branchcache: filter obsolete revisions sooner Since we won't do anything with the obsolete revisions, we can just ignore them sooner.

branchcache: skip entries that are topological heads in the on disk file In the majority of cases, topological heads are also branch heads. We have efficient way to get the topological heads and efficient way to retrieve their branch information. So there is little value in putting them in the branch cache file explicitly. On the contrary, writing them explicitly tend to create very large cache file that are inefficient to read and update. So the branch cache v3 format is no longer including them. This changeset focus on the format aspect and have no focus on the performance aspect. We will cover that later.

branchcache: simplify the branch rev cache test We don't need that many content dump and this gets in the way in change in access pattern (e.g. accessing revision in a different order change the order of branches in the "names" file). So we simplify this test in advance.

branchcache: store filtered hash and obsolete hash independently for V3 This will avoid the bug covered in tests/test-branches-obsolete.t when we stop storing all heads explicitly in V3.

branchcache: show the cache file content in test-branches-obsoletes.t This help to track the changes in format between v2 and v3.

branchcache: rework the `filteredhash` logic to be more generic We now have a more flexible `key_hashes` tuple. We duplicated various logic in the V2 and V3 version of the cache as the goal is to start changing the logic for V3 in the next few changesets.

filteredhash: rename the filteredhash function The new name is less ambiguous, as we are about to introduce an alternative function it seems like a good idea to have clearer name to distinct the two.

filteredhash: split the computation of revision sets The branch2's filteredhash combines the filtered revisions and the obsolete ones, this will creates issue for implicit reference to heads we want to introduce for the v3 of the branch cache format. So we isolate this logic for alternative use.

filteredhash: move the hashing in its own function This will help us to reuse this logic in variants of the hashes used for branch cache validation.

branchcache: cleanup the final key generation after update A lot of duplicated work seemed to be done, as we already update the tiprev and tipnode when needed right before. So we simplify that part to focus on the filtered hash. See inline comment for details.

branchcache: add more test for the logic around obsolescence and branch heads While working on branch-cache-v3, we noticed some ambiguity in the filtered+obsolete hash. However this was only caught by a rebase test by chance. It seems important to explicitly tests these cases.

branchcache-v3: use more explicit header line The key-value approach is clearer and gives more rooms to have the format evolve in a clear way. It also provides extension (like topic) simpler way to extend the validation scheme. This is just a small evolution, the V3 format is still a work in progress.

branchcache-v3: introduce a v3 format For now the format is the very same, however we will start changing it in future changesets.

branchcache: use an explicit class for the v2 version This prepare the introduction of an experimental v3 format version. In the process, we move the description of the format in that new class.

branchcache: add some blank line in a test This helps each section to stand out.

phases: update the phase set as we go during retract boundary Apparently iterating over the `changed_revs` dictionary is very expensive. On mozilla-try-2019-02-18, a perf::unbundle call with a 10 000 changesets bundle gives give use the following timing. e57d4b868a3e: 4.6 seconds ac1c75188440: 102.5 seconds prev-changeset: 30.0 seconds this-changeset: 4.6 seconds So, the performance regression is gone. Once again: thanks to marvelous Python!

phases: avoid a potentially costly dictionary interation in some case If we retract for the draft phase, there is not non-public item to be retracted and we can skip this part. This part is was apparently super costly thanks to Python. On mozilla-try-2019-02-18, a perf::unbundle call with a 10 000 changesets bundle gives give use the following timing. e57d4b868a3e: 4.6 seconds ac1c75188440: 102.5 seconds this-changeset: 30.0 seconds So we recovered about ⅔ of the regression, the next changeset will give us the rest back.

