Re: [PATCH 20/20] pack-revindex.c: avoid direct revindex access in 'offset_to_pack_pos()'

From: Jeff King <peff@peff.net>
Date: Jan 12, 2021, 09:37 UTC
Message-ID: <X/1txTIxV4pYt0Xo@coredump.intra.peff.net>
In-Reply-To: <eada1ffcfafc3fb57de80626e368672cb8b22318.1610129796.git.me@ttaylorr.com>

On Fri, Jan 08, 2021 at 01:18:06PM -0500, Taylor Blau wrote:

Show 13 quoted lines

> To prepare for on-disk reverse indexes, remove a spot in
> 'offset_to_pack_pos()' that looks at the 'revindex' array in 'struct
> packed_git'.
> 
> Even though this use of the revindex pointer is within pack-revindex.c,
> this clean up is still worth doing. Since the 'revindex' pointer will be
> NULL when reading from an on-disk reverse index (instead the
> 'revindex_data' pointer will be mmaped to the 'pack-*.rev' file), this
> call-site would have to include a conditional to lookup the offset for
> position 'mi' each iteration through the search.
> 
> So instead of open-coding 'pack_pos_to_offset()', call it directly from
> within 'offset_to_pack_pos()'.

This definitely makes sense in the long run. I could take or leave it as a final patch in _this_ series (as opposed to the first patch in a subsequent series adding the rev files).

>  	do {
>  		const unsigned mi = lo + (hi - lo) / 2;
> -		if (revindex[mi].offset == ofs) {
> +		off_t got = pack_pos_to_offset(p, mi);

They're both constant-time, so performance should be the same big-O. The function has extra BUG() checks. I doubt those are measurable in practice, though.

-Peff