uncommit: warn about new instability

uncommit: add --all option Convenient when combined with --exclude.

uncommit: handle bookmarks Extract the bookmarks handling code from amend and share it with uncommit.

evolve: add uncommit command

obsolete: enable rebase --keep again We still never strip anything. But using --keep on the command line dos not lay marker.

[obsolete] fix revset error message

evolve: warn when latecomer changeset are created

obsolete: detect late comer during push

obsolete: add latecomer computation and display

obsolete: add precursors and successors revset.

update revset vocabulary to use precursors, successors

obsolete: remove unused variable in rebase wrapping

evolve: tidy up a bit - long lines - extra blank lines - commands options descriptions

obsolete: make collapse create less markers Collapsing A and B into C was registering: A -> C B -> C A -> B At this point, the last relation brings nothing and is removed for the sake of simplicity: A -> C B -> C test-evolve.t is slightly modified as one of the relations let stabilize compute a target outside of the parent revision tree. This is a stabilize bug, not trivial to fix, so let's tweak the test for now.

obsolete: remove unused code from reposetup()

obsolete: suspended changeset with secret only children should be secret too test-obsolete-push.t: exhibit outgoing bug with suspended changeset in this case

obsolete: add 'obsolete' keyword The obsolete keyword displays the node status among 'stable', 'unstable', 'suspended', 'extinct'.

obsolete: unify collapsed revisions markers handling When collapsing A, B and C into D, amend was registering: A -> D B -> D A -> B C -> D A -> C while the rebase wrapper was doing: A -> D B -> D C -> D At this point, I have no argument to favor one or another or even a new one like: A -> B B -> C C -> D so I am aligning the rebase implementation on the older amend one. At least we can now change them all at once.

rebase: improve invalid revisions filtering - Filter rebasestate nullmerge and unrebased revision before considering updating obsolete. This simplifies the handling of obsolete and make it ignore unrebased revision rather than extinguishing them. - Apply obsolete markers only if rebasestate is not empty, testing rebasetarget was irrelevant.

test-evolve: extend output and exhibit an amend bug The extended section shows the parent revision stabilizing itself on another revision, without --any. In theory, only descendants of the parent predecessors or of its descendants should be stabilized, and be stabilized in parent subtree. Unfortunately, stabilize implementation correctly picks the changesets to stabilize but not their target. This problem is aggravated by amend registering too many obsolete markers which prevents the stabilize selection heuristic to prune several changesets.

evolve: remove --change . == --change .^ crazyness

evolve: use explicit phase constants

[doc] update readme

[doc] have people update on stable after cloning

merge with fix

merge with fix

[qsync] fix: 'newerversion' expect node id, not hexid

merge back with stable tag

Added tag 0.2.0 for changeset d3f20770b86a

Merge with Patrick improvement

obsolete: do not obsolete rebase --detach nullmerge revs Rebase state contains the changesets to rebase as well as 'nullmerge' entries used drive the merge strategy. These nullmerge were not rebased and should be ignored, and certainly not be marked obsolete.

evolve: fix relocate() with removed revision relocate() was not handling the case where the rebased revision was discarded and was raising when trying to update the phase of None.

test-obsolete-rebase.t: test --abort and --continue

obsolete: handle rebase --collapse Recording obsolete in concludenode() did not work for --collapse because only the first revision was passed to the call. The new strategy is to track the rebase state in defineparents() and concludenode() and to create markers only after a successful non-abort call. In theory, this should also fix --continue/--abort cases. The change in test-stabilize-order.t comes from concludenode() no longer creating obsolete marker. stabilize command was actually duplicating markers, once in concludenode(), once explicitely.

test-stabilize-order.t: trace obsolete markers Will prevent breakage in rebase.concludenode() overriding.

test-obsolete-rebase.t: move --keep test from test-obsolete.t

obsolete: handle rebase removing empty changesets The new revision identifier returned by concludenode() was None and was not correctly translated to nullid.

obsolete: add or fix debug commands help

obsolete: do not duplicate command help

obsolete: use new-style command registration

obsolete: make rebase abort with --keep obsolete modifies rebase to mark rebased changesets obsolete instead of stripping them, which makes --keep irrelevant.

obsolete: fix rebase wrapping ui and repo arguments were swapped which made --keep calls traceback.

obsolete: add debugsuccessors command It dumps obsolete changesets and their successors.

