stablerange: update the filename to avoid cache confusion The schema validation seems to not work as well as intended, we update the file name to make sure there will be no confusion.

stablesort: extract in its own module We are going to introduce variants, cache and more complex logic, we move that in its own repository beforehand.

compat: in depth cache, only use cachevfs when available We support older version without the new vfs.

depthcache: issue a progress bar when loading the depth cache The operation can be in the minute order of magnitude of some larger repository. We better keep the user entertain when he wait.

depthcache: move to a dedicated object and storage The SQL cache was overkill and fragile. We move to a dedicated storage using the new generic class.

cache: adds debug details about what the content of the update Having more data in debug is good, we add a small method to help providing them. This code was headed for core during the 4.3 cycle but never made it there. So we starts with a copy in the evolve repository.

cache: introduce a changelogsourcebase class This abstract class will help code that need a cache tracking the changelog content (eg: the branchmap cache). The cache key used is the same as what the branchmap uses. This code was headed for core during the 4.3 cycle but never made it there. So we starts with a copy in the evolve repository.

cache: introduce an abstract class for cache we can upgrade incrementally Right now each class implements their own mechanism for validation, and update. We start introducing abstract class to ultimately allow more unification of the cache code. The end goal of this series is to introduce a cache for some obsolescence property, not to actually implement the cache. However, taking advantage of adding a new cache to introduce the abstract class seems like a win. This code was headed for core during the 4.3 cycle but never made it there. So we starts with a copy in the evolve repository.

depth: basic debugdepth implementation and test This will come very handy to validate more advanced implementation later.

topics: add a config to reject draft changeset without topic on a server This patch adds a new config option experimental.topic-mode.server which if sets to True, the server won't accept any draft changeset without topic on it. In case both `experimental.topic-mode.server` and `experimental.topic.publish-bare-branch` are set to True, the enforce-topic one is respected. Tests are added for it. The CHANGELOG file is also updated mentioning about the config option.

test: update output to ad5f2b923b0d A new bundle2 capabilities have been added (related to bookmark). So we have to update the test output. CORE-TEST-OUTPUT-UPDATE: ad5f2b923b0d

touch: shorten touch's docstring to make it fit into a one line This make it show completely (untruncated) in `hg help -e evolve`. This follows up on fed946edc293 where the docstring got unwrap for that purpose and thus made the line length exceed 80 columns.

branching: merge with stable

help: promote "amend" to important command Recent change to what command we advertise as "basic". It turn out "amend" was not part of one of them. Since "amend" is one of the core command of a rewriting workflow, we promote it to "basic command".

help: fix output for `hg help -e evolve` wrt 'touch' command Currently, when running `hg help -e evolve`, the output looks like this on my screen: touch create successors that are identical to their predecessors except With this change, it looks like this: touch create successors that are identical to their predecessors except for the changeset ID

help: remove a few commands from `hg` (no args) command list According to `hg help -e evolve`, the following commands come from the evolve extension: amend evolve fold metaedit next obslog pdiff previous prune pstatus split touch uncommit If one runs `hg` without arguments, commands that are prefixed with a ^ character are shown. From the previous list, this includes: evolve fold metaedit next previous prune split touch uncommit It feels several of these commands are not "basic commands" that someone who has never used hg before might care about. They probably also do not come close to passing the "toothbrush test": things that users are likely to use every day. This commit removes these items from the list: metaedit, touch, uncommit This means the following are kept: evolve fold next previous prune split

typos: fix typos in several locations A user at Google attempted to fix our local copy because they had noticed a typo (accross instead of across), but this will just get overwritten on our next import. This commit fixes that case and a few others that my editor found. Most of the typos were in comments, but user-visible output is changed in a few cases: - accross -> across - splitted -> split - ambigious -> ambiguous - evolvestte -> evolvestate (this is in a ui.debug, so not often seen) There is another typo that I wanted to fix, but didn't: 'supercede' is spelled 'superseed' in a few locations. I believe this is only internal to the extension, instead of being user-visible, so while it could probably be fixed easily, I wasn't 100% sure it didn't end up in a file on disk or something and might cause problems, so I left it alone.

