Simplify identity source map generator

[aleclarson]

This should be a lot faster than using the source-map library, since all we're doing is generating an identity line map.

[alangpierce]

Thank you! And sorry for the delay, work stuff has been a bit crazy recently.

I worry a bit about this code being really mysterious to readers or fragile if it gets any more complex. Could you bring back source-map as a devDependency and add a test that creates the mappings using that library and asserts that the results are the same? (You'll still need the RawSourceMap explicitly written out like you have, since it's part of the public API.)

An alternative to consider is something like ";AACA".repeat(numLines), which theoretically could be faster since it knows up-front how big the string is going to be. But I just tested it, and it seemed either the same or slightly slower, so what you have looks good to me!

[MaxGraey]

Also I recommend use this:

const NEW_LINE_CODE = "\n".charCodeAt(0);
for (...) {
  if (code.charCodeAt(i) === NEW_LINE_CODE) {
    ...
  }
}

instead:

for (...) {
  if (code[i] === "\n") {
    ...
  }
}

This could be also fast:

const NEW_LINE_CODE = "\n".charCodeAt(0);
let numLines = 0;
for (...) {
   numLines += ((code.charCodeAt(i) === NEW_LINE_CODE) | 0);
}
mappings = ";AACA".repeat(numLines);

But need tests

[alangpierce]

@MaxGraey thanks! I was actually skeptical that it would really matter (V8 seems to generally be smart about small code snippets), but from my testing it looks like that is measurably faster. So @aleclarson would be great if you included that as well.

Some informal measurements from running the benchmark about 5 times each:
460ms baseline (no source maps)
505ms with === "\n"
485ms with === charCodes.lineFeed

There's already a charCodes const enum, so that would be the way to do it in this case:

import {charCodes} from "./parser/util/charcodes";
...
if (code.charCodeAt(i) === charCodes.lineFeed) {

[alangpierce]

Moving to "requested changes" state per comments above. @aleclarson, would be nice to get this merged and released reasonably soon, since I think it does improve perf for my main use case. I'd be happy to take over if you don't have bandwidth.

[alangpierce]

Thanks!

Still would be nice to have a test that source-map and this code produce the same result, but it's fine for that to happen in a follow-up. In general feels like there's still a little more to do before this source map stuff is settled.