fix: token after strict mode block is evaluated in strict mode

[kaicataldo]

Q	A
Fixed Issues?	Fixes #11183
Patch: Bug Fix?	👍
Major: Breaking Change?
Minor: New Feature?
Tests Added + Pass?	Yes
Documentation PR Link
Any Dependency Changes?
License	MIT

We're currently eating the closing bracket token of the class body before restoring the outer scope's strict mode state. This PR changes it so that we restore the out strict mode state before eating this final token.

Question: Does this seem like the correct way to fix this?

[JLHwung]

I suggest we delay throwing octal error when reading a number token. So even in this.nextToken() the next token is read in an incorrect strict mode, we can still throw later when we are sure this.state.strict is correct.

babel/packages/babel-parser/src/parser/expression.js

Line 1019 in d13fd7c

return this.parseLiteral(this.state.value, "NumericLiteral");

We can delay throwing octal literals error when we finished parsing a numeric literal

        const start = this.state.start;
        const strict = this.state.strict;
        node = this.parseLiteral(this.state.value, "NumericLiteral");
        if (strict && this.state.containsOctal) {
          this.raise(start, "Legacy octal literals are not allowed in strict mode");
        }
        this.state.containsOctal = null;
        return node;

And in the tokenizer part we can set containsOctal: true only and allow tokens be generated, so it will be checked later with correct strict.

babel/packages/babel-parser/src/tokenizer/index.js

Line 1045 in d13fd7c

if (this.state.strict) {

I think we should also get rid of

babel/packages/babel-parser/src/tokenizer/index.js

Lines 227 to 228 in d13fd7c

	this.state.containsOctal = false;
	this.state.octalPosition = null;

as long as we correctly reset these states after parsing literals.

Sidenote: other than tokenizer#readNumber(), we also risk incorrect state.strict in tokenizer#readEscapedChar.

babel/packages/babel-parser/src/tokenizer/index.js

Line 1309 in d13fd7c

} else if (this.state.strict) {

But I think it is okay to leave it as-is because this function will never called in the nextToken of a tt.braceR, which may be a boundary of strict mode. tokenizer#readEscapedChar will only called after consuming a tt.backslash token.

[JLHwung]

Can you also add a test case about function declaration, which is failing on master

function A() { "use strict"; }
05

[kaicataldo]

Thanks for the review! Will take a look later this afternoon.

[kaicataldo]

Waiting to land #11188 first since the two PRs actually overlap a bit.

[JLHwung]

FYI acornjs/acorn@44e52b2 is how acorn fixes this issue. That is a neat fix, maybe we can port it here?

[kaicataldo]

This is ready for re-review!

[nicolo-ribaudo]

Thanks!

[existentialism]