fix($interpolate): use the first end symbol that forms a valid expression

[lgalfaso]

When an interpolation string has multiple end symbol choices, pick the
occurrence that forms a valid expression

Closes #8642

[caitp]

Then we spend potentially a lot more time parsing =\ also try/catch blocks --- it looks like it will prevent us from throwing in the case of invalid expressions, am i wrong?

[lgalfaso]

Updated the PR with another approach that keeps track of (){}[] and strings. It uses some extra bytes but does not have a try..catch

[shahata]

This was kind of a hard read, I think I would refactor this to something like:

var brackets = [[0, 0, 0], [0, 0, 0]], bracketChars = '({[)}]', ch, quoteCh;
index = startIndex + startSymbolLength;
while (index !== endIndex || quoteCh || !equals(brackets[0], brackets[1])) {
  ch = text[index];
  if (quoteCh) {
    if (ch === '\\') {
      index++;
    } else if (ch === quoteCh) {
      quoteCh = undefined;
    }
  } else if (ch === "'" || ch === '"') {
    quoteCh = ch;
  } else if ((i = bracketChars.indexOf(ch)) != -1) {
    brackets[i/3][i%3]++;
  }
  index++;

  if (index > endIndex) {
    endIndex = text.indexOf(endSymbol, index);
    if (endIndex === -1) {
      break;
    }
  }
}

(haven't tested, I might have some bug there)

[lgalfaso]

The suggested code will fail with $interpolate('{{"\\}}"}}'), anyhow it is possible to fix it, but it does add some other oddities

[shahata]

As I said, that's just a suggestion for a direction to get rid of this pretty cryptic switch statement, I didn't really test it. Anyway, I modified it a bit so it will work with the failure you mentioned.

[shahata]

I think I like the try/catch better since my guess is that this implementation is much less performant in "sunny day" cases (where the expression does not contain }}).

[caitp]

The try/catch case is broken because it causes us to silently fail when there are legitimate syntax errors.

I don't really expect this bug to actually get fixed, since it's been around for so long and people rarely get bit by it

[shahata]

Not sure what you mean, looking at lgalfaso@2c64dc4 it seems to me that it will re-throw the last exception when it runs out of endSymbols ahead. Regarding the chances of this getting fixed, you're probably right :)

[lgalfaso]

The PR is in place to start the discussion on if this should be fixed and if that is the case, the solution that should be implemented

[lgalfaso]

@caitp not sure why you say that the PR silently fails on syntax errors, it will still throw when these are found
Also updated the PR so it it cleaner and saves a few bytes

[caitp]

I saw that it rethrows the second time I looked at it, but I still wouldn't expect it to land