fs.copyFile - add option to keep timestamps

[metabench]

fs.copyFile - It would be useful to have options to keep date created and date modified timestamps.

Possibly this would be beyond trivial to implement, but in my opinion it would be an excellent extension to node's cross-platform system capabilities.

[tniessen]

I like the idea, many libraries / runtimes already have such options. There are some things to consider though: If I remember this correctly, Unix inodes support change, modification and access timestamps, so there is no guarantee that creation timestamps exist on Unix filesystems, whereas Windows uses creation, modification and access timestamps.

If we decide to support such an option, this will need to be implemented upstream (in libuv).

[bnoordhuis]

Libuv in general cannot do better than call futimes() afterwards. That's something node.js could do so there is no reason for it to live in libuv. I don't think there are platform-specific APIs that would justify that.

Of note is that futimes() won't preserve the change time or birth time.

[bnoordhuis]

It's been a month with no action, and in light of my previous comment, I'm closing this out. Reopen if you feel this is worth pursuing (but please motivate why.)

[System-Arch]

@bnoordhuis @metabench

I am not sure how to reopen this issue, but I would like to advocate for it being re-opened. At present the behavior on Linux differs from that on Windows and Mac in an unexpected manner. While all three platforms preserve file mode metadata, only Windows and Mac preserve the modification time. This discrepancy could be rectified by having node/deps/uv/src/unix/fs.c:uv__fs_copyfile() call futimes() similar to the way it calls fchmod(). If the existing behavior is deemed desirable under some circumstances, an additional flag could be added, e.g., COPYFILE_STAT (following the existing Mac-inspired paradigm of COPYFILE_EXCL) to control this behavior.

As for motivation, currently one must add as explicit call to fs.utimes() in their application code to achieve platform parity, and one must test their code on all three platforms to discover this undocumented variation in behavior.

[bnoordhuis]

@System-Arch I'm not sure if that is a change we'd accept it libuv because of the race window between the copy and the timestamp update.

Any way you slice it, there will be a window where another process can manipulate the file and cause the timestamps to be off. Better to be consistently wrong all the time than flaky wrong some of the time.

However, if you want to pursue this, please open an issue or pull request over at libuv/libuv.

[MylesBorins]

It looks like not having accurate time stamps on fs.copy has caused fs-extra to ship broken code... seems like expectations are not being met.

@bnoordhuis would you say this is worth exploring?

[bnoordhuis]

The race condition is intrinsic, not much to explore there. Likewise the birth time issue since you can't change that after the fact.

If someone still thinks it's an excellent idea, he or she should open a pull request and we'll see where the discussion leads.

[BYK]

@bnoordhuis this also hurt us at Yarn. Do you think we can at least add something to the docs regarding this in the meantime? (not sure if this is going to be addressed but it would have been great to know about it at before we got burnt :))

If you agree with the docs, I'll submit a PR for that.

[cjihrig]

@BYK I think a docs update calling out this corner case would be good.