Use mainReadsParams metadata from wasm-emscripten-finalize #9064
Conversation
… (this change avoids the binaryen addition of the flags causing tests to break, as we assert on surprising flags)
src/postamble.js
Outdated
| #endif | ||
| ]( | ||
| #if MAIN_READS_PARAMS |
There was a problem hiding this comment.
This is tricky to read, because we're parameterizing on two #ifs and this is better than having 4 slightly-different copies of the same statement.
Would this still work if we had earlier
#else // not MAIN_READS_PARAMS
var argc = 1;
var argv = 0;
#endif? Because then we can leave the _proxy_main/_main logic alone.
There was a problem hiding this comment.
Sounds reasonable, yeah - it will add 0,0 to the code size that way, but that might be slightly better for the VM anyhow (to not have to coerce an undefined). I'll change it.
src/postamble.js
Outdated
| #endif | ||
| ]( | ||
| #if MAIN_READS_PARAMS | ||
| argc, argv |
There was a problem hiding this comment.
This drops the 0, I assume that's fine? Presumably an undefined can get synthesized to a 0 in the case where we have a char* envp[] in main
| run_process([PYTHON, EMCC, 'yes.c', '-O3', '-o', 'yes.js']) | ||
| yes = os.path.getsize('yes.js') | ||
| # not having to set up argc/argv allows us to avoid including a | ||
| # significant amount of JS for string support (which is not needed |
There was a problem hiding this comment.
because metadce... that's really cool :)
…n-core#9064) This lets the wasm backend path not emit code to set up argc/argv if main doesn't need them. It also avoids that code bringing in some string and stack allocation support (metadce improvements are from the latter). This reduces --closure 1 -O3 size of hello world by almost 10%.
This lets the wasm backend path not emit code to set up argc/argv if main doesn't need them. It also avoids that code bringing in some string and stack allocation support (metadce improvements are from the latter).
This reduces
--closure 1 -O3size of hello world by almost 10%.