realpath in stdlib.h but not in library

[binji]

I assume we won't have realpath for WASI MVP, is that correct? It seems to included in stdlib.h.

[sunfishcode]

Yeah, WASI doesn't have absolute paths, so the standard realpath doesn't fit. I filed #82 to remove the declarations.

That said, besides making paths absolute, realpath also resolves symbolic links, resolves .. and . paths, and trims redundant / characters. We could conceivably have a realpath which does those things and maybe also resolves relative paths to be rooted at ., which might be enough for some use cases.

[sbc100]

It doesn't have absolute paths at the syscall layer, but doesn't it kind of emulate them in libpreopen? I mean if the concept of PWD exists then can't realpath just use getcwd() to construct an abspath?

[sunfishcode]

We don't currently have getcwd either :-). libpreopen maps from absolute path to base + relative path, which isn't enough to do getcwd.

Should we add a getcwd? It does leak information about the host, and we can't easily virtualize or remap it because the main purpose is to get a string that you can send to another process or serialize out somewhere to be read by the outside world, so it has to reflect the actual host path name.

[pchickey]

I think we should encourage the programs that require these sort of absolute path functions to implement them totally in userland, if possible. I wouldn't want to expose the host's "real" cwd at all, but it should be possible to define these functions such that it always pretends to be under /home/wasi or something, right?

[sbc100]

No, I'm not talking about exposing the real host cwd. But we can build cwd in userspace based on the pre-opened root fd can't we? I'm surprised libpreopen doesn't do this already. Wouldn'tit be as simple as doing: base + global_cwd + relative?

I don't understand what you mean about sending the string to the outside world. Paths only make sense within a given module, right? Or at least within modules that shared the same virtualized view of the filesystem.

[binji]

Right, I'd imagine it would be similar to running in a chroot.

[sunfishcode]

The original issue here is fixed in a413650; we no longer declare realpath.

If anyone's interested in introducing a concept of a cwd, even without exposing the host cwd, we'd need new WASI support, so we should discuss this in the WASI issue tracker.