"Native" emscripten build environment

[SoniEx2]

Emscripten should figure out a way to provide an execution environment with the following features:

• posix_spawn/system, using filesystem emulation and unsafe-eval if needed (to load wasm modules from the emulated filesystem).

This would enable building and running thngs like make and uname in emscripten, which in turn plays nicer with build systems, which in turn makes it possible to build more stuff. And because it's not seen as cross-compilation, the build system can avoid making assumptions such as whether fork works, and it can also generate artifacts as needed (e.g. libgcrypt/libgpg-error need to generate some stuff on the host machine, which isn't really possible with cross-compilation).

[sbc100]

Support for multiple "processes" is indeed something that emscripten lack any support for at all. I agree it would be quite an interesting and ambitious project to add such support.

Getting this working to extent where once can actually run "make" inside emscripten would be a lot of work indeed. There are some folks who have tried this with some success: https://browsix.org/

[Antony74]

Here's the current state of GNU Make built by emscripten and run with nodejs is, to the best of my knowledge:

I've asked around to if anything like this already existed: Adding make to the npm toolchain?

And no-one came up with anything. This also gave me the opportunity to defend this only-slightly-crazy idea on two different fronts.

GNU Make builds fine with emscripten, but running it with a very simple Makefile with nodejs on Windows we get

make.js: getcwd: No such file or directory
node reverse.js result.json input.json
make.js: *** [Makefile:2: all] Error 127

The getcwd thing is a curious side-effect of using what the build system thinks is a Posix build on a Windows file system and is easily patched on the JavaScript side.

The error is caused by the fact posix_spawn is stubbed as follows

weak int posix_spawn(pid_t *pid, const char *path,
                   const posix_spawn_file_actions_t *file_actions,
                   const posix_spawnattr_t *attrp,
                   char *const argv[], char *const envp[]) {
  errno = ENOSYS;
  return -1;
}

My main worry is what else might be lurking in the GNU Make source code which rather than giving a compile-time error, is stubbed by Emscripten to give a run-time error. We might not get to find out until a user falls over an obsure piece of Make functionality.

It looks possible to me to convert the way this is stubbed to the way getcwd is stubbed, and implement it on the nodejs side with child_process.spawn - I mean the difficultly is above average because there's callbacks for stdout, stderror, and an exit code, but I think it should probably be within my abilities. Currently I've only got as far as building LLVM as it's literal hours, and bootstrapping a fork of emscripten, but that's my plan :-)

[sbc100]

It seems reasonable to add support for posix_spawn under node, perhaps as past of NODERAWFS. However, if make is going to spawning all these native processes, why not just have make it self be native too? I.e. what is the point of have make itself be in wasm/js?

Regarding LLVM you should not need to build it yourself. You can get it from emsdk. For information on how to use the emsdk versions of binaryen and llvm with fork of emscripten see https://emscripten.org/docs/tools_reference/emsdk.html#how-do-i-use-my-own-emscripten-fork-with-the-sdk

[Antony74]

what is the point of have make itself be in wasm/js?

I think I answered that in the reddit thread, but basically just for the sake of having a clean npm toolchain with no additional dev tools required.

Regarding LLVM you should not need to build it yourself. You can get it from emsdk.

Thanks! I didn't get a chance to start looking at that until this evening... I think I'm gradually making headway with it...

[Antony74]

Sorry, I'm missing something - the part of the process where emscripten's system/lib/libc (where posix_spawn lives) gets built.

On consecutive docs pages we:

• Setup the toolchain
• Configure Emscripten settings
• Verify the Emscripten dev environment

I keep looking at these pages thinking I've missed a step.

There are no obvious configure or make files except the Makefile and bootstrap.py in the root folder, which don't build libc. I've introduced an intentional compile error in emscripten_libc_stubs.c and rerun them just to be sure.

I think I'm going to feel terribly silly when I hear the answer to this question, but how do I build libc (and anything that needs building along with it), please?

[kripken]

@Antony74 Normally the toolchain builds libc etc. for you automatically (and the emsdk builds contain precompiled libraries) but you can also do it manually using the embuilder tool. Do ./embuilder --help for docs, but basically embuilder build libc will build libc, etc.

[Antony74]

Thanks @kripken I'm afraid it's taken me a while to get back to this project, but I have managed to configure emscripten from source and get to where I was before with building GNU Make. Hopefully, from here on this is no longer a configuration problem (boo!) but a programming problem (hurray!).

