Remove recast and use ast-types directly

[danez]

This removes recast and only uses ast-types directly.

The performance increases significantly, but I haven't tested more different cases.

There are still some cases that do not yet work:

• Printing of nodes which were created in react-docgen by ast-types builders.
• Comments in printed code are not filtered out

Fixes #237

[fkling]

Amazing. Somehow I thought the whole NodePath business was part of recast (that would probably be more difficult to remove).

Looking forward to where this goes 😃

The performance increases significantly, but I haven't tested more different cases.

Can you share numbers?

[danez]

yeah I forgot to include that yesterday:

Before:

❯ node --expose-gc --max-old-space-size=2000 --max-semi-space-size=1500 --noconcurrent_sweeping ./benchmark/index.js
Node: v10.15.1
Running benchmark for CircularProgress.js ...
┌─────────────────────┬─────────────────────────────┐
│ fixture             │ timing                      │
├─────────────────────┼─────────────────────────────┤
│ CircularProgress.js │ 30.34 ops/sec ±7.82% (33ms) │
└─────────────────────┴─────────────────────────────┘

With PR:

❯ node --expose-gc --max-old-space-size=2000 --max-semi-space-size=1500 --noconcurrent_sweeping ./benchmark/index.js
Node: v10.15.1
Running benchmark for CircularProgress.js ...
┌─────────────────────┬─────────────────────────────┐
│ fixture             │ timing                      │
├─────────────────────┼─────────────────────────────┤
│ CircularProgress.js │ 40.55 ops/sec ±5.08% (25ms) │
└─────────────────────┴─────────────────────────────┘

The commandline options are to prevent gc to interfere, so the script runs without automatic gc.

The test file is taken from material ui. I need to find bigger or different (flow) files to test on. Would also be interesting to run this on a big codebase and see the difference.

[fkling]

Would also be interesting to run this on a big codebase and see the difference.

I get the hint 😉