Re: patching the stable branch (was: Re: Octave 3.1.52 available for ftp)

[Thomas Weber]

Am Donnerstag, den 12.02.2009, 11:53 -0500 schrieb John W. Eaton:
> On 12-Feb-2009, WMennerich wrote:
> 
> | Both branches come from 3.0.
> | Hence, the patches included in 3.0.1, 3.0.2, 3.0.3 and 3.0.4
> | could also be necessary for 3.1.5x as long as they touch only
> | parts of the sources which are not different between 3.0 and 3.1.5x .
> | Or are the two branches already to different?
> 
> The general idea is that bugs are fixed in the main development branch
> 3.1, and then patches that are appropriate for the stable branch are
> applied there.
> 
> In the 3.0.x branch, we applied a lot of patches, attempting to fix
> many of the bugs reported, possibly at the expense of stability.
> 
> In the future, I would like to see us only applying patches to the
> stable branch that fix regressions from previously released versions
> of Octave.  Otherwise, I think applying too many patches does not
> tend to improve the stability of the series of "stable" releases, and 
> takes up effort that could be better used on the main development
> branch.
> 
> So, I would like to see our guidelines for the next stable release
> series be
> 
>   * If the bug was present in past versions of Octave (so not a
>     regression), fix it in the development branch only.
> 
>   * If a bug appears in the stable version that was not present in
>     previous versions (a regression), then the problem should be fixed
>     in both the development and stable branches.
> 
> No other patches should be applied to the stable branch without some
> discussion, and I would argue against applying the patch in nearly all
> cases.  I would probably only argue in favor of applying a patch for a
> bug that is not a regression if it is causes a serious problem, like
> making Octave crash.  But it seems likely that problems like that
> would also be regressions.
> 
> Comments?

This boils down to not fixing bugs, but only errors introduced when
fixing other bugs. Frankly, I'd consider such releases pointless. This
forces anybody on the development branch if they want a fix for a bug
that was also in a previous revision. 

Additionally, the work for checking older versions is non-trivial. Which
older versions do you want to check? Who is going to check it?
Distributions will either have an older version of the stable branch or
only the current one, against which the bug is reported.

Do you have any indication for a loss in stability of the stable branch?
I'd say it's still stable as a rock.

        Thomas