branching: merge stable into default

Added signature for changeset 2e6fde2ed01e

Added tag 6.7.1 for changeset 2e6fde2ed01e

relnotes: add 6.7.1

subrepo: fix normalizing paths with scheme After revision 0afe96e374a7, subrepo paths were normalized using posixpath.normpath and that resulted in ssh paths being wrongly converted from ssh://host/path to ssh:/host/path This fix applies the same logic used in urlutil.url to split the path scheme from the rest and only use posixpath.normpath to the string after scheme://

sshpeer: fix path when handling invalid url exception In 73ed1d13c0bf the code was refactored but the error handling seems to have been missed (or maybe the object shoud have implemented __bytes__)

delta-search: fix crash caused by unbound variable This code path was apparently not tested. This fixes a crash when cloning the Tryton repo.

branching: merge stable into default

Added signature for changeset c9ceb4f60256

Added tag 6.7 for changeset c9ceb4f60256

phases: avoid N² behavior in `advanceboundary` We allowed duplicated entries in the deque, which each entry could potentially insert all its ancestors. So advancing boundary for the full repository would mean each revision would walk all its ancestors, resulting in O(N²) iteration. For repository of any decent size, N² is quickly insane. We introduce a simple set to avoid this and get back to reasonable performance.

relnotes: add 6.7

admin-commands: move the chainsaw extension to the admin commands module Activating an extension is always a little bit of a chore and the long name, options and "chainsaw" bits are deterrent enough. This also allows us to help the discoverability for people looking for repo "administration" tools, with the widest semantic of "administration".

obsutil: sort metadata before comparing in geteffectflag() This is probably less important now that we dropped Python 2. We do still support Python 3.6 though, and the dictionaries aren't ordered there either (that was a big change that came with 3.7). Still, maybe it's a good idea to sort metadata explicitly.

tests: disable revlog compression in test-generaldelta.t (issue6867) The revlog compression makes a lot of numbers unstable. Since checking revlog compression is not the goal of this test, we disable the compression to get stable numbers. This should avoid wasting more time on this kind of changes in the future.

test-general-delta: actually test optimize-delta-parent-choice=no Since the configuration was not explicit, the case stopped testing what it intended to test when the default value changed.

test-chg: stabilize the log checking The "worker process exited" line have been making the CI flaky for a long time. Lets sort this out.

tests: fix test-patchbomb-tls.t instability The flakiness on chg is caused by a client that exit faster than the server output log. So actively wait for the server to issue the expected output (with a small timeout)

test-lock: use synchronisation file instead of sleep This will prevent the test to be flaky on load.

branchcache: use update_disk to refresh 'served' and 'served.hidden' The `update_disk` method is dedicated to this kind of usecase. Now that the writting patterns are more consistent, we can use it to warm these two important cache. I am dropping the first comment about "refreshing all the others" because it is false. If a branchmap already exist for "served", none of the subset will be updated.

branchcache: explictly update disk state only if no transaction exist If a transaction exist the `write_dirty` call will eventually be done and the state will be synched on disk. It is better to no interfer with that.

branchcache: do not use `__getitem__` in updatecache The `update_disk` method uses `updatecache` and the point of `update_disk` is to be able to do alternative processing to the one we do in `__getitem__`. So we calling `__getitem__` in `updatecache` defeat this purpose. Instead we do the equivalent explicitly to preserve the spirit of `update_disk` (that we will actually put to use soon, I promise)

branchcache: explicitly track inheritence "state" We move from a binary "dirty" flag to a three value "state": "clean", "inherited", "dirty". The "inherited" means that the branch cache is not only "clean", but it is a duplicate of its parent filter. If a branch cache is "inherited", we can non only skip writing its value on disk, but it is a good idea to delete any stale value on disk, as those will just waste time (and possibly induce bug) in the future. We only do this in the update related to transaction or explicit cache update (e.g `hg debugupdatecache`). Deleting the file when we simply detected a stall cache during a read only operation seems more dangerous. We rename `copy` to `inherit_for` to clarify we associate a stronger semantic to the operation.

