sync-build is broken

[vbraun]

sage -sync-build doesn't understand wildcards, so it deletes a bunch of correctly-installed sage.symbolic modules.

Really, the whole concept is broken. We should at the very least record what is being installed instead of trying to guess it in sync-build. Even better, make it part of the build process.

There is quite a lot of setup code, none of which is doctested. And, therefore, broken. As this ticket shows. I've started to move setup code into a sage_setup package that is installed with Sage and can be doctested in the usual manner.

Also:

• Having SAGE_LIB point to SAGE_SRC is actually never useful, removed.
• re-enable out-of-tree cythonize

[vbraun]

New commits:

​d1078bb	integrate sync-build into setup.py
​b3679e5	Having SAGE_LIB point to SAGE_SRC is actually never useful, removed
​d6330c1	automatically doctest sage_setup
​af65bdd	make use of the new SAGE_LIB
​da33134	remove annoying warning from sage -ba

[mathzeta2]

I'm not up to review this, but I do have two questions:

1. Why not remove -ba-force if it is exactly the same as -ba? At least its description should be updated.
2. In find_python_sources and installed_files_by_module the use os.chdir seems uneeded. Can we just use a different argument to os.walk?

[vbraun]

1. removal needs deprecation etc. Really it doesn't hurt to keep that alias around.
2. its faster to construct the right relative path from the get-go than having to strip off a leading component.

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​7b26839

re-enable out-of-tree cythonize

[leif]

I think you can remove the sdist variable entirely from setup.py, i.e., also

if len(sys.argv) > 1 and sys.argv[1] == "sdist":
    sdist = True
else:
    sdist = False

[vbraun]

Yes, the setup.py should be broken up and doctested. And we need to decide whether we want to support sdist or not (in which case the Manifest.in can probably go, too). But thats for another ticket.

[leif]

Replying to vbraun:

And we need to decide whether we want to support sdist or not (in which case the Manifest.in can probably go, too).

Is that (still) used anywhere in Sage? (I mean calling Sage's setup.py with sdist.)

[vbraun]

Off-topic, but as far as I know we don't currently use distutils sdist anywhere.

[leif]

Just for the record: This doesn't fix my docbuild-broken upgraded Sage version (but I didn't really expect it to either; I think it's rather some Sphinx-brokenness w.r.t. Python's multiprocessing).

[vbraun]

No, thats fixed in #16440

[ohanar]

Mostly looks good, I haven't checked cython modules, but it doesn't seem like it really handles them at the moment (which would agree with some TODO I think I read in the documentation of some function). Are you planning on adding support for them at the moment, or on a future ticket?

I added a few minor reviewer changes.

---

New commits:

​e9ad958

small changes to initial sage_setup package

[vbraun]

Just to clarify: Building / cleaning of Cython modules works. Only the doctest that no cython files are being cleaned after a successful build needs work. In particular, the module_list.py would need to be importable after building. Could be done with a sys.path hack, but I'd rather do it properly later on.

[vbraun]

PS: any further functionality should go to a future ticket.

[ohanar]

Ah, yes. Looking at it again, seems fine. Everything works as expected.