Re: Install python modules from build directory (issue 575960043 by address@hidden)

[hanwenn]

On 2020/04/06 20:28:59, hahnjo wrote:
>
https://codereview.appspot.com/575960043/diff/557650043/make/lilypond-vars.make
> File make/lilypond-vars.make (right):
> 
>
https://codereview.appspot.com/575960043/diff/557650043/make/lilypond-vars.make#newcode20
> make/lilypond-vars.make:20: export
> PYTHONPATH:=$(top-build-dir)/python/$(outconfbase):$(PYTHONPATH)
> On 2020/04/06 19:50:35, hanwenn wrote:
> > On 2020/04/05 21:54:40, hahnjo wrote:
> > > On 2020/04/05 21:48:46, hanwenn wrote:
> > > > This will potentially make the out-www build fail. It's much
better to
> load
> > > the
> > > > source files directly, because they're always there and they're
always up
> to
> > > > date. 
> > > 
> > > I don't see why this should break: It's now back to the state it
was before
> > you
> > > changed it. But I can change it to out/ if you prefer.
> > 
> > Loading .py files from out/ directories is an antipattern in the
lilypond
> build
> > that makes working on the scripts a PITA, because the dependency
tracking
> > doesn't work, and you have to remember to rebuild them by hand.
Please don't
> > reintroduce this.
> 
> I don't really get that argument. As far as I understand, your recent
change
> ensures that python/ is built first to get the modules before they are
needed.
> AFAICT this dependency works correctly, and if you touch a .py file in
the
> source directory it is correctly copied over and byte-compiled. Maybe
you can
> describe what's an "antipattern" about this?

That assumes that you are building a complete build from the toplevel
directory. When working on (say) the mf/ directory (which needs tweaks
to the scripts under scripts/build/ ), one needs to do "make -C
../scripts/" by hand, because mf/GNUmakefile doesn't declare
dependencies on scripts/build/ .

Running scripts directly from source reduces confusion because effects
take place immediately after saving the source file.  
 
> > > > if you do it like this, this should depend on
$(outdir)/book_base_test.py
> > and
> > > > the rule should use $< . Otherwise, editing the test file
doesn't redo the
> > > copy.
> > > 
> > > No, default depends on $(OUT_PY_MODULES). And we really need all
modules
> > because
> > > book_base_test.py includes other files.
> > 
> > you're right.  I do think that this is not the way to go to suppress
the
> > __pycache__ directory. If that is a concern, we should just pass -B
to the
> > commandline.
> 
> It doesn't surpress __pycache__, it just uses the already compiled
modules in
> the build directory. This way there is exactly one location where the
cache is.

Sure, but in practical day to day development, we'll be running python
on source files directly, so I think we'll continue to see __pycache__
directories. It's probably simpler to put __pycache__ into .gitignore.

https://codereview.appspot.com/575960043/