Also generate a tar.xz when running paver sdist

[tomspur]

Would it be possible to also generate a tar.xz, when generating sdists?

This patch generates a tar.xz when running paver sdist as a first step.

[rgommers]

Can you explain why you need this format?

And would it have to be paver sdist itself or would you be equally helped by an extra paver xsdist command for example? With this patch I'd have to delete this format straight after creating it in the release scripts, because I don't need it for uploading to sourceforge/pypi.

[coveralls]

Coverage remained the same when pulling 9006e0a on tomspur:tar.xz into 5309706 on scipy:master.

[tomspur]

With a current checkout, the tar.xz is about 30% smaller than the tar.gz. Offering a tar.xz file for downloading would therefore need less space and bandwidth and especially the former is often an issue, when offering a source distribution (see e.g. here).

That's why I added it topaver sdist on purpose, so it is always generated and it can be directly uploaded...

It would be very nice to have it on sourceforge/pypi (and many other projects also provide tar.xz as download option - and many other don't), but is not absolutely needed :)

[rgommers]

For me it's the latter right now, downloads time out regularly here. So there is a benefit to this, I'm fine with uploading .tar.xz to PyPi and SF. Although .tag.gz should remain the default.

[rgommers]

I'll merge this in a day or two unless anyone sees an issue with providing these downloads.

[tomspur]

Great. Thanks! :)

[rgommers]

Tried this and it fails for me on OS X 10.9:

xz dist/scipy-0.14.0.dev-dc79d97.tar
Build failed running pavement.sdist: Subprocess return code: 127

No time to debug right now.

[tomspur]

It seems you don't have xz installed on your system. Do you? I can modify pavement.py, so that paver tries to build it, but doesn't fail with a build error, when that didn't work.

The latest commit in this branch now only warns, when there is no tar.xz, but continues to create the rest of the usual paver sdist stuff.

[coveralls]

Coverage remained the same when pulling 351ac0d on tomspur:tar.xz into 5309706 on scipy:master.

[rgommers]

Indeed, last commit fixes the issue. Squashed and merged in 0da153e. Thanks @tomspur

[tomspur]

Thanks @rgommers