Clarify shared file handler behavior of File::try_clone.

[frewsxcv]

Fixes #46578.

[rust-highfive]

r? @aidanhs

(rust_highfive has picked a reviewer for you, use r? to override)

[frewsxcv]

not a big fan of my wording here or in the previous section. lemme know if y'all have better suggestions

[abonander]

Maybe something along the lines of:

This means that both File instance shares the same cursor; reads, writes and seeks will affect both instances simultaneously.

[frewsxcv]

Thanks for the suggestion! Latest force push incorporates some of that wording

[abonander]

I just realized this could be a bad footgun for someone using a cloned handle with BufReader<File>; the buffer won't know to flush and re-read after a seek or modification. Should we warn the reader about this?

[frewsxcv]

I just realized this could be a bad footgun for someone using a cloned handle with BufReader; the buffer won't know to flush and re-read after a seek or modification. Should we warn the reader about this?

@abonander I think that might be a good idea. Any suggestions on how to word this?

[abonander]

Something like

Note: BufReader Data Corruption

Using a cloned handle with BufReader could cause data corruption when the other handle is read from, seeked or written to, as the buffer won't know that the data it contains is stale.

--

However, I think this merits some discussion as BufReader could instead remember the cursor position from its previous read and check it against the current cursor position before yielding data out of the buffer. However, this would mean a syscall on every fill_buf() invocation (?). (Also if the other handle is read or written and then seeked back to the original position then the buffer wouldn't know the difference anyway.)

[abonander]

We already have a warning about a similar footgun with BufReader.get_mut() so I don't think this requires special treatment.

[frewsxcv]

Okay so it sounds like we might want to open a separate issue for that then? What do you think?

[aidanhs]

@bors r+ rollup

This looks good to me, and it sounds like the bufreader thing may need more discussion elsewhere.

[bors]

📌 Commit d597da3 has been approved by aidanhs

[bors]

💡 This pull request was already approved, no need to approve it again.

• There's another pull request that is currently being tested, blocking this pull request: Generator bugfixes #47845

[bors]

📌 Commit d597da3 has been approved by aidanhs