perf(semantic): use faster indextree building

[YoniFeng]

Related to issue #115, but not necessarily the final change.

Local profiling shows improvement as expected (see comment in the issue), which isn't large/significant.
I opened an indextree PR, will need to wait for feedback.

[Boshen]

I'll setup the CI benchmarks today. I also invited you to the repo so github can post benchmark results.

[YoniFeng]

Running the linter on typescript.js (which has 3 lint errors and runs <1s):
main

You can see the top hit's callstack originates in create_ast_node() -> append()

pr

Notice I couldn't figure out how to group based on "indextree"/call stack.
If I manually sum everything >1ms originating in append vs append_new_node(the pr version), the reduction is from 26ms to 13ms.

A few notes:

1. This improvement is only noticeable with no/few errors. When there are many errors, miette used to print the diagnostics dominates time/CPU usage by an order of magnitude, making the linting work itself negligible.
   (The screenshot profiles were run linting typescript.js. It's rather large, but has a lot of comments (which aren't added to the AST, right?) and only 3 lint errors. When I run for babylon.max.js (which has hundreds of errors), miette dominates all time significantly.)
2. These are one-off profiles. A proper benchmark will give an indication of actual bottom-line gain.

[Boshen]

The CI actions has proper benchmarks now https://github.com/Boshen/oxc/actions/runs/4336428902

The numbers are really significant!

[YoniFeng]

Ha I just noticed right before you posted.
I rebased on main to make sure the parser isn't actually regressed.

[Boshen]

I rebased on main to make sure the parser isn't actually regressed.

~2% change is just noise.

---

Shall we wait for a week for indextree or do we start looking for alternatives?

[Boshen]

I can use your clone for the time being, since indextree is rarely updated.

[YoniFeng]

Let's wait.
I believe indextree's maintainer will work tomorrow and respond.

[YoniFeng]

Squashed, ready for merge

saschagrunert/indextree#92 was merged

[Boshen]

https://github.com/Boshen/oxc/blob/bc2c17547291b1127bd4ec1feb2c6c8f4ebc027e/crates/oxc_semantic/src/scope/builder.rs#L39-L40

The scope tree is also an indextree, let's update it as well and see the overall benchmark result.

[Boshen]

Parser Benchmark Results - ubuntu-latest

semantic/babylon.max.js    1.73    244.0±6.85ms    42.3 MB/sec    1.00    141.0±7.08ms    73.2 MB/sec
semantic/d3.js             1.97     19.8±0.38ms    27.6 MB/sec    1.00     10.1±0.46ms    54.2 MB/sec
semantic/lodash.js         2.25      5.7±0.13ms    89.9 MB/sec    1.00      2.5±0.24ms   202.3 MB/sec
semantic/pdf.js            2.17     14.9±0.67ms    27.0 MB/sec    1.00      6.8±0.20ms    58.7 MB/sec
semantic/typescript.js     1.64    160.5±7.80ms    59.9 MB/sec    1.00     97.6±6.74ms    98.5 MB/sec

[Boshen]

Thank you so much for working on this @YoniFeng I'll praise you whenever I can.