Simple spike at separating the core SpotlessTask from check and apply

[bigdaz]

• SpotlessTask runs the formatter for each source file, and generates an
  output file with the formatted results (if different)
  • If the task is not incremental, all prior outputs are removed
  • For each added file, the formatter is run, and the output is
    written if different from the original
  • For each removed file, the previous output is removed, if present
• SpotlessCheck will fail if there are any outputs from SpotlessTask
• SpotlessApply will copy any formatted outputs back over the original source

This mechanism has a number of benefits and makes things a lot simpler:

• The spotless cache is not required
• SpotlessTask is now a simple incremental task and benefits from the
  usual up-to-date checking. It should also support being made cacheable.

Note that a number of things were ignored for this spike:

• Anything to do with PaddedCell
• Check does not print out the actual formatting issues
• I didn't run any of the tests

[bigdaz]

@nedtwigg I spent some time experimenting with a solution to the cacheability of Spotless in Gradle. This is far from a finished solution, but demonstrates how I think it could work. In the end it's a lot simpler because there are no tasks that declare the same values as input and output.

Please take a look and let me know if you'd be interested in a more complete solution following along this vein. Thanks.

[nedtwigg]

Looks great, thanks very much! Unless we hit a showstopper (I don't see any), this will definitely get merged as we work through any issues, I definitely agree that it simplifies things. I'll take a deeper look later and leave comments and perhaps push a commit or two - there are a few things that pop out where I think I can simultaneously turn the unit tests green and make the total diff a little smaller. I guarantee a turnaround by tomorrow (Thurs) am.

[bigdaz]

Thanks Ned. This is definitely spike code, but any feedback/improvements are welcome.
I've tried not to use any "modern" Gradle features (e.g. Provider) but I didn't verify that they didn't sneak in.

[nedtwigg]

I wanted to dig in to the padded cell stuff, because I had a vague recollection of tech debt there. The story is:

• users expect spotless apply to be idempotent, but lots of formatters have edge cases
• we added paddedCell mode to fix them
• we added a way to detect if you don't need paddedCell anymore, since it was slower
• it turned out that people only need paddedCell for a single run usually. That one run would get them out of the buggy-zone of the formatter, and they didn't need it anymore
• so we tried to accomodate that common-case automatically, and ended up with paddedCell smeared all over everything

Looking at it with fresh eyes, we can leave paddedCell on at all times without any penalty performance compared to the smeared-thing we ended up doing, and that simplifies things greatly. There are a few things I'd like to assign myself in this PR:

• fix the tests
• cleanup our paddedCell mess

A question I have for you is: it doesn't make sense to send non-formatted spotlessApply results to buildcache, imo. Can we use cacheIf to disable buildcache if the output directory isn't empty? Is that crazy?

[nedtwigg]

More follow-up - this PR is much easier if PaddedCell is just always on. The refactor is easier/safer if you remove PaddedCell on its own, and then do the task restructuring that you've outlined here. I tried to do both at once, but I'm uncomfortable with how many tests are turning red for different unrelated reasons - we've got hard-earned tests for subtle gotchas in PaddedCell and in up-to-date checking, I don't want to accidentally throw any out by conflating them.

I'm hoping to push up the PaddedCell PR today, and once that's merged we can take a second pass on this PR.

[bigdaz]

OK, thanks Ned. So as I understand it, you're going to push a PR so that PaddedCell is always enabled. After that I can take another pass at this PR, this time in a more rigorous fashion.

[nedtwigg]

Perfect!

[nedtwigg]

@bigdaz okay! master is now ready. Should be a lot easier this time around :)