dirfs.OpenFile can return ENAMETOOLONG even if that is not defined FS.OpenFile in the interface documentation

[karelbilek]

This is similar to #2101 but it's more general and it basically contains all other errnos.

Basically, dirFS can return a lot more other errnos - not just the few defined here

https://github.com/tetratelabs/wazero/blob/ba569623fcb7b32bb8fc64dadc842672befd6819/imports/wasi_snapshot_preview1/fs.go#L1534...L1540

when the os returns them as a syscall error, and it's in the list here

wazero/experimental/sys/syscall_errno.go

Line 7 in cb3dc29

func syscallToErrno(err error) (Errno, bool) {

A trivial repro on macOS (need to run from somewhere where you have access to internal/sysfs/dirfs)

rootFS := dirfs.DirFS("/")
res, err := rootFS.OpenFile("tmp/"+strings.Repeat("a", 400), sys.O_CREAT, 0)
fmt.Println(res, err)

This should not be possible according to FS.OpenFile documentation:

https://github.com/tetratelabs/wazero/blob/ba569623fcb7b32bb8fc64dadc842672befd6819/experimental/sys/fs.go#L34...L39

Again I hit this when I implement my own FS implementation and I am not sure if I should adhere to the doc with the errnos, or just return whatever.

(Sorry if this is duplicate of #2101 and should not be a different issue)

[karelbilek]

The same repro should work on Linux, max filename length is 255.

[evacchi]

I'd say it is not necessary to open an issue for each new error that is not documented :D let's keep it tidy, as I think I have mentioned in #2076 this feature is still experimental so you might find more of these idiosyncrasies. We can continue on this or #2101, as essentially the issue boils down to limited documentation and/or unspecified behavior.

Do follow up with your questions and feedback, all it's welcome.

[karelbilek]

sorry:) yeah ok will close this and move the comment there

…

On Thu 29. 2. 2024 at 18:46, Edoardo Vacchi ***@***.***> wrote: I'd say it is not necessary to open an issue for each new error that is not documented :D let's keep it tidy, as I think I have mentioned in #2076 <#2076> this feature is still experimental so you might find more of these idiosyncrasies. We can continue on this or #2101 <#2101>, as essentially the issue boils down to limited documentation and/or unspecified behavior. Do follow up with your questions and feedback, all it's welcome. — Reply to this email directly, view it on GitHub <#2104 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAZT4IHLXBFSZLJ5N5UND3YV5URFAVCNFSM6AAAAABD77DUJSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSNZRGY2DSMRTGU> . You are receiving this because you authored the thread.Message ID: ***@***.***>