Make submodule checks work under Python 3.

[takluyver]

My shot at fixing gh-3385.

• Ditch the imports from utils.submodule
• Move the is_repo check up: distro packaging will often not be running from a git repository, in which case packagers won't need to do anything. If we're not in a git repository, there's nothing sensible we can do about submodules anyway.
• Clearly label where packagers can patch to disable the submodule tests completely.

[minrk]

There is one case that the previous addressed that this no longer does: if someone downloads a tag from GitHub, it will be missing components and not be a repo. This should not be allowed to be installed, but this PR will allow an incomplete IPython to be installed without warning in that case.

[juliantaylor]

the daily build patches out the runtime git submodule check and will probably continue to do so for robustness.
so if it causes issues you can revert the original merge that caused the problems instead.

[takluyver]

It should be allowed to be installed from a github download - don't
forget the terminal IPython and the Qt console are perfectly usable without
the notebook static files. If there's a robust way of providing a warning
that those files are missing, we can do that, but it would need to:

• Work without importing the rest of the codebase
• Not require any interactive acknowledgement, because there are plenty of
  scenarios where you might be running setup.py without a user prodding it
  directly.

On 30 May 2013 17:53, Julian Taylor notifications@github.com wrote:

the daily build patches out the runtime git submodule check and will
probably continue to do so for robustness.
so if it causes issues you can revert the original merge that caused the
problems instead.

—
Reply to this email directly or view it on GitHubhttps://github.com//pull/3387#issuecomment-18693447
.

[minrk]

It should be allowed to be installed from a github download

No, it should not because you cannot download a complete sdist of IPython from GitHub.

[takluyver]

You can download a package which can successfully install everything apart from the Notebook. We should make it clear why the Notebook isn't working, e.g. with a runtime check for the notebook static files, but we shouldn't completely cripple that file just because it doesn't work for the notebook. Especially since detecting the case to cripple it makes life more complicated.

[minrk]

it should be a safe assumption that any install of IPython is complete.

[takluyver]

If you don't have tornado/pyzmq, you can install IPython but not run the notebook. The notebook static components function like just another dependency, except that they're in a git submodule. We can easily add a check that will prevent the notebook server from starting if you don't have those files, just like it doesn't start if tornado isn't installed.

Anyway, any objections to landing this PR?

[minrk]

The notebook static components function like just another dependency

Critical difference: you can't install ipython-components after the fact, you would have to remove and reinstall IPython. Only if we make the components a separate Python package would that be true.

Anyway, any objections to landing this PR?

No, I'll merge this now since it's critical, and figure out a way to prevent installation missing components later.

[takluyver]

Thanks.

The static components could be installed later, it's just slightly more complicated because you'd have to download, unzip and copy the files to the right location by hand. We could easily provide a wrapper script/function to automate that if it looks like many people will end up in that situation.