So there's a docs page which says

Emscripten itself is written in Python and JavaScript so it does not need to be compiled.

which is 100% true, but potentially confusing for anyone who's peeked inside the repo and noticed there are C/C++ system libraries. My best attempt to write clarifying text can be found here

Antony74@8da276e

Very happy to open an issue/pull request as required, and work with whoever to achieve help-text which works for everyone.

(I've got a very minor PR open here for a few spelling mistakes I found in README.md files and source code banner comments. I don't know what's going on with the CI).

Hope that's helpful, and thank you very much for your help in getting this far! :-)

[kripken]

@Antony74 Good idea in Antony74@8da276e - I think that is worth adding! The only change I'd suggest is to mention embuilder there also.

[Antony74]

Thanks @kripken, PR for docs page enhancement opened.

[Antony74]

Above we've been assuming GNU Make uses posix_spawn to run the commands within a Makefile. By default this is not true - it actually uses vfork and execve. I think it would be a little bit crazy to try and simulate those in nodejs, so a better approach might be

#define USE_POSIX_SPAWN

(And at some point I need to figure out why nothing happens when I add -D USE_POSIX_SPAWN to my emcc command line)

When I do this I get a startup-time error. i.e. after compile-time but before run-time

    failed to asynchronously prepare wasm: LinkError: WebAssembly.instantiate(): Import #16 module="env" function="__syscall_faccessat" error: function import requires a callable
    Aborted(LinkError: WebAssembly.instantiate(): Import #16 module="env" function="__syscall_faccessat" error: function import requires a callable)
    /git/make/make.js:219
        var e = new WebAssembly.RuntimeError(what);
                ^
    
    RuntimeError: Aborted(LinkError: WebAssembly.instantiate(): Import #16 module="env" function="__syscall_faccessat" error: function import requires a callable). Build with -sASSERTIONS for more info.
        at abort (/git/make/make.js:219:13)
        at /git/make/make.js:256:13
    
    Node.js v20.18.0

And I'd like to file a bug report saying this is the wrong error at the wrong time, but I'm having trouble isolating it.

GNU Make calls eaccess (lib/findprog-in.c). I can comment them all out and it runs (past that point at least). But I had to comment quite a lot out to make that happen, so I uncomment just the first one, and the problem returns. So that's sort of isolated, but when I try to write a test program it compiles, starts, and runs no problem

// emcc -O2 -s NODERAWFS=1 access.c

#define _GNU_SOURCE

#include <stdio.h>
#include <unistd.h>
#include <fcntl.h>

int main() {
    printf("Calling an access function\n");
//    int result = access("./result.json", X_OK); // Works OK
//    int result = faccessat(AT_FDCWD, "./result.json", X_OK, AT_EACCESS); // Works OK
    int result = eaccess("./result.json", X_OK); // Also works OK :-(
    printf("Called with result %d\n", result);
    return 0;
}

So I'm afraid I'm a bit stuck and not sure what to try next. The commands I'm using to build GNU Make are

emconfigure ./configure
emmake make
find . -name "*.o" -type f | xargs emcc -O2 -s NODERAWFS=1 -o make.js

[Antony74]

Slightly improved commands for building GNU Make:

CFLAGS="-DUSE_POSIX_SPAWN -sASSERTIONS" emconfigure ./configure
emmake make
find . -name "*.o" -type f | xargs emcc -O2 -s NODERAWFS=1 -o make.js

That answers my "how to I set USE_POSIX_SPAWN" question, but leaves the error message completely unchanged. Which also means I've somehow failed to follow the instruction to "Build with -sASSERTIONS for more info" correctly.

[Antony74]

I can confirm that the equivelent gcc commands produce an apparently well-behaved ./make executable.

CFLAGS="-DUSE_POSIX_SPAWN" ./configure
./make

Apologies for not double-checking this the moment I hit the present error, but I've done it now: this an emscripten-problem(or a my-ignorance-of-emscripten-problem).

[sbc100]

Which also means I've somehow failed to follow the instruction to "Build with -sASSERTIONS for more info" correctly.

-sASSERTIONS is a link time flag not a compile time flag. Are you sure its getting passed to the linker? Can you see the final link command in the output of make?

[sbc100]

It sounds like you fixed the initial problem? If so I guess we can close this issue?