Re: Stable releases

[Ludovic Courtès]

Hi,

Neil Jerram <address@hidden> writes:

> I think we probably have taken a wrong turn, because I don't think the
> 1.8.x that we are on the verge of producing can be described any more
> as a "stable" series.  Surely the common connotations of "stable" are
> that the API is as unchanging as possible, and that the code is only
> changed in order to fix non-trivial bugs?
>
> And on the other hand, if 1.8.x isn't a "stable" series, how does it
> differ usefully from HEAD?

My viewpoint, based on the observation of the current and past release
process, was that HEAD should contain "revolutionary" changes like,
e.g., switching to GMP, providing a replacement to the GH API, and
eventually things like integrating Unicode support, integrating R6RS
library support, changing GCs (?), etc.

Conversely, the "stable" branch would not change such major components.
OTOH, since it may take a while before the unstable branch is usable, I
was happy with the integration of "minor" functionalities such as text
collation into the "stable" series.  But...

> Therefore, my feeling now is that we should revert to traditional
> "stable" handling for 1.8.x.  This would mean not merging enhancements
> from HEAD such as my debugging stuff and Ludovic's text collation
> work.  It would also mean that Rob's comments about limited testing
> requirement hold.

Adding new C code (as is the case with the text collation bug) might
indeed break builds on some platforms.  If this is the case, then it may
be the case that the series can hardly be regarded as "stable".  Adding
new Scheme modules, however, is unlikely to break builds.

To summarize: I think we could well have an "in-between" policy, that
is, allowing a little more than just bug fixes in the "stable" branch.
This would require careful evaluation of the problems/breakages that
could be caused by each individual new functionality, and this would
make the release process slightly heavier since more testing would be
required.  I believe many other large software projects (e.g., the Linux
kernel) have a similar policy.

BTW, I haven't (yet?) merged `(ice-9 i18n)' into 1.8.

> We certainly need at least one familiar person, but I'm sure it would
> be even better to have more than one.

Yeah, documenting and automating all this would be very helpful.

Thanks,
Ludovic.