fulfill: try using SmallVec or Box for stalled_on

[lcnr]

Tested both Box and SmallVec for stalled_on, with both resulting in a perf loss.
Adds a comment mentioning this and removes an now outdated FIXME.

Logging the length of stalled_on resulted in the following distribution while building a part of stage 1 libs:

22627647 counts:
(  1) 20983696 (92.7%, 92.7%): process_obligation_len: 1
(  2)   959711 ( 4.2%, 97.0%): process_obligation_len: 2
(  3)   682326 ( 3.0%,100.0%): process_obligation_len: 0
(  4)     1914 ( 0.0%,100.0%): process_obligation_len: 3

cc @eddyb
r? @nnethercote

[nnethercote]

r=me if the perf CI run looks good:

@bors try @rust-timer queue

[rust-timer]

Awaiting bors try build completion

[bors]

⌛ Trying commit 8be007786cb0aa917163757d9ee5fa7d8a7a7d4a with merge 95fdf3ad69575b3abd39f50ddc1c669115d57664...

[bors]

☀️ Try build successful - checks-azure
Build commit: 95fdf3ad69575b3abd39f50ddc1c669115d57664 (95fdf3ad69575b3abd39f50ddc1c669115d57664)

[rust-timer]

Queued 95fdf3ad69575b3abd39f50ddc1c669115d57664 with parent 91fb72a, future comparison URL.

[rust-timer]

Finished benchmarking try commit 95fdf3ad69575b3abd39f50ddc1c669115d57664, comparison URL.

[lcnr]

Well, I did not expect that 😆

So my understanding is that we very frequently look at the elements of stalled_on without allocating anything. As SmallVec has to first check the current len to get the relevant pointer this
is actually slower than using a Vec.

I think I will go ahead and update the docs, so we don't forget this.

[lcnr]

I now changed stalled_on to Box<[TyOrConstVar<'tcx>]> and updated the comment for stalled_on.

Let's see if this changes perf 🤷

[jonas-schievink]

@bors try @rust-timer queue

[rust-timer]

Awaiting bors try build completion

[bors]

⌛ Trying commit a7c4b936ea3b3f79883bfc7eb7be0c76027cb664 with merge 9c4cd0972aa23430f490bc23cab84804323de3de...

[bors]

☀️ Try build successful - checks-azure
Build commit: 9c4cd0972aa23430f490bc23cab84804323de3de (9c4cd0972aa23430f490bc23cab84804323de3de)

[rust-timer]

Queued 9c4cd0972aa23430f490bc23cab84804323de3de with parent 74e8046, future comparison URL.

[rust-timer]

Finished benchmarking try commit 9c4cd0972aa23430f490bc23cab84804323de3de, comparison URL.

[nnethercote]

If you have a Linux machine I recommend getting familiar with local benchmarking and profiling, it'll be much faster and easier than having to create PRs and do perf runs on CI. Instructions are here.

[lcnr]

Updated the comment of stalled_on and didn't change anything else.

[nnethercote]

The new comment in the code looks good, thanks.

The first comment in this PR will be used in the merge commit. Can you update that so it describes the new change rather than the old change? That way the VCS history will be clearer.

[lcnr]

👍 Like this?

[nnethercote]

@bors r+ rollup=always

Thanks!

[bors]

📌 Commit 0441763 has been approved by nnethercote