handleBracketingOperators (src/support/string.h) is broken, and list parsing in asyncify is not functioning correctly.

[caiiiycuk]

Hi, I've come across a corner case where asyncify list parsing is failing with an error:

 Fatal: Asyncify: failed to parse lists

The list parsing mechanism is attempting to calculate nesting operators from the code:

Handles bracketing in a list initially split by ",", but the list may contain nested ","s. For example,
void foo(int, double)
must be kept together because of the "(". Likewise, "{", "<", "[" are handled.

The implementation performs simple nesting counting:

    for (const char c : part) {
      if (c == '(' || c == '<' || c == '[' || c == '{') {
        nesting++;
      } else if (c == ')' || c == '>' || c == ']' || c == '}') {
        nesting--;
      }
    }

This doesn't work for the function XStream::operator>(int&). Obviously, the nesting will be -1 at the end, and a fatal error will occur.

I don't have an idea on how to easily fix this. Moreover, I've been using asyncify for a long time, and this list format is hard to maintain. I think that using a one-line-one-function file format is much easier to maintain and will not have this issue.

What do you think? Any suggestion how to fix this?

[caiiiycuk]

link to another asyncify parsing problem #5271

[caiiiycuk]

I think we can add another option, like -sASYNCIFY_STRING_DELIM=newline, and if it is present, then use a one-line-one-function file format.

[kripken]

How about if we find a newline, then we assume that we should split by newlines (and do no other nesting computations, as you said)? That would avoid adding a new option.

[caiiiycuk]

It appears that -sASYNCIFY_WHITELIST=@file is parsing the file on its own because Binaryen receives the function list as a string, such as func1, func2, func3, instead of a file. Is it possible that the same bug exists on the Emscripten side as well?

[kripken]

I don't think the emscripten side does any parsing of bracketing, but it does expand response files, so @file would turn into the contents of that file. But that can be adjusted on a per-setting basis.