Re: Python 3

[Jonas Hahnfeld]

Am Samstag, den 21.09.2019, 11:25 +0200 schrieb David Kastrup:
> Jonas Hahnfeld via lilypond-devel <
> address@hidden
> > writes:
> 
> > Am Samstag, den 21.09.2019, 12:09 +1000 schrieb Andrew Bernard:
> > > So let me get this straight.
> > > 
> > > It's not the case that GUB is completely broken. We can still build 
> > > releases.
> > > 
> > > DK is working steadily to cherry pick items for 2.20.
> > > 
> > > Python 2 to Python 3 is a major issue.
> > > 
> > > So, I offered to do the 2->3 port a long time ago but circumstances 
> > > prevented me from doing so. Would it be constructive if I launched into 
> > > that aspect? I cant understand the lilypond internals code but I have 
> > > extensive Python experience.
> > > 
> > > Would this be helpful to moving forward?
> > > 
> > > But have people already started on this, I thought?
> > 
> > I've also started looking into this and used the branch
> > dev/knupero/lilypy3devel as a starting point (see also 
> > https://lists.gnu.org/archive/html/bug-lilypond/2019-08/msg00014.html
> > ).
> > 
> > On top of that I've worked on the attached patches which brings the
> > make targets check / test and doc back to life with Python 3.7.4. Maybe
> > they can be added to the branch mentioned above to serve as a single
> > source of truth? I know the third patch is pretty large, let me know if
> > I should try to split it up...
> > 
> > As I'm pretty new to the development of Lilypond, I'm not really sure
> > if there's something else to do in the source code repository itself?
> > I'm pretty sure I didn't get to run through all error branches in all
> > scripts, but the targets mentioned in the documentation work for me
> > right now. If not, sounds like working on GUB would be next?
> 
> I haven't checked yet, but at the current point of time, the best
> patches will be those running under both Python 2 and Python 3 without
> having to special-case code.  Those can be applied to master and thus
> minimize the actual amount of code switching we ultimately need to do.

I agree that this would be ideal, but pretty hard: Already the result
of running 2to3 can't be executed under Python 2 in some cases or may
start to give different results. Examples include
 * "isinstance(s, unicode)" -> "isinstance(s, str)", which is something
different in Python 2, and
 * "import StringIO" -> "import io", which didn't have all functions in
Python 2.

If that is the way required to get Lilypond ported to Python 3, I can
try to work into that direction. But finally, I guess there will need
to be a hard switch to Python 3...

My 2 cents,
Jonas

signature.asc
Description: This is a digitally signed message part