bpo-10572 : Move sqlite3 tests to Lib/test/test_sqlite3

[erlend-aasland]

https://bugs.python.org/issue10572

[erlend-aasland]

I know you find this change controversial, @berkerpeksag, but I think it'll be nice in the end. Motivations:

• https://bugs.python.org/issue10572#msg179596
• test suites are simplified and less error prone (for example forgetting to include a test suite)
• it'll be possible to do python3 Lib/test/test_sqlite3/test_types.py, for example
• no special treatment in Makefile

What do you think?

[erlend-aasland]

BTW, history is preserved by git mv.

[berkerpeksag]

https://bugs.python.org/issue10572#msg179596

Unfortunately, I don't find Brett's arguments convincing. Linux distros are making too many customizations that would make stripping tests a trivial change compared to others.

test suites are simplified and less error prone (for example forgetting to include a test suite)

I don't even remember last time when this was an issue, to be honest.

it'll be possible to do python3 Lib/test/test_sqlite3/test_types.py, for example

This is a nice UX improvement. However, sqlite3 tests take almost no time to run and it's relatively easy to disable other test cases.

no special treatment in Makefile

Well, sqlite3 is hardly the only exception there :)

Also, we already made backporting fixes to maintenance branches automatically harder by renaming the test names. Moving tests around would make auto backporting almost impossible which I would like to avoid doing.

I'm sure you can find more interesting stuff to work on :)

[erlend-aasland]

test suites are simplified and less error prone (for example forgetting to include a test suite)
I don't even remember last time when this was an issue, to be honest.

Ok, but they're still simplified :) This PR removed a lot of lines.

it'll be possible to do python3 Lib/test/test_sqlite3/test_types.py, for example
This is a nice UX improvement. However, sqlite3 tests take almost no time to run and it's relatively easy to disable other test cases.

That's true

no special treatment in Makefile
Well, sqlite3 is hardly the only exception there :)

True, but it still declutters the Makefile (by one line)

Also, we already made backporting fixes to maintenance branches automatically harder by renaming the test names. Moving tests around would make auto backporting almost impossible which I would like to avoid doing.

Not impossible, but cumbersome. Anyway, it's a strong argument. However, the recent test changes will also make backporting cumbersome (for the next two releases only), which is a minor argument pro this change.

OTOH, this is no big deal for me. I'm ok with letting it go for now. It was a nice exercise, though :)

I'm sure you can find more interesting stuff to work on :)

I'm twiddling thumbs, waiting for #24135 ;)

Yes, there's a lot of stuff to work on. After the AC transition, I've got PR's for module state and multi-phase init. When we're there, I would like to get rid of the cyclic ref hack between Cache and Connection. I had a PR for it, but for some reason it failed on Windows. I haven't spent much time figuring out that issue, though. New features can wait for now, IMO.

[berkerpeksag]

I had a PR for it, but for some reason it failed on Windows. I haven't spent much time figuring out that issue, though.

I'm currently on Windows and I can take a look. It may take a while though because I mainly use it for gaming and I need to set up my development environment.

I'm closing this for now. We can consider it later. Thank you for your effort! :)

[erlend-aasland]

I had a PR for it, but for some reason it failed on Windows. I haven't spent much time figuring out that issue, though.
I'm currently on Windows and I can take a look. It may take a while though because I mainly use it for gaming and I need to set up my development environment.

Great, thanks!

We can consider it later. Thank you for your effort! :)

I sense a sliver of hope for this bpo issue ;) I'll keep the branch around