Mercurial > public > mercurial-scm > hg
view mercurial/pure/mpatch.py @ 26117:4dc5b51f38fe
revlog: change generaldelta delta parent heuristic
The old generaldelta heuristic was "if p1 (or p2) was closer than the last full text,
use it, otherwise use prev". This was problematic when a repo contained multiple
branches that were very different. If commits to branch A were pushed, and the
last full text was branch B, it would generate a fulltext. Then if branch B was
pushed, it would generate another fulltext. The problem is that the last
fulltext (and delta'ing against `prev` in general) has no correlation with the
contents of the incoming revision, and therefore will always have degenerate
cases.
According to the blame, that algorithm was chosen to minimize the chain length.
Since there is already code that protects against that (the delta-vs-fulltext
code), and since it has been improved since the original generaldelta algorithm
went in (2011), I believe the chain length criteria will still be preserved.
The new algorithm always diffs against p1 (or p2 if it's closer), unless the
resulting delta will fail the delta-vs-fulltext check, in which case we delta
against prev.
Some before and after stats on manifest.d size.
internal large repo
old heuristic - 2.0 GB
new heuristic - 1.2 GB
mozilla-central
old heuristic - 242 MB
new heuristic - 261 MB
The regression in mozilla central is due to the new heuristic choosing p2r as
the delta when it's closer to the tip. Switching the algorithm to always prefer
p1r brings the size back down (242 MB). This is result of the way in which
mozilla does merges and pushes, and the result could easily swing the other
direction in other repos (depending on if they merge X into Y or Y into X), but
will never be as degenerate as before.
I future patch will address the regression by introducing an optional, even more
aggressive delta heuristic which will knock the mozilla manifest size down
dramatically.
author | Durham Goode <durham@fb.com> |
---|---|
date | Sun, 30 Aug 2015 13:58:11 -0700 |
parents | 525fdb738975 |
children | 9a17576103a4 |
line wrap: on
line source
# mpatch.py - Python implementation of mpatch.c # # Copyright 2009 Matt Mackall <mpm@selenic.com> and others # # This software may be used and distributed according to the terms of the # GNU General Public License version 2 or any later version. import struct try: from cStringIO import StringIO except ImportError: from StringIO import StringIO # This attempts to apply a series of patches in time proportional to # the total size of the patches, rather than patches * len(text). This # means rather than shuffling strings around, we shuffle around # pointers to fragments with fragment lists. # # When the fragment lists get too long, we collapse them. To do this # efficiently, we do all our operations inside a buffer created by # mmap and simply use memmove. This avoids creating a bunch of large # temporary string buffers. def patches(a, bins): if not bins: return a plens = [len(x) for x in bins] pl = sum(plens) bl = len(a) + pl tl = bl + bl + pl # enough for the patches and two working texts b1, b2 = 0, bl if not tl: return a m = StringIO() def move(dest, src, count): """move count bytes from src to dest The file pointer is left at the end of dest. """ m.seek(src) buf = m.read(count) m.seek(dest) m.write(buf) # load our original text m.write(a) frags = [(len(a), b1)] # copy all the patches into our segment so we can memmove from them pos = b2 + bl m.seek(pos) for p in bins: m.write(p) def pull(dst, src, l): # pull l bytes from src while l: f = src.pop() if f[0] > l: # do we need to split? src.append((f[0] - l, f[1] + l)) dst.append((l, f[1])) return dst.append(f) l -= f[0] def collect(buf, list): start = buf for l, p in reversed(list): move(buf, p, l) buf += l return (buf - start, start) for plen in plens: # if our list gets too long, execute it if len(frags) > 128: b2, b1 = b1, b2 frags = [collect(b1, frags)] new = [] end = pos + plen last = 0 while pos < end: m.seek(pos) p1, p2, l = struct.unpack(">lll", m.read(12)) pull(new, frags, p1 - last) # what didn't change pull([], frags, p2 - p1) # what got deleted new.append((l, pos + 12)) # what got added pos += l + 12 last = p2 frags.extend(reversed(new)) # what was left at the end t = collect(b2, frags) m.seek(t[1]) return m.read(t[0]) def patchedsize(orig, delta): outlen, last, bin = 0, 0, 0 binend = len(delta) data = 12 while data <= binend: decode = delta[bin:bin + 12] start, end, length = struct.unpack(">lll", decode) if start > end: break bin = data + length data = bin + 12 outlen += start - last last = end outlen += length if bin != binend: raise ValueError("patch cannot be decoded") outlen += orig - last return outlen