mercurial: Changelog

mq: use dirstateguard instead of dirstate.invalidate (qrefresh) Before this patch, "mq.queue.refresh()" uses "dirstate.invalidate()" as a kind of "restore .hg/dirstate to the original status" during a failure. But it just discards changes in memory, and doesn't actually restore ".hg/dirstate". Then, it can't work as expected, if "dirstate.write()" is executed while processing. This patch uses "dirstateguard" instead of "dirstate.invalidate()" to restore ".hg/dirstate" during a failure even if "dirstate.write()" is executed before a failure. This patch also removes "beginparentchage()" and "endparentchange()", because "dirstateguard" makes them useless. This is a part of preparations to fix the issue that the recent (in memory) dirstate isn't visible to external processes (e.g. "precommit" hook).

mq: use dirstateguard instead of dirstate.invalidate (qpush) Before this patch, "mq.queue.apply()" uses "dirstate.invalidate()" as a kind of "restore .hg/dirstate to the original status" during afailure. But it just discards changes in memory, and doesn't actually restore ".hg/dirstate". Then, it can't work as expected, if "dirstate.write()" is executed while processing. This patch uses "dirstateguard" instead of "dirstate.invalidate()" to restore ".hg/dirstate" at failure even if "dirstate.write()" is executed before failure. This is a part of preparations to fix the issue that the recent (in memory) dirstate isn't visible to external processes (e.g. "precommit" hook).

tryimportone: use dirstateguard instead of beginparentchange/endparentchange To fix the issue that the recent (in memory) dirstate isn't visible to external process (e.g. "precommit" hook), a subsequent patch makes "localrepository.commit()" invoke "dirstate.write()" in it. This change will make "beginparentchange()" and "endparentchange()" on dirstate in "cmdutil.tryimportone()" meaningless, because: - "dirstate.write()" writes changed data into ".hg/dirstate", but - aborting between "beginparentchange()" and "endparentchange()" doesn't cause any restoring ".hg/dirstate" it just discards changes in memory. This patch uses "dirstateguard" instead of "beginparentchange()" and "endparentchange()" in "cmdutil.tryimportone()" to restore ".hg/dirstate" during a failure even if "dirstate.write()" is executed before a failure. This patch uses "lockmod.release(dsguard)" instead of "dsguard.release()", because processing may be aborted before assignment to "dsguard" , and the "if dsguard" examination for safety is redundant.

import: use dirstateguard instead of dirstate.invalidate Before this patch, "commands.import()" uses "dirstate.invalidate()" as a kind of "restore .hg/dirstate to the original status" during a failure. But it just discards changes in memory, and doesn't actually restore ".hg/dirstate". Then, it can't work as expected, if "dirstate.write()" is executed while processing. This patch uses "dirstateguard" instead of "dirstate.invalidate()" to restore ".hg/dirstate" at failure even if "dirstate.write()" is executed before failure. This patch also removes "beginparentchage()" and "endparentchange()", because "dirstateguard" makes them useless, too. This is a part of preparations to fix the issue that the recent (in memory) dirstate isn't visible to external process (e.g. "precommit" hook).

amend: use dirstateguard instead of dirstate.invalidate Before this patch, "cmdutil.amend()" uses "dirstate.invalidate()" as a kind of "restore .hg/dirstate to the original status" during a failure. But it just discards changes in memory, and doesn't actually restore ".hg/dirstate". Then, it can't work as expected, if "dirstate.write()" is executed while processing. This patch uses "dirstateguard" instead of "dirstate.invalidate()" to restore ".hg/dirstate" at failure even if "dirstate.write()" is executed before failure. This is a part of preparations to fix the issue that the recent (in memory) dirstate isn't visible to external process (e.g. "precommit" hook).

localrepo: use changelog.hasnode instead of self.__contains__ Before this patch, releasing the store lock implies the actions below, when the transaction is aborted: 1. "commithook()" scheduled in "localrepository.commit()" is invoked 2. "changectx.__init__()" is invoked via "self.__contains__()" 3. specified ID is examined against "repo.dirstate.p1()" 4. validation function is invoked in "dirstate.p1()" In subsequent patches, "dirstate.invalidate()" invocations for discarding changes are replaced with "dirstateguard", but discarding changes by "dirstateguard" is executed after releasing the store lock: resources are acquired in "wlock => dirstateguard => store lock" order, and are released in reverse order. This may cause that "dirstate.p1()" still refers to the changeset to be rolled-back at (4) above: pushing multiple patches by "hg qpush" is a typical case. When releasing the store lock, such changesets are: - not contained in "repo.changelog", if it is reloaded from ".hg/00changelog.i", as that file was already truncated by "transaction.abort()" - still contained in it, otherwise (this "dirty read" problem is discussed in "Transaction Plan" http://mercurial.selenic.com/wiki/TransactionPlan) Validation function shows "unknown working parent" warning in the former case, but reloading "repo.changelog" depends on the timestamp of ".hg/00changelog.i". This causes occasional test failures. In the case of scheduled "commithook()", it just wants to examine whether "node ID" of committed changeset is still valid or not. Other examinations implied in "changectx.__init__()" are meaningless. To avoid showing the "unknown working parent" warning irregularly, this patch uses "changelog.hasnode()" instead of "node in self" to examine existence of committed changeset.

cmdutil: add class to restore dirstate during unexpected failure Before this patch, after "dirstate.write()" execution, there was no way to restore dirstate to the original status before "dirstate.write()". In some code paths, "dirstate.invalidate()" is used as a kind of "restore .hg/dirstate to the original status", but it just avoids writing changes in memory out, and doesn't actually restore the ".hg/dirstate" file. To fix the issue that the recent (in memory) dirstate isn't visible to external processes (e.g. "precommit" hooks), "dirstate.write()" should be invoked before invocation of external processes. But at the same time, ".hg/dirstate" should be restored to its content before "dirstate.write()" during an unexpected failure in some cases. This patch adds the class "dirstateguard" to easily restore ".hg/dirstate" during unexpected failures. Typical usecase of it is: # (1) build dirstate up .... # (2) write dirstate out, and backup ".hg/dirstate" dsguard = dirstateguard(repo, 'scopename') try: # (3) execute somethig to do: # this may imply making some additional changes on dirstate .... # (4) unlink backed-up dirstate file at the end of dsguard scope dsguard.close() finally: # (5) if execution is aborted before "dsguard.close()", # ".hg/dirstate" is restored from the backup dsguard.release() For this kind of issue, an "extending transaction" approach (in https://titanpad.com/mercurial32-sprint) seems to not be suitable, because: - transaction nesting occurs in some cases (e.g. "shelve => rebase"), and - "dirstate" may be already modified since the beginning of OUTER transaction scope, then - dirstate should be backed up into the file other than "dirstate.journal" at the beginning of INNER transaction scope, but - such alternative backup files are useless for transaction itself, and increases complication of its implementation "transaction" and "dirstateguard" differ from each other also in "what it should do for .hg/dirstate" in cases other than success. ============== ======= ======== ============= type success fail "hg rollback" ============== ======= ======== ============= transaction keep keep restore dirstateguard keep restore (not implied) ============== ======= ======== ============= Some collaboration between transaction and dirstate will probably be introduced in the future. But this layer is needed in all cases.