evolve: use registrar.revsetpredicate to register revset predicate functions Now, using registrar.revsetpredicate of Mercurial directly in evolve extension should be safe enough. because it has been available since Mercurial 3.8, and minimum Mercurial version for evolve extension is 3.8, too.

topic: use registrar.revsetpredicate to register revset predicate functions Now, using registrar.revsetpredicate of Mercurial directly in topic extension should be safe enough. because it has been available since Mercurial 3.8, and minimum Mercurial version for topic extension is 4.0. This patch also removes modsetup() in topic/revset.py, because it is useless. BTW, this patch fixes an issue of "hg help revsets" output that custom revset predicates are not displayed as same as others, because they are not "inline literal" in reST syntax (``NAME``).

doc: make paragraphs before example code end with "::" for reST syntax This prevents lines in an example code from being treated as a part of normal paragraph. For issue example, lines are folded into one paragraph.

topic: centralize compatibility logic between hg versions into compat module This patch can delay loading obsutil and obsolete modules until they are actually used at "hg topics" or so, if demandimport of Mercurial is enabled.

readme: mention the improvement to stack in the readme

test: issue a clarifying graph print in tests/test-stack-branch.t

stack: properly order stack when gaps existing inside it We transitively search for the next "stack" ancestors, of changeset in the stack not based on other revision of the stack. This should help having a consistent display when topic are interleaved.

stack: display both parent with displaying merge This will help understanding merge in the stack. More test involving merge are adding in the next changeset.

tests: add tests for showing stack when merge is involved This patch adds tests of showing stack when merges are involved. This will help us in changing the behaviour and realising that in later patches.

stack: move data computation on the object We gather multiple data about the stack on the object. Let us move this on the object and keep it cached there.

stack: replace 'getstack' with direct call to 'stack' The dedicated function do not add any value. We can update all the callers.

stack: introduce a rich stack object The logic around stack object become more and more complex. We great and object to gather and abstract it.

tests: add tests for `hg stack` showing surprising behavior with gaps The test will help us to analyse in next commit if the things are fixed or not.

topic: check availability of obsutil.getmarkers() for portability Before this patch, topic extension causes unintentional failure with Mercurial earlier than 4.3, because obsutil.getmarkers() has been available since Mercurial 4.3 (tests for topic on mercurial-4.* branches fail, too). This breaks "minimumhgversion = '4.0'" declaration of topic extension. This patch fixes this issue in a straightforward way for simplicity on stable branch. I'm planning to centralize such portability logic in topic extension into topic/compat.py or so on default branch for efficiency at runtime, like as evolve/compat.py.

tests: add more explicit test about empty topic from publishing

convertbookmark: advertise the new command in the readme The command is in a good enough shape now.

convertbookmark: properly convert stacked bookmarks We ignores pure descendant when looking for revision to convert. This adds support for cases were a linear stack contains multiple bookmark (or un-bookmarked descendants exists). In this case, some changeset are left orphan, but we'll consider the issue as minor for now.

convertbookmark: perform all actions at the end This will make the whole operation more deterministic and robust.

converbookmark: split target computation from actual update This will allow for further refactoring helping consistency.

convertbookmark: extract the revset into a module level constant This will make it easier to read and modify the query.

convertbookmark: properly exclude public changeset Public changeset are immutable and cannot have a topic anyway.

test: add a new root changeset to all debugconvertbookmark test case We'll make this root changeset public to check the debugconvertbookmark logic in that regards. It is added in its own changeset since it massively impact all output.

tests: import section in 'test-topic-debugcb.t' We now have two levels of titles to clarify some aspect of the tests.

test: avoid using debugdrawdag in 'test-topic-debugcb.t' The 'debugdrawdag' utility adds local tags that prevent hidden changeset to get hidden. We will do without that extension for now.

tests: add more tests for debugconvertbookmark

convertbookmark: add logic to skip revisions with multiple bookmarks When a user tries to converts a bookmark to topic which is on a changeset with more bookmarks we skip that.

topics: add a new debugconvertbookmark command to convert bookmarks to topics This new command, aliased to debugcb converts bookmarks to topics. This command either accept name of a bookmark using flag '-b' or '--all'. If a bookmark is passed using '-b', the whole stack defined by that bookmark is converted to topics. If '--all' is passed, the above thing is repeated for every bookmark in the repository. If the revset which we are using unable to decides revisions of the stack and return an expty set, in that cases we are not deleting the bookmark as we are not writing any topics of the same name as that of bookmark. Also currently if a changeset has two bookmarks, than there is buggy behaviour. So these are known lackings in the current implementation which will be fixed in upcoming patches. This also adds tests for the same.

topic: support --rev argument and properly process then as user input Revisions inputed by the users should be processed using `scmutil.revrange`. This will take alias into account and allow extensions to monitor the user inputs. As a side effect, the '_changetopics' function takes revisions as argument. make the API clearer.

topic: move a status message in the right scope The finally get runs on failure so we should not advertise rewrite topic in the case. In addition the 'rewrote' variable can be undefined in the finally.

obsfate: use effect flag information for computing a better verb