On Sat, 25 May 2013, Jakub Wilk wrote:

Welcome to DPMT!

Thanks!

I don't intend to sponsor this package, but here's my review:

I have a couple of DD friends I can bug to do the actual sponsored upload, but review from folks with more knowledge of Python packaging and team procedures is very useful, thanks.

Upstream provides documentation. It might be a good idea to build and ship it.

Lintian says:
I: pygithub source: debian-watch-file-is-missing

Fixed, using githubredir.debian.net.

P: python-github: no-upstream-changelog
P: python3-github: no-upstream-changelog

There isn't one. I assume I shouldn't be creating one out of git shortlog or something... although that does remind me that the upstream README.rst wasn't getting installed, which is now fixed.

P: python-github: no-homepage-field
P: python3-github: no-homepage-field

Fixed.

I would drop the Provides fields:
http://lists.debian.org/debian-python/2011/03/msg00139.html

Oh, I see. Good to know. (That also makes moot my question of why ${python3:Provides} was not being set.)

Re override_dh_auto_install: -O0 is no-op when used together with --no-compile. But then, I wouldn't use --no-compile, as is thwarts possibility of spotting byte-compilation errors early.

I copied that from ScottK's packaging for python-ipaddr. I'm not sure why it's there, but I can remove both of those since I don't think there's any reason for them in this package.

Do tests require Internet connectivity? If no, then it would be good to run them at build time.

Only one of them does (JSON encoding, github/tests/Issue142.py). I can try to arrange to skip that one; the rest all pass under `unshare -n net`.
(There's some mechanism for mocking the API's responses.)

That said, I thought the eventual goal was to move tests to autopkgtest instead of the build process? Are we sufficiently far away from that reality that today I should still be putting this in the build process?

Typos in upstream code:
explicitely -> explicitly
instanciate -> instantiate

The code uses assert to validate types of method arguments. That's not what assert is for.

I'll file issues / patches upstream about both of these, although no promises about upstream agreeing with the latter stylistically. (These should be converted into TypeErrors, right?)

Does pygithub validate SSL certificates?

Ugh. Good call, and I'm embarrassed not to have checked. Looks like it uses httplib.HTTPSConnection, which doesn't. I will definitely come up with a patch and file a pull request upstream before uploading!

Don't ignore errors from "rm -rf build". "-f" takes care of ENOENT and you failures certainly should not go unnoticed.

Oh, good to know. I'd always seen people ignoring errors from rm in Makefile clean targets, but your reasoning makes sense.

--
Geoffrey Thomas
http://ldpreload.com
geo...@ldpreload.com


--
To UNSUBSCRIBE, email to debian-python-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Archive: 
http://lists.debian.org/alpine.deb.2.00.1305251554530.27...@dr-wily.mit.edu

Reply via email to