How does MPV deal with opened files after they are deleted/moved?

[ruben1PvP]

I realized that if I open a file with mpv and I delete it, Windows deletes it as normal and doesn't complain about it being in use, unlike other players. Knowing that I can seek in the whole file (tried with 3 hour videos) without any issues, I don't think it's just a small buffer on a copy in RAM.

How does mpv do this then?

[ghost]

#16172 (comment)

[ruben1PvP]

I suspected using inodes was the cause, but does Windows have them in the first place?

[ruben1PvP]

Oh well, I'm going to respond myself with #16172 (reply in thread), but I'm still in doubts about how it works with Windows' kernel.

[ruben1PvP]

I actually asked around and I had the following response from another person:

Yeah, there are no inodes on NTFS. That part of the explanation was only referring to UNIXoid systems. But the general concepts of "a file with a specific name and path" and "the data in the file" still exists, so conceptually you can think similarly about it. The mentioned FILE_SHARE_DELETE is not equivalent to the UNIX behavior though because it doesn't actually free the name, it only hides it. For example if you delete a file while someone still has a handle open with FILE_SHARE_DELETE, you can't create a file under the same name until they close the file. So for a robust delete you would have to rename to file to a random name (possibly one that includes a timestamp to avoid conflicts) first. Then you can delete it and the open handle will occupy the random name, but not the target name you want to reuse.

That is unless you are on a recent Windows version (not sure exactly which one) where allegedly they made FILE_DISPOSITION_POSIX_SEMANTICS the default for some APIs. Or if you are using that API yourself to enable it on older version. But I haven't played with FILE_DISPOSITION_POSIX_SEMANTICS myself, so take that last part with a grain of salt.

When I asked him about what could be the replacement for inodes in Windows that allows for the "the general concepts of "a file with a specific name and path" and "the data in the file"" to exist, he commented that it could be the Master File Table on NTFS.

After testing it with the latest Windows 10 version, I can say that at least in this version the last paragraph (using FILE_DISPOSITION_POSIX_SEMANTICS instead of FILE_SHARE_DELETE) applies.