MAINT: Remove manual expansion of template loop for some ufuncs

[eric-wieser]

There's templating here that loops over @S@ = 'U', ''. yet we don't actually use it to generate some of the functions.

It's a lot less obvious where to add new ufuncs before this change.

Extracted from another PR, because this part needs no discussion

[charris]

LGTM pending tests.

[eric-wieser]

Appveyor looks to be broken in some way - none of the jobs seem to have run for any PRs...

[eric-wieser]

Why do some appveyor builds run at numpy/numpy, yet others at charris/numpy?

[charris]

Because appveyor didn't support groups when we started using it. If I want to edit tests I need to log into appveyor as charris, where I run the numpy tests as a member of the numpy team. I don't know what the current situation with appveyor is, their web interface and documentation sucks is lacking, but ideally the tests would be run by the numpy project and any developer could deal with them. If that is now possible we should set it up.

[charris]

Oh, and if I want to switch my appveyor login, I need to delete their cookies. Perhaps I should disable cookies for that site.

[charris]

Appveyor looks to be rerunning tests for the past months/years. it did that once before, I had to spend a long time cancelling hundreds of tests to get it back to working. I think it is time to look for another place to run windows testing. Scipy looks to use https://circleci.com/. @rgommers Would you recommend circleci.com? Any oddities?

[rgommers]

SciPy uses CircleCI to build the html docs; it's nice for that because it keeps build artifacts around for inspection.

Timeouts are pretty tight IIRC, and we haven't considered CircleCI for Windows builds.

[rgommers]

Appveyor is a massive pain to admin, but it's still the best option I think.

[eric-wieser]

Appveyor looks to be rerunning tests for the past months/years.

Specifically, numpy/numpy seems to be. charris/numpy is up to date, roughly. How is github deciding whether to hit the webhook for your account or numpy?

[charris]

Beats me. Looks like the numpy account broke about 9 mos ago. I suspect that accounts for why merged PRs immediately show broken tests, which suggests that github merges are getting tested.

[eric-wieser]

Restarting it caused it to use the other hook, I think.

Any objection to removing the numpy/numpy hook? Seems like there's a race condition about which one is picked, and most of the time we get lucky

[charris]

The numpy project has a hook, which I deleted and readded, which cleared the queue. We weren't able to get it to display last time. It would be nice if everything ran through a single interface, but that will require some experimentation to figure out what works. It gets worse, the reason that I need to delete cookies to login to a different account is that there is no way to log out that I can discover.

[charris]

OK, I deleted the numpy project from the numpy account. Let's see what that does.

[charris]

And later we can try deleting from the charris account and adding to the numpy account.

[eric-wieser]

Good to merge?

[charris]

Yep, go ahead.

[charris]

Looks like if we want to set up an appveyor numpy organization account we will need a special email account, as appveyor accounts are identified with email addresses.