Thanks for catching this oversight. Shouldn't we be zero'ing the nanoseconds field in this case? eg:

#if defined(GIT_USE_NSEC)
    iter->entry.ctime.nanoseconds = entry->st.st_ctime_nsec;
    iter->entry.mtime.nanoseconds = entry->st.st_mtime_nsec;
#else
    iter->entry.ctime.nanoseconds = 0;
    iter->entry.mtime.nanoseconds = 0;
#endif

@ethomson I guess. I just followed the implementation on the original PR though, so that would also require updating. :)

Sorry, which original change are you looking at?

Generally we zero structs, but I think that in this particular case, we're iterating an index entry from an on-disk index that we've loaded. That on-disk index may have been written with a nanosecond supporting git client. If so, the nanoseconds field in the entry would be nonzero. If that's true, then we need to clear it, and not persist it...

This one.

git_index_entry__from_stat is called on new, zero'd out index entries that we're creating. They're generated in index_entry_create which callocs its memory.

The index iterators are very different beasts - we don't create new index entries with them, that we know will have zero'd nanosecond entries - we're refreshing index entries that we've loaded from the index on disk. They may have non-zero nanosecond entries that we need to clear.

Alrighty. Just updated the patch.

Thanks!