branchcache: stop writing more branchcache file on disk than needed Before this change, we were unconditionally writing a branchmap file for the filter level passed to `update_disk`. This is actually counter productive if no update were needed for this filter level. In many case, the branch cache for a filter level is identical to its parent "subset" and it is better to simply keep the subset update and reuse it every time instead of having to do identical work for similar subset. So we change the `update_disk` method to only write a file when that filter level differ from its parent. This removes many cases where identical files were written, requiring multiple boring update in the test suite. The only notable changes is the change to `test-strip-branch-cache.t`, this case was checking a scenario that no longer reproduce the bug as writing less branchmap file result in less stalled cache on disk. Strictly speaking, we could create a more convoluted scenario that create a similar issue. However the next changeset would also cover that scenario so we directly updated that test case to a "no longer buggy" state.

branchcache: do not copy the `_dirty` flag If the inherited branch cache is dirty, it will be written on disk, and the super-set did not need to modify it, the on disk value for the subset will be re-useable as is. So the super set does not needs to write the very same content itself.

branchcache: explicitly assert that copy is always about inheritance This would catch cases where copy is used for something else if any existed.

branchcache: stop using `copy(…)` in `replace(…)` The `copy` method is mostly used for a filter level to inherit the branchmap from a subset. So we stop using (abusing) it in "replace" to ensure `copy` is used only for inheritance purposes. Since `replace` is a method of the BranchMapCache, it seems fine to do lower level operation there.

branchcache: change the _delayed flag to an explicit `_dirty` flag This is more consistent with the logic we use for other object and it open the way to a clearer management of the cache state. Now, cache are created clean, cache update mark them dirty, writing them on disk mark them clean again.

branchcache: write branchmap in subset inheritance order This way, we can guarantee a valid subset has been written before touching the branchmap of another filter. This is especially useful as we are bout to start deleting outdated branchmap file.

branchcache: do not accept "empty update" This currently does not happens and it will be simpler that is remains that way. If all update do something, we will be able to simply declare, in a later changesets, that all update to result in a dirty branchcache.

branchcache: avoid created a `None` filter repoview when writing The repoview class is not intended to be used for unfiltered repository.

stream-clone-tests: stop filtering non existent warning This filtering was introduced in 74c004a515bc, however there is already no warning in that changeset. So I guess the warnings existed when we the patch was created but the problem was solved in another changeset that 74c004a515bc, rebased on.

stream-clone-test: simplify case testing obsolescence There is only two important things in this test: - the number of file we send, to show we picked the obsstore. - the resulting state, to show we did alter things in the process. The rest are of the number are very fragile and consume a lot of time for little value when adjusting formats, caches, and protocol.

stream-clone-test: simplify the case testing phases There is only two important things in this test: - the number of file we send, to show we picked the phase roots. - the resulting phases, to show we did not modified them. The rest are of the number are very fragile and consume a lot of time for little value when adjusting formats, caches, and protocol.

stream-clone-test: simplify bookmark clone The important things to test here is the number of file included (to catch that the bookmark file was sent). So we keep that part non glob'ed but glob the rest. The glob'ed numbers are very fragile and consume a lot of time for little value when adjusting formats, caches, and protocol.

stream-clone-test: add a verify call to the "clone while changing" case It seems useful to very that the clone did not result in a corrupted copy.

stream-clone-test: add title to various test cases These case are fine as is, but as we are adding title to all the other as we simplify them, lets add title for all cases.

stream-clone-test: simplify testing of secret cloning restriction Here, we just want to check if the streaming clone is allowed and used or not. We do not care about the details of the clone itself.

stream-clone-test: simplify the background file closing test Here we just care about the fact the background file closing logic actually ran. We don't need to check the details of the cloning. The details of the output is very fragile and consume a lot of time for little value when adjusting formats, caches, and protocol. So we filter it out.

stream-clone-test: simplify the --uncompressed alias check To check that --uncompressed is an alias we just need to check it trigger a stream clone, we don't need to check anything else.

