Re: Your Friendly Neighborhood LSR

[Mats Bengtsson]

Valentin Villenave wrote:

> > The version line might be good information to have - at least in a
> > comment. If a snippet stops behaving correctly, the original author of
> > the snippet is gone, nobody any longer really remembers what the snippet
> > was supposed to do - then the version line will give a hint on which
> > ancient lilypond to use in order to see the snippet as the contributor
> > intended.
> >    
>
> No. If we ever find such a snippet, and if we really have *no idea* of
> what it's supposed to do, *no way* to contact its author and *not a
> clue* about how to make it work, we just delete it.
>
> The last time I deleted a snippet that was not a duplicate, it was an
> obviously deprecated TeX-based snippet that, frankly I don't care if
> it was built on 2.2 or 2.4.
>  
Right, but if it had been marked as version 2.10 and it
looked suspicious in 2.10.30 (or whatever version is
used in LSR), then I'm sure you would have spent some
more time investigating. Considering the amount of bug
fixes done during the 2.10 series, I wouldn't be surprised
if there are some snippets that really illustrate bug workarounds
and have become obsolete even during 2.10. However, this is
not my main point, see below.

> The current LSR version could be more visible, I agree.
> Now, should the snippets include a \version? I'm not sure; this would
> imply that *all* snippets, excepted those marked as "version-specific"
> would begin with the same \version "2.10" or \version "2.12"...  After
> all, verbatim fragments from the manual never include a \version line,
> do they?
>  
No, but when you read the manual, you have explicitly chosen
what version of the manual you are looking at, in contrast to
LSR, which only exists in one version.
Also, we have had quite a number of questions on the mailing
list over the years, related to the fact that people haven't
realized that an example they used didn't match their LilyPond 
installation.
So, I strongly urge you to show a \version line for every snippet.
However, if the database internally stores the snippets with or
without this \version line, does not matter for me. If it's not
included within the database, then the web server can
add the line before showing the code in the user interface.

>  
>
> > However, Rune has also got a point that it would
> > be good to know at least about the last version where we are sure that the
> > snippet worked as originally intended. Valentin, I think it was you who
> > sent some questions some half year ago when you tried to understand and
> > update snippets that were outdated (and had not even been updated
> > correctly with convert-ly). In the upgrade to the next stable version, I'm
> > sure that you will make sure to use convert-ly and manually review every
> > snippet to make sure that it produces correct output, but especially for
> > more
> > complex snippets it may still be easy to miss that there have been some
> > unintentional changes in the result.
> >    
>
> Hence the "version-specific" tag. Suppose some day we drop the native
> Postscript markup backend like we did for TeX. As soon as we did, it
> would be really easy to search every snippets including a \postscript
> command, and tag them as "version-specific", before considering
> modifying or removing them.
>  
I think many bug fixes and updates are much more subtle than
so and you really have to compare the output of every snippet
between the two versions to be 100% sure that it really works
as intended. You might want to consider some automated
regression test like is done in the LilyPond distribution, though,
to simplify this task.

>  
>
> > Does code ever work in one version, then not work in a subsequent
> > version, then work again in an even later version? I can see where it
> > might, at least at the x.x.xx level. If that's the case, snippets could
> > end up with multiple version lines.
> >    
This has happened over the years, where some bug was
introduced and then fixed some versions later. However,
collecting and keeping track of such fine-grained information
is virtually impossible.

  /Mats