panic on mismatched size? about memmap2-rs

[zkat]

ah interesting. I guess it's a bug on my end. I'm already working on fixing it. Thanks so much, and sorry for the bother!

[adamreichold]

@zkat Could you elaborate on how mmap is used by cacache? Looking at the linked issue, the Write impl looks suspicious, i.e.

mmap.copy_from_slice(buf);

can only work if the mapped region has exactly the length of buf. Shouldn't there be some kind of cursor involved when implementing Write?

[zkat]

from reading the docs, it looks like it? I'm just surprised it never panicked like this before. 🤔

[adamreichold]

from reading the docs, it looks like it? I'm just surprised it never panicked like this before. thinking

One thing I could image is that the mmap call in

let mmap = if let Some(size) = size {
    if size <= MAX_MMAP_SIZE {
        unsafe { MmapMut::map_mut(tmpfile.as_file()).ok() }
    } else {
        None
    }
} else {
    None
};

was silently failing in the past (i.e. mmap == None even though the file size was below MAX_MMAP_SIZE) and updating the dependency exposed the code path which was never taken before?

Alternatively, writes did happen in a single call in the past and so the lengths did match but that happened "by accident" in the above layers and some other change made this happen in smaller chunks now?