topic: use more protective code to access current topic There are some code path where the operation is not properly initialized and lack the new attributes. While the lack of initialization, is suspicious and I've not be able to pin point case where this happens. Currently the code crash when it encounter this situation. So we make the code more resistant to this case to prevent bad user experience.

topic: use more protective code to access publishing code There are some code path where the operation is not properly initialized and lack the new attributes. While the lack of initialization, is suspicious and I've not be able to pin point case where this happens. Currently the code crash when it encounter this situation. So we make the code more resistant to this case to prevent bad user experience.

typos: fix typos in several locations A user at Google attempted to fix our local copy because they had noticed a typo (accross instead of across), but this will just get overwritten on our next import. This commit fixes that case and a few others that my editor found. Most of the typos were in comments, but user-visible output is changed in a few cases: - accross -> across - splitted -> split - ambigious -> ambiguous - evolvestte -> evolvestate (this is in a ui.debug, so not often seen) There is another typo that I wanted to fix, but didn't: 'supercede' is spelled 'superseed' in a few locations. I believe this is only internal to the extension, instead of being user-visible, so while it could probably be fixed easily, I wasn't 100% sure it didn't end up in a file on disk or something and might cause problems, so I left it alone.

tests: add a test for newline in obsnote

obsnote: don't add '' to note while showing it in obslog

obsnote: add an entry in CHANGELOG about it

amend: add support for storing a note in obsmarker

uncommit: add support for storing a note in obsmarker

fold: add support for storing a note in obsmarker

metaedit: add support for storing a note in obsmarker

prune: add support for storing a note in obsmarker

split: add support for storing a note in obsmarker

touch: add support for storing a note in obsmarker

obslog: add initial support for showing note in obslog This add a very basic and initial support for showing the note stored in the obsmarker in obslog to test the upcoming changes which adds support. The UI is not finalised and will be improved in the patches after support is added and we have tests with the current output.

cmdrewrite: add a utility function to make sure note is of valid format Upcoming patches will add support to store a note on various evolution related commands by storing that in obsmarker. Since the obsmarker metadata field can't store things above 255 bytes, this patch introduces a function which will be used to make sure note is under 255 bytes and does not contain newlines. We don't want notes to be multilines.

branching: merge with stable

test: remove dangling --config in test-metaedit.t

tests: fix previously accidentally matching regexes in test-touch.t Before Mercurial commit 6d5718e39657 (run-tests: make "| foo (re)" not match everything, 2017-11-29), regexes like "| foo", which are common in graphlog output, would incorrectly match anything. After that commit, they match as they should, meaning that either "" or " foo" would match, but not a "| foo". This commit fixes the test by replacing the "|" by "." rather than escaping with "\|" since that would make the graph harder to read.

test: update test-discovery-obshashrange.t

test: remove dangling --config in test-metaedit.t

serverminitopic: also avoid reading Avoiding to write cache for filter where topic applies is a good step, but we need to also avoid reading it. Existing branchmap cache might exists before the turned the feature on.

topic: introduce a minimal extensions to enable topic on the server This small extensions simply expose topic in the branch. This should help getting minimal support for topic from various hosting solution (eg: bitbucket maybe ?) See extensions help for details.

branching: merge stable into default

pusoperation: wrap pushoperation __init__ directly This is how remote names does it. Using the same method should helps with compatibility when mixing the two extensions.

hgignore: fix .err matching patterns We don't want to ignore .err file outside of the 'tests/' directory.

changelog: fix 7.0.0 release date We date was the same as 6.8.0 for some unkown reason. We now use the date of the tag.

changelog: mention the improved verbosity handling

branching: merge with stable

changelog: update missing release date

summaryhook: respect -q flag There's no test coverage, but it seems pretty obvious and I don't care enough to write a test.

summaryhook: remove unused function

evolve: make prev/next respect --quiet flag (issue5742)

packaging: merge 7.0.1 into default

packaging: mark version as development version. This helps reduce confusion when installing non tagged version.

