minification error breaks wasm2js initialization with -O2 --memory-init-file 0

[bvibber]

I'm trying to update ogv.js's upstream-llvm compatible builds to work with the latest goodies, including JS builds using wasm2js, but I seem to be having some trouble with minification at -O2 or above breaking the memory init functions when using --memory-init-file 0.

Using a latest-upstream build on macOS:

emcc (Emscripten gcc/clang-like replacement) 1.38.37 (commit d33f7a29002a2463a7956cd53a559b97a52f9560)

This code fragment from wasm2js.h:

    function(mem) {
      var _mem = new Uint8Array(mem);
      return function(offset, s) {
        if (typeof Buffer === 'undefined') {
          var bytes = atob(s);
          for (var i = 0; i < bytes.length; i++)
            _mem[offset + i] = bytes.charCodeAt(i);
        } else {
          var bytes = Buffer.from(s, 'base64');
          for (var i = 0; i < bytes.length; i++)
            _mem[offset + i] = bytes[i];
        }
      }
    }

ends up becoming this in output (exact var names vary depending on code built):

                    function(mem) {
                        var N = new Uint8Array(mem);
                        return (function(offset, s) {
                            if (typeof Buffer === "undefined") {
                                var O = atob(s);
                                for (var P = 0; R < Q.length; R++)
                                    N[offset + R] = Q.charCodeAt(R)
                            } else {
                                var Q = Buffer.from(s, "base64");
                                for (var R = 0; R < Q.length; R++)
                                    N[offset + R] = Q[R]
                            }
                        }
                        )
                    }

Note that the var bytes definitions becomes both a var O and a var Q in the separate if/else branches, but the uses in both branches are for Q. This causes an exception during initialization:

Uncaught (in promise) abort("TypeError: Cannot read property 'length' of undefined"). Build with -s ASSERTIONS=1 for more info.

because Q is hoisted and left at undefined rather than being set to the expected value.

Example test case:

#include <stdio.h>

int main(int argc, const char **argv) {
    printf("Hello, world\n");
    return 0;
}

emcc --memory-init-file 0 -O2 -s WASM=0 -o hi.html hi.c

[sbc100]

Presumably this is a bug in the upstream minifier.. :(

[bvibber]

Looks like my patch didn't quite fix it right. :(

I find this is still failing for me now, but in a silent way where it corrupts memory and later dies with an OOM. The for loops use the same variable name and it has the same problem, getting minified into the wrong variable for the non-Node case.

(Workaround is to disable minification, such as using --profiling, or use a memory init file which is probably more efficient anyway.)

I'll see if I can chase down the actual minifier bug at some point rather than chasing more symptoms...

[bvibber]

I'm also looking a little at the minifier; I think it's js-optimizer.js not acorn-optimizer.js that's being used here, but I'm not sure if anything from upstream uglify-js is doing that or if it's all the local optimization passes.

[sbc100]

Presumably we should address that bug in the optimizer rather than trying to tip toe around it, no? (Sorry I don't know this area of the code well so it not obvious to me who is to blame).

[bvibber]

Yes, I'm not sure yet either where the fault lies, but I think it's some of the operations in js-optimizer.js where it's renaming variables.

[stale]

This issue has been automatically marked as stale because there has been no activity in the past year. It will be closed automatically if no further activity occurs in the next 30 days. Feel free to re-open at any time if this issue is still relevant.

[kripken]

We've fixed a bunch of minification issues over the last year in wasm2js, perhaps this has been resolved?