evolve: add alias for kill and stabilize

merge with 0.1.0 tag

Added tag 0.1.0 for changeset 6c6bb7a23bb5

stabilize: improve unstable selection heuristic Without argument, stabilize was picking the first in: "unstable() and ((suspended() or obsancestors(::.))::)" which usually returned the "oldest" unstable revision in parent predecessors descendants. This revision is interesting because it usually gives "soft" merges but rebasing it left the working directory on a remote branch, which was very confusing. The new heuristic picks an unstable changeset which can be rebased on top of the parent revision, or on top of one of its descendants (selected in revision order). This has the advantage of selecting a revision which can be rebased on the current subtree, and leave the working directory in a more convenient location.

amend: remove useless repo.changelog alias

README: some more changes

merge with Arne changes

amend: do not traceback on no-ops If rewrite() generated changeset happened to be an existing one, the call would traceback when trying to obsolete the changeset with itself. Instead, leave gracefully, marking any intermediate changeset extinct.

amend: do not rely on revsingle() default for --change

amend: drop --branches, pick it from working directory The amended changeset branch was picked either from --branch or from the first parent, but the actual working directory branch was ignored. The behaviour is changed so the amended revision branch is picked from the intermediate revision, which branch comes from the usual working directory rules. --branch is thus replaced by "hg branch".

.hgignore: add .rej/.err/.orig and emacs tmp files

added “pull here is OK, too” to the README.

evolve: fix stabilize --any short option It was displayed as --A instead of -A.

tutorial: zsh fix

Fix amend and relocate who picked phases data after extinct wher made secret

zsh does not like "`" in << EOF

readlinks does not do what we wnat on BSD

evolve: update rebase usage to match 2.3

obsolete: fix error in cache invalidation

[obsolete] fix new api: be more flexible on rebase's buildstate signature Introduced by `d1afbf03e69a@mercurial <http://selenic.com/repo/hg/rev/d1afbf03e69a>`_

[obsolete] fix new api: use the new phase cache api Introduced by `b6081c2c4647@mercurial <http://selenic.com/repo/hg/rev/b6081c2c4647>`_

[qsync] abort properly of a patches repository does not exists

[doc] Specify that tuto hg configuration is used for testing.

qsync: fix mq commit message generation A CR was missing when both applied and ready to review patches are generated during a single qsync.

[doc] changes location of repo We now have a "stable" repo on bitbucket and a dev repo on logilab hg lab.

change version definition

doc: idea about OnDisk Storage

doc: add the RoadMap

[doc] add Question and Answer about Obsolet Marker Implementation

fix typo in file name

[doc] change doc generation directory

[doc] add good practice section

[doc] fix unstable code block

[readme] fix stupid typo

more read me update

update README

[doc] fix tutorial warning

[doc] add `.t` to `.rst` converteur Update make file accordingly.

[doc] fix bad include

[doc] add a simple make file

docs: proof-read and review

restore unstability content

less verbose table

[doc] update tutorial text.

proof-read documentation and docstrings

doc: several update and review.

obsolete: export pretty printed json Will help unavoidable hand fixing.

accept test change (not very happy about it)

new obsolete store The new formats allows multiple "new" changeset and various meta-data. Forcing extinct to secret phase happen now on an irregular basis A command exist to migrate to the new format

enable.sh: enable rebase because evolve need it

ignore .pyc files

doc: unstability principle

qsync: support for synchronisation with applied patches from outer space and more :more: - also add some tests - kill OLDBRANCHNAME - several minor cleaning

evolve-amend: properly handle amending file renames and additions.

obsolete: do no complain when push create a new head but obsolete an old one. Do not read the code or your eyes will burn. The (+1 heads) message still appear (see mercurial issue 3394).

obsolete: refuse to push unstable changeset without -f We do not alter phase of suspended and unstable changeset anymore. But push refuse to push them without force. The extinct part of the history stay secret for simplicity shake.

qsync: add a warning file in qsync changeset.

obsolete: fix changelog reload This changelog was not properly reloaded. This leaded to bug with rollback and strip.

obsolete: do not warn about unpushed extinct changeset.

obsolete: add an extinct property on context

obsolete: to not overwrite cancopy This does not bring anything and this confuse verify.

evolve: issue the "%i new unstable changeset" in more place most command able to create unstability now issue the warning

obsolete: blindly obsolete unstability cache when changeset are added

tutorial: make it pass again More work is needed but the tutorial work as a functionnal test.

docs: fix broken alias for qseries template for description in `{desc}` not `{description}`

fix enable.sh script it was printing the mercurial version in to HGRC file

test: graphlog honor --hidden honor in 2.2

obsolete: fix rollback wrapper for 2.2-rc

2.2 compat: nodemap.get take a single argument now.

[obsolete] add supprt for futur Hg 2.2

[obsolete] simplify by using the Hg 2.1 api

qsync: display short node repr on error

FAQ: add en entry about rebasing after a pull

FAQ: typos

FAQ: small typo

stronger try except for store cache

amend: add a warning after amend added unstablility