WasmFS NodeRawFS is not consistent with NodeRawFS (non-WasmFS version)

[trybka]

Please include the following in your bug report:

Version of emscripten/emsdk:

emcc (Emscripten gcc/clang-like replacement + linker emulating GNU ld) 4.0.11 (d3432e7a2a6585331c94b041973fb16c9f332ce4)
clang version 21.0.0git (https:/github.com/llvm/llvm-project 0f0079c29da4b4d5bbd43dced1db9ad6c6d11008)
Target: wasm32-unknown-emscripten
Thread model: posix

Consider this example:

/* test.cc */
#include <stdio.h>
#include <limits.h>
#include <unistd.h>

int main(int argc, char** argv) {
  char buf[PATH_MAX + 1];
  ssize_t len = readlink("/proc/self/exe", buf, sizeof(buf));
  if (len > PATH_MAX) {
    return 2;
  }
  printf("self exe: %s", buf);
  fflush(stdout);
  return 0;
}

% emcc -sEXIT_RUNTIME -sNODERAWFS test.cc && node a.out.js
self exe: /home/trybka/.nvm/versions/node/v22.13.1/bin/node

vs.

% emcc -sEXIT_RUNTIME -sNODERAWFS -sWASMFS test.cc && node a.out.js
self exe: 

This is because the "root" of the VFS for NodeRawFS should be /, but when using WasmFS it is . (I assume, via

emscripten/system/lib/wasmfs/backends/noderawfs_root.cpp

Line 11 in e5c41db

return wasmfs_create_node_backend(".");

)

If I change the . to / then it works the same in both cases.

[kleisauke]

PR #24733 also includes some failing tests for this, for anyone interested in fixing it.

[sbc100]

I tried the suggested fix but it looks more complicated that I thought. Turns out wasmfs has a hardcoded idea that the current working directory and the root directory should always be the same. Obviously that is not the case for NODERAWFS but we don't currently have any way to express this.

[sbc100]

Long term I think we probably need to take a different approach to WAMFS+NODERAWFS, and completely avoid the in-memory FS in this case (like we do for the old/current NODERAWFS). @tlively @kripken WDTY?

[kripken]

@sbc100

Turns out wasmfs has a hardcoded idea that the current working directory and the root directory should always be the same.

Sorry, I don't follow this. The current working directory changes over time, while the root is fixed - how can WasmFS assume they are the same? Where do you see that assumption?

Long term I think we probably need to take a different approach to WAMFS+NODERAWFS, and completely avoid the in-memory FS in this case (like we do for the old/current NODERAWFS)

I think that makes sense.

[sbc100]

@sbc100

Turns out wasmfs has a hardcoded idea that the current working directory and the root directory should always be the same.

Sorry, I don't follow this. The current working directory changes over time, while the root is fixed - how can WasmFS assume they are the same? Where do you see that assumption?

Sorry, I meant the initial current working directory. To get NODERAWFS compatibility we would need the root to be / and the CWD to start at whatever proccess.cwd() is. Right now there is now way to set an initial CWD.

Long term I think we probably need to take a different approach to WAMFS+NODERAWFS, and completely avoid the in-memory FS in this case (like we do for the old/current NODERAWFS)

I think that makes sense.

[kripken]

Right now there is now way to set an initial CWD.

I see, thanks.

We have a few hooks in wasmfs.cpp (wasmfs_before_preload and wasmfs_create_root_dir), used e.g. for backends to install themselves at the root, and other things that must happen during WasmFS startup. Perhaps one of them can be used for this, or a new one added.

[sbc100]

Is wasmfs_before_preload available for use by the backend? (i.e. is it not something that end users can/will override?)

[kripken]

Ah, I guess that one is meant for users. wasmfs_create_root_dir is an example of one meant for backends.