Use JS module's length calculation for FormData values

[FreeMasen]

This is an attempt to address #211 by using JS::GetDeflatedUTF8StringLength to avoid encoding while calculating the length of values.

This change doesn't follow the suggestion to use variant<JS::Heap<JS::Value>, HostString> but I believe this would be equivalent to the existing implementation. If it would be preferable to use the memoization approach suggested I would be happy to close this and re-open when I can figure out how to implement that version.

Testing

I've added a new test to the integration tests explicitly to target length calculation in this path, please feel free to request more test cases.

[andreiltd]

Hey, thank for looking into this!

I had to refresh my memory on this one. Generally speaking I don't think Variant<JS::Heap<JS::Value>, HostString> will work here because we would need both. Otherwise when calling FormData#get or similar, for the variant holding HostString we would need to call core::decode to get the Value from string.

Overall I see few issues with the code in this patch:

• each call to JS_GetStringCharAt is internally calling JS_EnsureLinearString which potentially allocates and is very slow for iterating over characters, instead we should get linear string once and operate on its memory directly,
• JS_EnsureLinearString is fallible and its return value should be checked against nullptr
• compute_unencoded_normalized_len function should itself be fallible (because of potential allocation failures) so it should return mozilla::Result.

e.g:

  JSString *linear = JS_EnsureLinearString(cx, value);
  if (!linear) {
    return mozilla::Err;
  }

  JS::AutoCheckCannotGC nogc;
  const char16_t *chars;
  size_t length;
  if (!JS_GetTwoByteLinearStringCharsAndLength(cx, linear, &chars, &length)) {
    return mozilla::Err;
  }

  size_t defalted_len = JS::GetDeflatedUTF8StringLength(linear);
  for (size_t i = 0; i < length; ++i) {
    auto c = chars[i];
    ...
  }

I appreciate the tests, although in this case I think the WPT test suite coverage for FormData is quite good and the added test feel redundant. You can run the WPT tests like this:

just wpt-setup
just wpt-test

You should see the number of failures if your function is incorrect.

[FreeMasen]

Thank you very much for the feedback. I believe I have the changes locally working but will likely not be able to clean them up to push until the evening US Central Time today or tomorrow.

[FreeMasen]

I ran into an issue resolving with the mozilla::Result type at this level and instead am using std::nullopt as an OOM indicator. If that isn't the right choice I can move this function into the namespace where OutOfMemory is defined.

[andreiltd]

Thank you!

[FreeMasen]

All squashed up and ready to merge, thank you for all your support and suggestions while I worked through this