Fix zombie files in merge
# HG changeset patch
# User maf46@burn.cl.cam.ac.uk
# Node ID 57667c9b93a5a743e4629d15a0e6bd76699130c3
# Parent d2994b5298fb20f87dc1d4747635b280db3c0526
Fix zombie files in merge
Keir Fraser observed the following:
> I made a small test case that illustrates the bug in merging changesets
> with 'hg remove's in them:
>
> 1. Create a repository A containing files foo & bar.
> 2. Create clone called B.
> 3. A removes file bar, and commits this removal.
> 4. B edits file foo, and commits this edit.
>
> Now, if B:
> # hg pull ../A; hg update -m; hg commit
> Then bar remains deleted.
>
> If A:
> # hg pull ../B; hg update -m; hg commit
> Then bar is resurrected!
>
> It looks as though, when you merge across a branch, any deletions in
> your own branch are forgotten.
> ...
> Fixing this is a must, as zombie files are a real pain. :-)
Keir later patched our local copy of hg as shown below, which fixes
the problem. I've also enclosed a test which captures the test Keir
outlined...
Files deleted on a branch should not automatically reappear in a merge
Patch notes:
1. The first chunk does not change behaviour, but cleans up the code
to more closely match check of 'force' in the second chunk. I
think it makes the code clearer.
2. The second chunk fixes two bugs --
i. If we choose to keep a remotely-changed locally-deleted file,
then we need to 'get' that file. If we choose to delete it
then no action need be taken (it is already deleted in the
working manifest). Without this fix, choosing to delete would
get a Python traceback.
ii. The test for whether the file was remotely-created is
insufficient. It is only true if f is not in the common
ancestor. Otherwise the file was deleted locally, and should
remain deleted. (this is the most important fix!)
Index: hg/tests/test-merge6
===================================================================
# transaction.py - simple journalling scheme for mercurial
#
# This transaction scheme is intended to gracefully handle program
# errors and interruptions. More serious failures like system crashes
# can be recovered with an fsck-like tool. As the whole repository is
# effectively log-structured, this should amount to simply truncating
# anything that isn't referenced in the changelog.
#
# Copyright 2005 Matt Mackall <mpm@selenic.com>
#
# This software may be used and distributed according to the terms
# of the GNU General Public License, incorporated herein by reference.
import os
import util
class transaction:
def __init__(self, report, opener, journal, after = None):
self.journal = None
# abort here if the journal already exists
if os.path.exists(journal):
raise "journal already exists - run hg recover"
self.report = report
self.opener = opener
self.after = after
self.entries = []
self.map = {}
self.journal = journal
self.file = open(self.journal, "w")
def __del__(self):
if self.journal:
if self.entries: self.abort()
self.file.close()
try: os.unlink(self.journal)
except: pass
def add(self, file, offset):
if file in self.map: return
self.entries.append((file, offset))
self.map[file] = 1
# add enough data to the journal to do the truncate
self.file.write("%s\0%d\n" % (file, offset))
self.file.flush()
def close(self):
self.file.close()
self.entries = []
if self.after:
util.rename(self.journal, self.after)
else:
os.unlink(self.journal)
self.journal = None
def abort(self):
if not self.entries: return
self.report("transaction abort!\n")
for f, o in self.entries:
try:
self.opener(f, "a").truncate(o)
except:
self.report("failed to truncate %s\n" % f)
self.entries = []
self.report("rollback completed\n")
def rollback(opener, file):
for l in open(file).readlines():
f, o = l.split('\0')
opener(f, "a").truncate(int(o))
os.unlink(file)