push: hackish handeling of new branch head from phase move The current head checking mechanism is not expecting "head change" from phase movement. Topic allows that, changeset with a topic moving to public can create a new head. We introduce a hack to double check that no head were added at the transaction level to work around this.

push: allow pushing new topic to non-publishing server by default This improves and fix the behavior change introduced by the new "topicmap". * Topics are properly ignored when pushing to a publishing server, * pushing new topics is allowed without --force a non-publishing server, * Pushing extra heads on a topic requires --force. Create of new head on a branch by phase movement is not properly detected for now. We'll improve that part in a later changesets. There is more awful monkey patching going on. We'll have to refactor core to get rid of them.

topic: take topic in account for all branch head computation This changeset introduce a "topicmap" that is tracking not just the head of all branches, but the heads of all branch+topic pair. Including the head of the part of the branch without any topic. In practice this means that BRANCHNAME now resolve to the tipmost part for the branch without topic and impact various other logic like head checking during push and default destination for update and merge (these aspect will need adjustment in later changesets). The on-the-fly-temporary-monkey-patching process is pretty horrible, but allow to move forward without waiting on having core patched. We use 'branch:topic' as the branchmap key, this is a small and easy hack that help use a lot for (future) support of heads discovery/checking and on disc cache. I'm not sure it is worthwhile to improve this until an implementation into core. Note that this changeset change the branchmap in all cases, including during exchange, see next changeset for improved behavior. We also currently have the on-disk cache disabled because the core branchmap is lacking phase information in its cache key. This will get done in a later changesets

testedwith: declare compatibility with Mercurial 3.7

Makefile: update all-version-tests to explicitly test 3.7

hgignore: also ignore testtimes

killdaemons: discard now that it's unused

test: drop custom run-tests.py We should use the one provided by Mercurial core directly.

Makefile: rework running of tests This makes running tests easier without depending on our own copy of run-tests.py (though I'd still prefer to keep that around.)

compat: adapt to change in mercurial core The code around merge destination have been changed in core, we do some basic adaptation to make sure we don't crash. Further rework to adapt//take advantage of core change would probably be useful but this is not the point of this commit.

topic: handle merge.update function signature change I think this is backwards compatible with how it used to be called, but I didn't worry about it too much either.

Merge.

test-topic-dest: test fixes I should have put in the previous change I split a bigger change from marmoute that included some rebase support that I wanted to discuss more, but I didn't correctly commit my fixes to the tests before pushing. Oops.

topic: fix changing topics of commits that have an amend_source Includes a test for this specific case, but we should probably spend more time squinting at the interactions between the current topic wrapper function and commitctx for memctx instances.

topic: remove useless call to c.parents()

topic: fix up change logic a little and add debug logging

merge: use topic to pick default destination If the topic have multiple heads, reduce that. Otherwise, pick the branch we are on.

topic: introduce a 'ngtip' concept The concept is to be massively used in naming and default destination logic. The name is horrible so that people find a better one.

template: test that topic is accessible through template This comes free with the namespace feature but it is good to have.

test: adapt to newer mercurial

patch: import topic from patch header If the used Mercurial supports it, we import topic from patch headers.

patch: add topic to exported patch If used Mercurial supports it, we add a topic header to patches.

topic: work around stringmatcher moving during development of hg 3.6

topic: allow use of topic(.) to match the p1 topic As far as I can tell this matches the behavior of branch(.) in core hg in tht it matches p1's value rather than the wdir value.

topic: fix missing error import

amend: allow clearing topics on amend

changectx: add topic method

topics: consistently use empty string instead of None This agrees with repo.currenttopic which uses ''. A subsequent patch is going to provide a context topic method that returns the empty string.

topics: drop debugging print in rebase

topics: allow commits that only change topic (issue4725) This allows amend to change topics. It also matches the behavior of branches.

topics: add listnames hook so completion works

rewrite: use a lock and transaction as spotted by devel warnings

topics: only apply topic to commits of the working copy

rebase: teach rebase how to copy topics

topic: wrap the underlying update function rather than the command This more properly manages the updates done by things like rebase.

commit: add a topic field to the in-editor commit text

topic: only require obsolete support for --change

commit: add a --topic flag

topic: note when switching topics

summary: add topic summary hook

topics: add command summary

docs: fix format of extension docstring

constants: extract key for extra into a constant It might be prudent to prefix it, and regardless it was bugging me to have it repeated everywhere.

topic: add ability to change topic of non-public changes This is a little crude, but it gets the job done. You probably don't want to use this without evolution for now.

revset: add a topic() revset for querying topics

topic: disallow use of topics without obsolete enabled This is a prelude to allowing changing topics. The experience will be disastrous without obsolete markers, so let's just preempt any potential problems.

topic: use repo.set() where we need a changectx anyway I always forget repo.set() exists and use repo.revs(). Sigh.

testlib: enable rebase and obsolete for easier testing

Initial dumb version of topics.

serveronly: deduplicate code with the main evolve extension Now that 'serveonly' is a sub module of evolve. We can access its content in evolve. As a result we stop duplicating code all common parts are defined in the 'serveronly' module and imported by evolve when needed.

serveronly: clean up imports remove unused one and arrange them

compat: just use '_callcompressable' That API is available in all version we supports.

compat: drop bookmark compatibility layer All the version we now supports have the 'new' API.

compat: drop compat for older wireproto implementation We only support Mercurial version with these attributes.

compat: always use the strip revset from core We only support versions that defines it.

compat: drop support for ancestral pushop object Pushoperation object have a 'cgresult' attribute in all the versions we support.

compat: drop compatibility layer around update We now support only versions with the new API.

compat: drop compatibility layer for 'tryimportone' Return is now homogeneous for all the version we support.

compat: drop superfluous summary output All versions we support issues this output.

compat: drop compatibility layer around memctx The version we support all have the same API now.

compat: drop handling of ancestral 'helploader' caller We only support recent only version on Mercurial now.

compat: drop hack around a 'graft' signature change We only support recent only version of Mercurial now.

evolve: proactively detect bad version early The extensions have been providing such error to the user for some time. So we keep this explicit check (instead of some later errors). The check is update to match out new version constraint.

evolve: drop support for version older than 3.8 The support for 'hgext3rd' was introduced in 3.8. Dropping support for older version seems reasonable.