mail: document behavior of Python 3 test-notify.t (and possibly other tests) are failing on Python 3 because email.message.Message is now aware of encodings and attempts to roundtrip values with the specified message encoding. Python 2 doesn't perform this roundtripping. We have tests with non-ascii data being serialized to a message that claims to use ascii encoding. I /think/ Mercurial's behavior may be buggy here. But I'm not sure. I'm documenting the behavior so the next person who looks into this doesn't start from scratch like I did. Differential Revision: https://phab.mercurial-scm.org/D5714

notify: be more defensive aboute None values encoding.strtolocal is the identity function on Python 2 but an actual string manipulation routine on Python 3. In some cases, we were passing None, which caused Python 3 to barf. Let's change the code to react properly when the value is None. Differential Revision: https://phab.mercurial-scm.org/D5713

wireprotov2server: use our JSON encoder Python's json module doesn't like to encode bytes instances. This makes this code difficult to work with Python 3. We simply swap in Mercurial's JSON encoder to work around it. Differential Revision: https://phab.mercurial-scm.org/D5712

tests: add optional setsockopt() lines for Python 3 Differential Revision: https://phab.mercurial-scm.org/D5711

diffstat: support filenames with whitespaces on renames This is a follow-up patch to D5628. `line.split()` cannot get filenames with whitespaces as mentioned by @yuja. This patch replaces `split()` method with `slice`. Corresponding tests were also added. Differential Revision: https://phab.mercurial-scm.org/D5709

largefiles: avoid walking full manifest When using treemanifest, diffs between two manifests can often be much cheaper than iterating the full manifests (because common subtrees are skipped). Differential Revision: https://phab.mercurial-scm.org/D5644

cleanup: use p1() instead of parents() when we only need the first parent Differential Revision: https://phab.mercurial-scm.org/D5708

cleanup: use repo['.'] instead of repo[None].p1() Differential Revision: https://phab.mercurial-scm.org/D5707

cleanup: use p1() and p2() instead of parents()[0] and parents()[1] We have had these methods on both contexts and dirstate for a long time now. Differential Revision: https://phab.mercurial-scm.org/D5706

branchmap: rename partial -> bcache The name better reflects that we have a branchcache object here, even if at *some point* it is still being constructed. Differential Revision: https://phab.mercurial-scm.org/D5290

context: delete mistaken comment about return value of renamed() The comment seems to think that the return value of renamed() is a tuple of source pathsx in the parents, but it's actually a pair of (rename source path, file nodeid). Differential Revision: https://phab.mercurial-scm.org/D5639

copies: get working copy parents from wctx, not dirstate, to make in-mem work Before this patch, `run-tests.py test-rebase-conflicts.t --extra-config-opt rebase.experimental.inmemory=1` would have some lines that say "searching for copies back to rev 3" changed to "... rev 1". I don't know of a better way of testing this patch than doing that manually. There are many other differences, so we're not ready to use #testcases syntax on the entire test-rebase-conflicts.t yet. Differential Revision: https://phab.mercurial-scm.org/D5596

copies: pass contexts into _findlimit() Just a little refactoring to make the next patch simpler. Differential Revision: https://phab.mercurial-scm.org/D5595

copies: consider nullrev a common ancestor I've seen many bugs in the git codebase that were caused by it not having a null revision and being forced to treat root commits differently. Mercurial has a null revision and I think it's generally a bug to treat it differently from other commits in graph algorithms. This effectively undoes 83cfa1baf8ad (copies: don't report copies with unrelated branch, 2010-01-01). The test cases that that commit added still passes. I suspect some other fix after that commit made it unnecessary. Differential Revision: https://phab.mercurial-scm.org/D5594

unshare: use context manager for locks Differential Revision: https://phab.mercurial-scm.org/D5695

py3: new tests from the ratchet Thanks to indygreg for doing the work on these! Differential Revision: https://phab.mercurial-scm.org/D5705

dagop: check if stopdepth is greater than or equal to maxlogdepth Might prevent off-by-one errors.

revset: move dagop.maxlogdepth usage to generationsrel() Not all subscript operations would want (maxlogdepth - 1) in place of None.

test-revset: check that #generations[1:-1] doesn't result in any errors

py3: fix a bytes/str mingling in test-install.t

tests: conditionalize Windows difference in test-dispatch.t The output matches the comment added 6 years ago in ef60083b5536, so I'm assuming that difference was the reason for the over-globbing removed in 93a7539a546e. There's still a bit of py3 instability here, but I assume those are just more Windows issues: @@ -95,7 +95,7 @@ [255] $ hg log -b --cwd=inexistent default - abort: $ENOENT$: 'inexistent' + abort: $ENOENT$: inexistent [255] $ hg log -b '--config=ui.traceback=yes' 2>&1 | grep '^Traceback' @@ -189,7 +189,7 @@ $ hg -R a archive '' abort: $ENOENT$: '' (no-windows !) - abort: $ENOTDIR$: '' (windows !) + abort: $ENOTDIR$ [255] #if no-outer-repo

util: cast memoryview to bytes Python 3 uses readinto() instead of read() in places. And taking a slice of the buffer passed to readinto() will produce a memoryview. _writedata() then gets confused when testing for `b'\n' in data` because memoryview is an iterable over ints instead of 1 character bytes. We work around by casting a memoryview to bytes. Differential Revision: https://phab.mercurial-scm.org/D5704

tests: add b'' prefixes to ui.configbool() call Otherwise the call fails due to using str on Python 3. # skip-blame: just b'' prefixes Differential Revision: https://phab.mercurial-scm.org/D5703

commands: check for modheads being None Python 2 allows the > operator to be used with a None and an int. Python 3 does not. So we need to ensure the value isn't None before comparing with >. Differential Revision: https://phab.mercurial-scm.org/D5702

tests: add b'' when testing for tls1.2 The dict keys are bytes. The test was always failing due to looking for a str key. skip-blame: just b'' prefix Differential Revision: https://phab.mercurial-scm.org/D5701

wireprotov2peer: make "received frame(...)" messages debug level We don't need to show these during --verbose, it's more of a debugging thing. Differential Revision: https://phab.mercurial-scm.org/D5700

tests: add optional setsockopt() output on Python 3 Tests still don't pass. But this gets us a little closer. Differential Revision: https://phab.mercurial-scm.org/D5699

wireprotov2peer: rewrite character traversal to use slices Otherwise on Python 3 we iterate over integers instead of a bytes instance and the comparison fails. Differential Revision: https://phab.mercurial-scm.org/D5698