On Wed, Mar 01, 2017 at 01:54:26PM -0800, Junio C Hamano wrote:
> -- >8 --
> Subject: [PATCH] diff: do not short-cut CHECK_SIZE_ONLY check in
> diff_populate_filespec()
Thanks, this is well-explained, and the new comments in the code really
help.
I wondered if we should be checking would_convert_to_git() in
reuse_worktree_file(), but we already do. It's just that we may still
end up in this code-path when we're _actually_ diffing the working tree
file, not just trying to optimize.
> diff --git a/diff.c b/diff.c
> index 8c78fce49d..dc51dceb44 100644
> --- a/diff.c
> +++ b/diff.c
> @@ -2792,8 +2792,25 @@ int diff_populate_filespec(struct diff_filespec *s,
> unsigned int flags)
> s->should_free = 1;
> return 0;
> }
> - if (size_only)
> +
> + /*
> + * Even if the caller would be happy with getting
> + * only the size, we cannot return early at this
> + * point if the path requires us to run the content
> + * conversion.
> + */
> + if (!would_convert_to_git(s->path) && size_only)
> return 0;
The would_convert_to_git() function is a little expensive (it may have
to do an attribute lookup). It may be worth swapping the two halves of
the conditional here to get the short-circuit.
It may not matter much in practice, though, because in the !size_only
case we'd make the same query lower a few lines later (and in theory
expensive bits of the attr lookup are cached).
> +
> + /*
> + * Note: this check uses xsize_t(st.st_size) that may
> + * not be the true size of the blob after it goes
> + * through convert_to_git(). This may not strictly be
> + * correct, but the whole point of big_file_threashold
s/threashold/threshold/
> + * and is_binary check being that we want to avoid
> + * opening the file and inspecting the contents, this
> + * is probably fine.
> + */
> if ((flags & CHECK_BINARY) &&
> s->size > big_file_threshold && s->is_binary == -1) {
> s->is_binary = 1;
I'm trying to think how this "not strictly correct" could bite us. For
line-ending conversion, I'd say that the before/after are going to be
approximately the same size. But what about something like LFS? If I
have a 600MB file that convert_to_git() filters into a short LFS
pointer, I think this changes the behavior. Before, we would diff the
pointer file, but now we'll get "binary file changed".
I wonder if we should take the opposite approach, and ignore
big_file_threshold for converted files. One assumes that such gigantic
files are binary, and therefore do not have line endings to convert. And
any filtering has a reasonable chance of condensing them to something
much smaller.
I dunno. I'm sure somebody has some horrific 500MB-filtering example
that can prove me wrong.
-Peff
