Added try_exists() method to std::path::Path

[Kixunil]

This method is similar to the existing exists() method, except it
doesn't silently ignore the errors, leading to less error-prone code.

This change intentionally does NOT touch the documentation of exists()
nor recommend people to use this method while it's unstable.
Such changes are reserved for stabilization to prevent confusing people.

Apart from that it avoids conflicts with #80979.

@joshtriplett requested this PR in internals discussion

[rust-highfive]

r? @kennytm

(rust-highfive has picked a reviewer for you, use r? to override)

[ChrisDenton]

One thing I'm unclear on is how this should work on other platforms. I'm thinking in particular of Windows which does not have the same folder issue as Linux but has its own way of working. E.g. GetFileAttributes can have one of the following outcomes:

• Success: the file exists
• Access denied error: the file exists but the attributes can't be read.
• File not found error: the file does not exist
• Other error: Something very unexpected happened.

If a file does exist but reading its metadata fails, then should it return an error? Or true?

[Kixunil]

@ChrisDenton access denied is just one (obvious) example, there can be various other reasons depending on the underlying OS.

If a file does exist but reading its metadata fails

If metadata() fails with any error other than NotFound then we don't know if it exists. The only reasonable return value is thus Err. That was the entire point of previous discussion.

[ChrisDenton]

there can be various other reasons depending on the underlying OS

Yes the behaviour can be very OS specific, which is the point I was trying to make using Windows as an example. But if the idea is to return Ok(true) only if the file's metadata is readable then I do think the documentation should make that clear in some way.

[joshtriplett]

This looks good to me (with the suggested fix applied).

Thank you for working on this!

[Kixunil]

@joshtriplett I guess you meant fix that match, so I did. Happy to have helped!

Should I open the tracking issue already or wait for something first?

[Kixunil]

@ChrisDenton do I understand correctly that Windows has some other function to check if file exists and also a function to read metadata and the latter could fail even if the former doesn't?

[ChrisDenton]

@ChrisDenton do I understand correctly that Windows has some other function to check if file exists and also a function to read metadata and the latter could fail even if the former doesn't?

There are two issues I think. The first one is really about how exists is implemented in terms of fs::metadata. The Windows implementation of fs::metadata involves two or more OS calls (first to open the file and then to query metadata). Querying metadata can fail but we only need to open the file to confirm the file exists.

The second issue is that exists and try_exists can fail for files that do exist and that the error itself indicates the file does indeed exist. For example, when accessing a local file both ERROR_SHARING_VIOLATION and ERROR_ACCESS_DENIED indicate the file exists but cannot be opened. Note that, unlike Linux, opening won't be denied by permissions of the parent directory because only the file's permissions are taking into account.

I don't think we necessarily need to handle or document all possibilities but I think there could be some more robust wording about what an error can mean. Though I'm struggling with how to phrase it.

[Kixunil]

@ChrisDenton thank you for explaining!

I'd actually prefer to handle those cases properly. If the function name indicates checking of existence of a file it should be as close to reality as possible. It also avoids wordy documentation of such behavior.

[kennytm]

@rust-lang/libs if no one opposes i'll r+ after the tracking issue is filed.

[BurntSushi]

LGTM.

[Kixunil]

Took it as "yes" answer to my question about opening the tracking issue, so I opened it (#83186) and filled.

[kennytm]

@bors r+ rollup

[bors]

📌 Commit 4330268 has been approved by kennytm