PEP 545

[JulienPalard]

This implements https://www.python.org/dev/peps/pep-0545/, building all known translations by default (only french for the moment, japanese may arrive soon.

This implements no crosslink between versions (no language switcher in the graphical interface) allowing to test for a period of time before going really public about it.

I tested it with:

$ mkdir /tmp/build_root  /tmp/www_root /tmp/log_root
$ git clone https://github.com/python/cpython.git --branch 3.6 --depth 1 /tmp/build_root/python36
$ git clone https://github.com/python/cpython.git --branch 2.7 --depth 1 /tmp/build_root/python27
$ git clone https://github.com/python/cpython.git --branch 3.5 --depth 1 /tmp/build_root/python35
$ git clone https://github.com/python/cpython.git --branch master --depth 1 /tmp/build_root/python37
$ python3 -m venv /tmp/build_root/environment/
$ . /tmp/build_root/environment/bin/activate
$ python3 -m pip install -r requirements.txt
$ python3 ./build_docs.py --git --build-root /tmp/build_root/ --www-root /tmp/www_root/ --log-directory /tmp/log_root/ --skip-cache-invalidation --group $(id -gn)

And it went well, properly skipping untranslated versions, building a correct hierarchy in www_root. It'll probably miss some work before implementing cross-linking like setting symlinks like /3/, and /dev/, which are currently done manually if I'm right, that I should implement in salt or in this script in a near future.

[JulienPalard]

highlighting @ned-deily @ewdurbin @benjaminp for review.

[vstinner]

Would it be possible to pass a list of arguments rather than a string to avoid issues with shell?

[vstinner]

Oh, wait, the "{}" or "%s" in shell commands are not new: the current code already use them. You can ignore my comment. It can be enhanced in a different PR ;-)

[JulienPalard]

Will gladly fix this in another PR yes.

[vstinner]

Can't use use cwd parameter of subprocess.Popen to avoid the "cd"?

Note: make also has a -C parameter to make the "cd".

Same question for string vs list.

[JulienPalard]

That's right, it will even allow running make directly (without a shell) once your previous comment fixed, which is nice. I tried to keep my changes minimal so I didn't changed this. Will do in another PR, later.

(As I don't have access to the logs or the server, I prefer keeping my chances atomic, so if something break we all know exactly what caused it).

[vstinner]

Ok, it perfectly makes sense. But it would help if someone could give you access to the logs since you seem to be one of the people working on this project :-)

@ewdurbin: Can you please give access to Julien to the logs? Or maybe send him logs on demand? Thanks.

FYI I proposed to give the commit bit to @JulienPalard on this project :-)
https://mail.python.org/pipermail/python-committers/2017-June/004583.html

[MarkMangoba]

@Haypo I can help get the logs, can you point me to what logs you'll need?

[JulienPalard]

@MarkMangoba @Haypo is speaking about the output of this crontab: https://github.com/python/psf-salt/blob/master/salt/docs/init.sls#L47 and /var/log/docsbuild/.

Having a copy of the logs right now is not needed, I'd only need them if I break something in one of my PRs, to diagnose.

[MarkMangoba]

@JulienPalard understood. @ewdurbin do you have a chance to review?

[vstinner]

Ok, LGTM.

[vstinner]

@MarkMangoba, @ewdurbin & others: how can we unblock this situation please?

[ewdurbin]

Running test builds now! Thanks @JulienPalard

[vstinner]

Cool, thanks @ewdurbin! How can we check for failure? Are you going to notify us? @JulienPalard: maybe join the #python-infra channel on Freenode? ;-)

[ewdurbin]

@Haypo indeed, unfortunately it's just logs on the docs server when the build job kicks off via cron.