stream-clone-test: drop an automatic pattern replacement That pattern is nice, but it prevent us to glob the number of bytes when we don't care about them. We don't care about them more often that what we currently checks so dropping this pattern will help use to simplify various tests.

stream-clone-test: simplify the test for getbundle with stream=1 The core of this tests is about checking we receive a stream bundle with such request. We don't need to look at too much of the details of the stream itself. Since the content of the stream if shifting overtime, Such check is very fragile and consume a lot of time for little value when adjusting formats, caches, and protocol. So we reduce the size of what we check to focus on "is this a stream clone" question.

stream-clone-test: factor some piece of basic clone test out Multiple parts of this case (listing cache, checking error) are common to all cases and don't need to be in the conditionnal block. This simplify the test update.

stream-clone-test: simplify the case where server disabled it We have an option to disable it, we don't need to test it with all protocol variants. In addition there is little value in looking at the bytes to bytes details of the reply. Such check is very fragile and consume a lot of time for little value when adjusting formats, caches, and protocol.

rust-matchers: raw regular expression builder Extracting this `re_builder()` from `re_matcher()` makes it reusable in more general cases than matching `HgPath` instances and would help reducing code duplication in RHGitaly.

rust-filepatterns: export glob_to_re function Making this function public should not risk freezing the internal API, and it can be useful for all downstream code that needs to perform glob matching against byte strings, such as RHGitaly where it will be useful to match on branches and tags.

repoview: prevent `None` to be passed as the filtername We let such instantiation slip in a previous commit, so we add an explicit check to prevent it to happen in the future.

branchcache: avoid created a `None` filter repoview when writing The repoview class is not intended to be used for unfiltered repository.

rust-index: don't use mutable borrow to computed filtered heads This does not need to mutate the index. This is the prime suspect for some RuntimeError raised during some pushes.

rust-index: don't use mutable borrow for head-diff computation It does not needs to mutate the index. This is one of the two suspects of RuntimeError being thrown during push.

branchcache: move head writing in a `_write_headers` method Same rational: this will help having format variants.

branchcache: move head writing in a `_write_heads` method Same rational: this will help having format variants.

branchcache: move the header loading in a `_load_header` class method This will help changing header parsing in format variants.

branchcache: simplify a long line Gratuitous change to help code readability.

branchcache: rename `load` to `_load_heads` We are about to have more similar function, we rename the existing one to a more meaningful name and mark it private in the process.

branchcache: move the filename to a class attribute This prepare the introduction of more variant of cache.

test-clonebundles: simplify matching to be less flavor depends We keep the files and bytes output for the first call, but then we mostly check that we are being served a stream-clone bundle, not the actual content and size of the bundle. That aspect being tested by the stream clone test themselves.

repoview: fix changelog.__contains__ method This have been around for ten years, so we can safely that this method have few callers. However I am about to add one.

branchcache: unconditionally write delayed branchmap

branchcache: drop the unused `_verifyclosed` This code appears dead since its introduction about 5 years ago in this three consecutive commits: - 6578654916ae → introduce the method with two calls - 7c9d4cf23adf → remove first call - be5eeaf5c24a → remove second call o changeset: be5eeaf5c24a | user: Pulkit Goyal <pulkit@yandex-team.ru> | date: Fri Apr 05 15:57:09 2019 +0300 | summary: branchcache: don't verify closed nodes in _branchtip() | o changeset: 7c9d4cf23adf | user: Pulkit Goyal <pulkit@yandex-team.ru> | date: Fri Apr 05 15:56:33 2019 +0300 | summary: branchcache: don't verify closed nodes in iteropen() | o changeset: 6578654916ae | user: Pulkit Goyal <pulkit@yandex-team.ru> ~ date: Mon Apr 01 13:56:47 2019 +0300 summary: branchcache: lazily validate nodes from the branchmap

branchcache: dispatch the code into the dedicated subclass The code useful only to the local brancache have now been moved into the dedicated subclass. This will help improving the branchcache code without subtle breaking the remote variants.

