feat: allow vfs implementations to specify m/b/a-times in file metadata

[kartonrad]

This small change adds the file modification birth and access times to the VfsMetadata struct.

These times are encapsulated in Option<>, as they are not guaranteed to be available even in std::fs::metadata.

I have implemented the metadata timestamps for PhysicalFS, though it may be possible to support them for the in memory database as well.

Motivation

I am planning on writing a minimal vfs implementation for MTS (Media Transfer Protocol).
I don't know yet if this is going to work, especially since there are few crates for MTS - but maybe i can make it work.

In the final product i will need access to the mtime field.

[kartonrad]

Nooo i forgot to run cargo fmt
Maybe tomorrow

[manuel-woelker]

Hi, thanks for the PR! Definitely useful!
I think adding some unit tests for the time fields is probably a good idea, even though the implementation is fairly simple.
Having support for timestamp information in MemoryFS would be quite neat as well!

[kartonrad]

I think adding some unit tests for the time fields is probably a good idea, even though the implementation is fairly simple.

Good Idea! I don't know if we should necessarily expect ModifyTimes to update when we write to a file - maybe that isnt guaranteed on all operating systems...
Do you have an idea of how to test that?

Having support for timestamp information in MemoryFS would be quite neat as well!

I just looked into it, and realized the trait definition immutably borrows the Filesystem
I could use a ref-cel i guess - what do you think

[manuel-woelker]

Build looks good now 👍

Yeah, a lot of systems also mount with noatime for performance reasons.
I think the best approach might be to set the times manually via the "normal" std::fs mechanisms (on systems that support it) and then verify that the VFS abstraction returns the correct values.

It might make sense to have a way to modify these times as well, that would certainly make testing easier.

Regarding the MemoryFS implementation, I think this should work without refcell. AsyncMemoryFile would need those additional fields to track the values, and then copy them out in metadata(). But it also might make sense to tackle that in a separate PR.

[kartonrad]

The diff has grown 😯

I've added your suggestions to the non-async version
Namely:

• The optional set_creation/modification/access_time functions (path.rs and filesystem.rs)
  • should these remain optional? or do we force downstream libraries to think about this issue
• Implementations for physical and memory, as well as redirections in altroot and overlay fs
  • i used the std::fs::File::set_times() Method, that was stabilized in Rust Version 1.75.0
    do we update the 4th pipeline (1.63.0), or do we replace my std call with an external crate?
• Tests in the test_vfs!() macro, that only run fully if the call to set_..._time doesnt return VfsErrorKind::NotSupported
  • do these tests meet your standard?

[kartonrad]

Regarding the Pipeline: MSRV would now be 1.75.0

[kartonrad]

i could certaily do the work to implement this in async too sometime...

[kartonrad]

I have added the neccesary tests to async_vfs
I have added the needed methods in path.rs, and redirected the calls in overlay and altroot

However i have not been able to find a function in async_std that works like the new std::fs::File::set_times
And i have not implemented the behaviour for async memory.rs

So the tests right now are all passing, because they recieve VfsErrorKind::NotSupported

[manuel-woelker]

Thanks for the updates! I will take a closer look later this week.

[manuel-woelker]

Looking good, some small suggestions.

The merge conflicts with master only touch the imports and should be fairly simple to resolve.

If you have any questions, let me know!

[kartonrad]

Feel free to resolve the threads that i have fixed.

[manuel-woelker]

Merged! Thanks again for the great work, especially getting all the finer details sorted out 👍