Design question: why support just dirs and not files?

[oconnor663]

I'm curious about the reasons for focusing specifically on temp dirs and not temp files. Are there security benefits to always using a dir? Or simplicity/portability benefits? Asking out of historical interest more than anything else. (Apologies if this is an abuse of issues :) )

[alexcrichton]

At this point it's mostly just historical and not intentional. Long ago we had TempDir in the standard library, then the compiler, and then this external crate. I don't think this has been actively developed since its inception really, and a PR to implement temp files would be most welcome!

[oconnor663]

Ooo good to know. I bet a lot of @Stebalien's work on https://stebalien.github.io/tempfile/tempfile could be imported. (There are a bunch of places in those docs where it says "SECURITY WARNING" though, will want to think carefully about those one way or another.)

Edit: relevant context https://www.reddit.com/r/rust/comments/32n864/tempfile_temporary_file_library/cqd3ic0/?context=3

[alexcrichton]

Sounds plausible to me!

[opilar]

If we want to merge tempdir and tempfile would it be better to call it temp crate?

[KodrAus]

Is this something you guys would like to do before a 1.0 release? Or do you think it's sufficient to have an example somewhere using both tempdir and tempfile?

cc: @alexcrichton

[oconnor663]

It seems like most other languages I've used handle both in one library? New users might find it a little surprising to split them?

https://docs.python.org/3.6/library/tempfile.html
https://golang.org/pkg/io/ioutil/

[aturon]

cc me

[Stebalien]

SECURITY WARNING

Basically, relying on named temporary files in the presence of temporary file cleaners is tricky. However, to be honest, I could probably soften those warnings.

If we want to merge tempdir and tempfile would it be better to call it temp crate?

I'd also be happy to give up the tempfile crate name for the greater good. Alternatively, someone could write an fsutil crate for all these miscellaneous filesystem related tools.

[KodrAus]

I can think of a few paths forward:

1. Do nothing
2. Deprecate tempdir or tempfile and merge them into a single crate (in this case I think it's worth pointing out that tempfile is stable and tempdir isn't, but tempdir has more dependent crates)
3. Add some docs/examples to tempdir and tempfile but keep the crates separate
4. Keep the crates separate, but make another crate, temp or fsutil, or whatever that re-exports or wraps bits of tempdir and tempfile, maybe dealing with issues like conflicting destructors

I guess which option makes the most sense depends on how important this is, and whether we're happy to defer stabilising tempdir until there's a crate that offers both temporary files and temporary directories.

Maybe we should cast the net wider and see if other Rustacean's expect temporary file and directory handling in a single crate?

What do you all think?

[alexcrichton]

I'd be totally down for merging the two crates somehow if desired, it's sort of what I'd expect! I think tempfiles have the nice property that they can almost always have guaranteed cleanup on process exit, whereas temporary directories not so much :(

[KodrAus]

Having the one crate with a consistent API for temporary directories and files does sound nice.

I'm just not sure how much we should churn users of tempdir or tempfile. @Stebalien I don't think anyone's actually properly asked you what you think and whether you'd be on board with merging these crates in some way.

[Stebalien]

whether you'd be on board with merging these crates in some way.

Completely on board.

However, I'd really like a directory abstraction that allows one to securely create files/directories under a directory (guaranteeing that the files are actually created under the given directory) and have TempDir dereference to said directory abstraction. For example, on Linux, Directory would keep a file descriptor open for the directory and use openat/mkdirat to create files under it. On macos, one can use per-thread current working directories (😈) to achieve the same effect (at the cost of two additional syscalls...).

[KodrAus]

It looks like this is something folks are pretty much on-board with, but it's a pretty significant undertaking. Would a good next step be a pre-RFC to work out specifics like crate name and API?

[alexcrichton]

I wouldn't necessarily think this needs an RFC, but it probably couldn't hurt if you're feeling up to the task!

[KodrAus]

Sure, I'll get the ball rolling on this 👍

[KodrAus]

We've now got the tempdir API merged into tempfile 🎉

I'll go ahead and close this now, and we can continue tracking the progress of a new tempfile release including temporary files and directories in that repository.