branchcache: introduce a base class for branchmap This will help define a clear boundary between the two.

branchcache: fix the copy code We copy some internal attribute along too. This should prevent inconsistency in the resulting branchmap.

branchcache: pass a "verify_node" attribut to __init__ instead of hasnode The hasnode callback cannot be inherited and is dropped on copy, which seems like a bad idea. Instead we pass the actual semantic as a parameter and let the internal logic deal with it.

branchcache: stop storing a repository instance on the cache altogether We did not really needed it and we do not needs it anymore at all. So lets make things simpler for consistency and garbage collecting and stop storing it altogether.

branchcache: pass the target repository when copying Branchmap are usually copied to be used on a different repoview using a different filter level. Passing the repository around means the repository in `branchcache._repo` will drift from the actual branchmap filter. This is currently "fine" because the repo is only used to retrieve the `nullid` value. However, this is a fairly big trap for any extension or future code using the `_repo` attribute. The replace logic is now using a copy to ensure the right repository view is used to initialized the cached value. We add a couple of assert for make sure this inconsistency does not sneak back.

branchcache: have an explicit method to update the on disk cache Explicit is better and will give use more flexibility for future evolution of the storage.

crecord: drop calls to `curses.endwin()` We got a bug report where `curses.endwin()` failed with `_curses.error: endwin() returned ERR`. Looking at e306d552dfb12, it seems like we should be able to just remove these calls.

config: move the option to mmap rev branch cache in the storage section See previous commit for rational.

config: document the storage and format sections This should help people to put configuration in the right section.

rust-index: drop offset_override The inline `offsets` value diverge from the one on disk for added value, so the offset_override tricks is not going to work well once we start having the full revlog logic in Rust. We remove it beforehand and align the Rust logic to the Python one (adjusting the segment offset at read time for inline revlog).

rust-index: stop calling `with_offset` in the tests We are not adding any data, so why are we setting any offset?

branching: merge stable into default

Added signature for changeset d1d48d18db37

Added tag 6.7rc0 for changeset d1d48d18db37

relnotes: add 6.7rc0

relnotes: remove outdated message from `next`

branching: merge default into stable for 6.7rc0

branching: merge stable into default

perf: add a --as-push option to perf::unbundle This turned out to make a quite significant difference.

chainsaw-update: exit early if one of the intermediate command fails That will prevent the user to be presented with a start that pretend to be consistent with the request, but is not.

chainsaw-update: lock the repository for the duration of the operation This should prevent and catch some misusage where something else try to touch the repository.

chainsaw-update: taking care of initial cloning Perhaps we should go just a bit lower level than this `instance()`, since the main added value in our use-case is full path resolution, that we need to do anyway for the rmtree cleanup.

chainsaw-update: use a graph with branching in graph This will be relevant for the next improvement of `chainsaw-update`.

chainsaw-update: log actual locks breaking Previously, the command would simply state that it was about to break locks, not if there was actually some to break. This version is race-free. It would be also possible to display the content of the lock before hand (not race-free but informative in almost all cases).

vfs: have tryunlink tell what it did It is useful in certain circumstances to know whether vfs.tryunlink() actually removed something or not, be it for logging purposes.

chainsaw: new extension for dangerous operations The first provided command is `chainsaw-update`, whose one and single job is to make sure that it will pull, update and purge the target repository, no matter what may be in the way (locks, notably), see docstring for rationale.

rust: disable the RustIndex without persistent nodemap See rational inline.

rust: stop claiming the C index is compatible with the rust code This is no longer the case since the introduction of the pure Rust Index, and was probably not the case since the MixedIndex itself. So we fix the dedicated attribute value.

