll have many complaints from users.  I
> think allowing development with regressions in stable branch without
> having working regtests comparison is unsane.
> 
> 
I think that there are two different meanings for regressions, as I've
followed this thread.

The first is failed regression tests.  We should not release stable with
failed regression tests (and the regression test facility should allow us
to easily do that testing).

The second meaning is "failed complex examples, like we might see in the
introduction".  I think that complex example failures are a problem, not
with the application, but with the example.  As part of the documentation,
we need to be sure that the examples are adjusted to work with the current
lilypond version.  But I don't think that making the examples work (as
currently written) is a necessary part of updating the lilypond application.
Over time, we improve things (like the spacing engine) which could make the
layout change and eliminate (or change) the need for a specific tweak in a
complicated example.  The proper response there is to change the example,
not to change the application, IMO.

So, I have heartburn about regressions on the stable branch as indicated by
failure to pass the regtest.  But I have no heartburn about "regressions"
that mean a complicated example that uses tweaks doesn't look right any
more.

Thanks for listening,

Carl