Release 0.7.0

[kohr-h]

It's been quite a while since the last release, and it seems like still a while to go until 1.0, since there are a bunch of quite big and important changes still to be finalized and merged. See #1055.

So how about we release once before, say 0.9.0, with roughly the current status? I also seem to remember that there were some changes in the conda build process that need to be done, and we don't want to screw up 1.0 for that reason.

Edit: I'll start with this and maintain a checklist here that reflects the process outlined in the release documentation.

• 1. Agree on a release schedule
  • Open an issue on the issue tracker using the title "Release 0.7.0"
  • Discuss and agree on a set of open PRs that should be merged and issues that should be resolved before making a release
• 2. Make sure tests succeed (including large-scale tests) and docs are built properly
  • If anything fails, fix and make PR bevore proceeding (Fixes before release #1426)
  • Python 2.7
  • Python 3.{4,5,6}
  • Windows
• 3. Make a release branch release-0.7.0 off master, on a fork like for any other PR
• 4. Bump the master branch to the next development version (here '1.0.0.dev0')
  • odl/__init__.py
  • conda/meta.yaml
  • Make a PR and merge it
• 5. Compile and publish the release
  • Compile the release notes with user-visible changes
  • Increment the version number to '0.7.0' in odl/__init__.py and conda/meta.yaml
  • Change the git_rev field in conda/meta.yaml to 'v0.7.0'
  • Make a PR and fix review comments, using a single commit REL: bump version to 0.7.0 Done on master
  • Push the release branch to the main repository
  • Draft a new release on GitHub using the the release-0.7.0 branch and the short summary from the release notes file; add a link to the release notes page
• 6. Create packages for PyPI and Conda
• 7. Test installing the local packages and check them
  • From source
  • From wheel
  • From conda package
• 8. Upload the packages to the official locations
• 9. Merge the release branch Not merging

[aringh]

I was looking at this only a few weeks ago, and I agree that a release before the 1.0 release could be nice. We have a lot of new things since last release, but indeed there seems to be a bit left before 1.0.

Edit: we seem to have changes in over 400 files, with over 40 000 additions and 19 000 deletions...

[adler-j]

I agree that we should do this, but the version would be 0.7 according to standard numbering.

Do we have a list of things we want done here?

[kohr-h]

I'd suggest we only put stuff here that we can fix in relatively short time. Here's a list, feel free to edit:

PRs

• Add Mayo Reader #1321 (practically done)
• ENH: Add three solvers for differences of convex functions. #1307 (also done for all intents and purposes)
• Reconstruction pipeline #1280 (if possible)
• Default stepsize selection for optimizers #1286 (also close to done, and somehow good to have before ENH: Add three solvers for differences of convex functions. #1307)

The others are more fundamental or need significant work as far as I can judge.

Open issues

• Implementation of Bregman distances #1312 definitely since it's a new API, last chance to adjust

[adler-j]

That does look like a reasonable release schedule. For 1.0 we want tensor valued discrete lp as well as a cuda backend.

[kohr-h]

I'll go ahead with this.

[kohr-h]

Done.

[adler-j]

Freaking tremendous and "done" does not sufficiently convey the success :) Now lets put our aims towards a proper 1.0 release (and getting an article out as well).

Btw, what happened to: #1432

[kohr-h]

Btw, what happened to: #1432

That'll go in as a regular PR on master, tonight if it gets approval in time 😉
I'll copy over the "highlights" from the release page.