Support new import attributes "with" syntax.

[freshp86]

The new with syntax will soon be required by Chromium (and other Chromium based browsers), and the old assert syntax will no longer work (see https://chromium-review.googlesource.com/c/v8/v8/+/5512879).

Also see [1], [2] for the status of the "import attributes" feature. The older syntax had been implemented at #1095, and it seems that some similar updates are needed to support the new syntax.

[1] https://caniuse.com/mdn-javascript_statements_import_import_attributes
[2] https://chromestatus.com/feature/5205869105250304

Bug report or Feature request?

Bug.

Version (complete output of terser -V or specific git commit)

5.31.0

Complete CLI command or minify() options used

{
  module: true,
  compress: {},
  mangle: {},
  output: {},
  parse: {},
  rename: {},
}

terser input

import mysheet from './foo.css' with {type : 'css'};

terser output or error

ERROR: Unexpected token: keyword (with)

Expected result

Should not crash.

[freshp86]

Also possibly related bug at acornjs/acorn#1289.

[fabiosantoscode]

What should be the output for people who wrote assert? Should it be garbage-in-garbage-out or always output with? assert wasn't JS syntax proper, but it might be supported by some browsers that users of assert support, that wouldn't support with.

[fabiosantoscode]

This is why I should always wait for stage 4 :)

[freshp86]

What should be the output for people who wrote assert?

Whatever happens today? I don't think the feature request here to support with should affect what happens when assert is used. Terser authors can decide when/if support for assert should be completely removed.

Should it be garbage-in-garbage-out or always output with?

Personal opinion is that a minifier should not make assert -> with transformations to the code, as this is not strictly speaking a minifiers responsibility. If the code used assert it should output assert. Having said that, I am not a Terser author, so just my perspective.

[fabiosantoscode]

You are correct that the minifier shouldn't convert between them, at least in theory.

I'll wait for Stage 4 to implement, otherwise I might have to support 3 different versions of the feature.