rust-index: remove one collect when converting back Turns out this is slightly faster. Sending the results back to Python is still the most costly (like 75% of the time) of the whole method, but it's about as fast as it can be now. hg perf::phases on mozilla-try-2023-03-22 before: 0.267114 after: 0.247101

rust-index: improve phase computation speed While less memory efficient, using an array is *much* faster than using a HashMap, especially with the default hasher. It even makes the code simpler, so I'm not really sure what I was thinking in the first place, maybe it's more obvious now. This fix a significant performance regression when using the rust version of the code. (however, the C code still outperform rust on this operation) hg perf::phases on mozilla-try-2023-03-22 - 6.6.3: 0.451239 seconds - before: 0.982495 seconds - after: 0.265347 seconds - C code: 0.183241 second

phases: directly update the phase sets in advanceboundary This is similar to what we do in retractboundary. There is no need to invalidate the cache if we have everything at hand to update it.

phases: large rework of advance boundary In a similar spirit as the rework of retractboundary, the new algorithm is doing an amount of work in the order of magnitude of the amount of changeset that changes phases. (except to find new roots in impacted higher phases if any may exists). This result in a very significant speedup for repository with many old draft like mozilla try. runtime of perf:unbundle for a bundle constaining a single changeset (C code): before 6.7 phase work: 14.497 seconds before this change: 6.311 seconds (-55%) with this change: 2.240 seconds (-85%) Combined with the other patches that fixes the phases computation in the Rust index, the rust code with a persistent nodemap get back to quite interresting performances with 2.026 seconds for the same operation, about 10% faster than the C code.

phases: apply similar early filtering to advanceboundary advanceboundary is called the push's unbundle (but not the other unbundle) so advanceboundary did not show up the profile I looked at so far. We start with simple pre-filtering to avoid doing any work if we don't needs too.

phases: filter revision that are already in the right phase No need to compute new roots if everything is already in order.

phases: invalidate the phases set less often on retract boundary We already have the information to update the phase set, so we do so directly instead of invalidating the cache. This show a sizeable speedup in our `perf::unbundle` benchmark on the many-draft mozilla-try repository. ### data-env-vars.name = mozilla-try-2023-03-22-zstd-sparse-revlog # benchmark.name = hg.perf.perf-unbundle # bin-env-vars.hg.flavor = no-rust # bin-env-vars.hg.py-re2-module = default # benchmark.variants.issue6528 = disabled # benchmark.variants.revs = last-10 before: 2.055259 seconds after: 1.887064 seconds (-8.18%) # benchmark.variants.revs = last-100 before: 2.409239 seconds after: 2.222429 seconds (-7.75%) # benchmark.variants.revs = last-1000 before: 3.945648 seconds after: 3.762480 seconds (-4.64%)

phases: incrementally update the phase sets when reasonable When the amount of manual walking is small, we update the phases set manually instead of computing them from scratch. This should help small update. The next changesets will make this used more often by reducing the amount of full invalidation we do on roots upgrade. The criteria for using an incremental upgrade are arbitrary, however, it "should never hurt".

phasees: properly shallow caopy the phase sets dictionary We are about to increments the set more incrementally in some case, so we need to make a proper shallow copy of it.

phases: pass an unfiltered repository to _ensure_phase_sets It seems better for such a low level function to be able to assume it operate on a real repository.

phases: drop set building in `hasnonpublicphases` We don't actually use the set, so why do we ensure they are built? (we should also clean up the use of repository argument but that's a quest for later).

phases: gather the logic for phasesets update in a single method This logic is duplicated around for no good reason, we gather it in a single place. The conditional is the new function are a bit weird as we about going to extend it soon.

phases: change the way we warm the phasecache in repocache Same logic as for the previous chngeset. We are about to rename and change the method used here.

phases: use a more generic way to trigger a phases computation for perf Querying the tip most revision will require the cache to warm the same as calling the dedicated method. This avoid using a method that is mostly meant for internal use and will be renamed in a coming changesets.