Use a simple arena for annotatable items

[Kobzol]

When profiling bors with dhat, this was one of the places that were kinda hot for allocations. So I tried to use an amortized Vec for the annotatable items. I didn't store it in MacroExpander, because those are created repeatedly, but rather in ExtCtxt, so that it survives over multiple expansion calls, and the memory can be better reused.

Since we already pass &mut ExtCtxt to the expansion functions, I couldn't easily pass &mut Vec<Annotatable> to all the expansion functions too, because that would be a double &mut reference. So as an experiment for a benchmark, I just switched all the functions to directly push to the arena in ExtCtxt. This is not super pretty and only works since the ExtCtxt is currently not used from multiple threads. I wanted to refactor it, but I figured that it's maybe not so egregious? So first wanted to get a second opinion before I do that.

If we wanted to do this via &mut Vec<Annotatable>, I could store RefCell<Vec<Annotatable>> in ExtCtxt, and borrow_mut() it just before calling the expansion. But for that to work, I'd probably have to replace &mut ExtCtxt in the expansion calls (which might? be possible), or store Rc<RefCell<Vec<Annotatable>>> instead.

r? nnethercote

[rustbot]

Some changes occurred in compiler/rustc_builtin_macros/src/autodiff.rs

cc @ZuseZ4

[Kobzol]

@bors try @rust-timer queue

[rust-log-analyzer]

The job tidy failed! Check out the build log: (web) (plain enhanced) (plain)

Click to see the possible cause of the failure (guessed by this bot)

Diff in /checkout/compiler/rustc_builtin_macros/src/deriving/generic/mod.rs:546:
                         })
                         .cloned(),
                 );
-                cx.annotatable_arena
-                    .push(transform(Annotatable::Item(Box::new(ast::Item { attrs, ..(*newitem).clone() }))));
+                cx.annotatable_arena.push(transform(Annotatable::Item(Box::new(ast::Item {
+                    attrs,
+                    ..(*newitem).clone()
+                }))));
             }
             _ => unreachable!(),
         }
fmt: checked 7206 files
Bootstrap failed while executing `test src/tools/tidy tidyselftest --extra-checks=py,cpp,js,spellcheck`

[rust-bors]

☀️ Try build successful (CI)
Build commit: fc9f757 (fc9f7579dad04fdfc8d4534004d9a05d9f9f7555)
Base parent: 3cabe36 (3cabe36ceb022e2f56d4d330b1e2886f31117f18)

[rust-timer]

Finished benchmarking commit (fc9f757): comparison URL.

Overall result: no relevant changes - no action needed

Benchmarking means the PR may be perf-sensitive. Consider adding rollup=never if this change is not fit for rolling up.

@rustbot label: -S-waiting-on-perf -perf-regression

Instruction count

This perf run didn't have relevant results for this metric.

Max RSS (memory usage)

Results (primary 3.5%, secondary -3.4%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	3.5%	[2.9%, 4.1%]	2
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-3.4%	[-4.6%, -2.1%]	3
All ❌✅ (primary)	3.5%	[2.9%, 4.1%]	2

Cycles

Results (secondary 2.2%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	2.2%	[2.0%, 2.4%]	2
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Binary size

Results (secondary -0.0%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-0.0%	[-0.0%, -0.0%]	3
All ❌✅ (primary)	-	-	0

Bootstrap: 475.389s -> 502.254s (5.65%)
Artifact size: 402.85 MiB -> 402.88 MiB (0.01%)

[Kobzol]

Damn, I think I benchmarked this with an unrelated change before, and the win came from that change alone. Good thing I didn't do the painful refactoring :D Nevermind, closing.