babelPlugin fix proposal and benchmarking setup

[naugtur]

Note this PR has been opened against naugtur-import-meta-url branch. I wanted to discuss this commit separately

Working on import.meta I stumbled upon a case where if import.meta is used in an export statement it's not being reached by the visitor.
This is one of the possible implementations.

Other ideas include:

• calling path.traverse inside rewriteDeclaration, which makes the whole thing take 2x the time
• replacing import.meta in the first ('Analyze') pass, which is controversial.

I think this fix may also fix other issues we didn't discover yet.

I'm aware this has lint issues and a console.log leftover. I'm setting up the PR to make discussion easier

[naugtur]

Not sure if we want this.

I needed something to check if I'm not introducing an awful performance regression.

I'm not super happy with the benchmark. Differences between my implementation and previous implementation were less significant than differences between the same implementation executed at different times.

I tried running both implementations in one suite and warming them up first. I could commit that too, but I doubt we want to keep multiple implementations laying around.

Example:

names correspond with fixtures, 2 means it's a new implementation (specifically one iteration before what's committed here)

small x 698 ops/sec ±11.71% (75 runs sampled)
large x 68.98 ops/sec ±3.19% (63 runs sampled)
exportheavy x 1,032 ops/sec ±4.57% (82 runs sampled)
small2 x 790 ops/sec ±2.57% (82 runs sampled)
large2 x 69.91 ops/sec ±5.07% (60 runs sampled)
exportheavy2 x 1,060 ops/sec ±2.10% (82 runs sampled)

small2 x 632 ops/sec ±11.21% (75 runs sampled)
large2 x 69.29 ops/sec ±2.80% (66 runs sampled)
exportheavy2 x 993 ops/sec ±3.53% (81 runs sampled)
small x 803 ops/sec ±5.18% (78 runs sampled)
large x 75.10 ops/sec ±2.25% (67 runs sampled)
exportheavy x 1,069 ops/sec ±2.15% (84 runs sampled)

With warmup:

warmup x 2,371 ops/sec ±9.89% (78 runs sampled)
small2 x 745 ops/sec ±4.39% (79 runs sampled)
large2 x 70.02 ops/sec ±3.31% (63 runs sampled)
exportheavy2 x 1,038 ops/sec ±2.53% (81 runs sampled)
small x 848 ops/sec ±3.36% (82 runs sampled)
large x 64.16 ops/sec ±8.13% (56 runs sampled)
exportheavy x 939 ops/sec ±6.42% (78 runs sampled)

warmup x 27.49 ops/sec ±7.32% (51 runs sampled)
small2 x 766 ops/sec ±2.59% (79 runs sampled)
large2 x 74.26 ops/sec ±2.04% (64 runs sampled)
exportheavy2 x 984 ops/sec ±4.15% (78 runs sampled)
small x 849 ops/sec ±3.78% (82 runs sampled)
large x 70.81 ops/sec ±5.13% (61 runs sampled)
exportheavy x 896 ops/sec ±9.47% (71 runs sampled)

[naugtur]

just some code to parse

[naugtur]

just some code to parse

[michaelfig]

I won't be able to review this for a little while, as it will need some close attention.

Overall, I think doing some of the transformations during the "analyze" phase is not so bad. They are really just "phase1" and "phase2" because I couldn't figure out how to do things in one pass.

[michaelfig]

Thanks for the primer on what you did here. Looks great!