Added tag 7.0.1 for changeset e7a2a6158077

packaging: prepare version 7.0.1

packaging: update tested-with version Mercurial 4.4.1 is out.

changelog: mention the --publish bugfix

test-compat: merge mercurial-4.2 into mercurial-4.1

test-compat: merge mercurial-4.3 into mercurial-4.2

test-compat: merge stable into mercurial-4.3

compat: fix comp ability of the new phase logic for Mercurial < 4.4

config: clarify that config is only enabled on the repository is it setup for This changeset change nothing but clarify a few thing. The ui passed to 'reposetup' is the same as 'repo.ui'. However we would like to clarify things a bit.

branching: merge with an old and strange head

branching: default into stable It is time for a new release

topic: fix new head detection when using --publish on a topic I suspected a bug lurking around there and I found one.

effect-flag: fix obsfate verb hooking Mercurial core commit b81ad5b78a81 renamed a the function we hooked on and update its signature. Update the code to detect that, which fixes tests errors.

test: update test output for effect flag by default Now that effect flag is in core and turned on by default, this affect topic test that did not had the evolve extension turned on all the time. CORE-TEST-OUTPUT-UPDATE: a7e49a5b3e6f

tests: add more testing of pulling with various setting We checks for client/server with various combination: * extensions enabled, * extensions enabled but exchange disabled, * extensions disabled,

test-compat: merge mercurial-4.2 into mercurial-4.1

test-compat: merge mercurial-4.3 into mercurial-4.2

test-compat: merge stable into mercurial-4.3

branching: merge stable into default

changelog: add details about the pushkey related fix

wireproto: gracefully fail on http when trying to use older pushkey method Previously the HTTP server would return a 500. It now returns a 410.

wireproto: avoid exposing pushkey abort for server without evolution Client with evolution enabled will query the 'obsolete' pushley namespace if no other way to request obsmarkers have been found. They will do this unconditionally as long as no other obsolescence related capabilities have been advertised by the server. The call is "harmless" since the namespace will be empty if obsmarkers exchange is disabled. However, since we forbid the use of pushkey for obsmarkers this lead to unexpected crash report for up to date client talking to a server with the evolve extensions but exchange disabled. We fix the issue by serving the expected empty namespace in this case. We keep forbidding the query if obsmarker exchange is enabled since new client should have picked another more effective was to fetch the data first. We also add some tests for this logic.

obsdiscovery: enforce the config on the server too Right now, setting 'experimental.evolution.obsdiscovery=no' disable obsolescence markers discovery on client only. The server still advertise discovery related capabilities. We now also honor the config on the server side.

packaging: mark default branch as development version This help avoid confusion when installing version from the repository.

branching: merge back stable back into default

test-compat: merge mercurial-4.2 into mercurial-4.1

test-compat: merge mercurial-4.3 into mercurial-4.2

test-compat: merge stable into mercurial-4.3

packaging: mark as development version This help avoid confusion when installing non tagged version.

Added tag 7.0.0 for changeset c56c028f3802

packaging: prepare version 7.0.0

debian: drop outdate part of the description

debian: update Mercurial version dependencies The debian package now reflect the actual requirement of the extension in its current version.

branching: merge with stable to prepare version 7.0.0

push: add a --publish flag The new flag turns all element in the push to public changeset. This will help experiment with diverse publishing workflow.

topic: add an option to automatically publish topic-less changeset This new option will help playing with merge based workflow.

topic: add an option to enforce a single head per name in a repository

revset: clean up some messy logic

evolve: rename --phasedivergent flag to --phase-divergent This makes the whole things easier to read.

evolve: rename --contentdivergent flag to --content-divergent This makes the whole things easier to read.

pushkey: forbid usage of pushley to exchange markers The pushkey based protocol lacks many important feature and is very slow. We disable it to prevent any usage by mistake.

changelog: add an entry for 7.0.0 with recent change

compat: drop compat code for smartrevs.first The 'first' method have been around for years.

compat: drop compatibility dance around configoverride This was introduced in hg-4.1 and we do no longer support lower versions.

compat: drop support for older exchange protocol Their have been better alternative for years. In addition they have been officially deprecated for quite some time. So we official stop support them to ease the maintainance.

compat: drop compatibility for 'streamres' prior to 4.1 We no longer support these version.

compat: drop the directaccess hacky extension The user of this extension has forked their own version. We lighten our maintenance burden and drop it.

compat: drop the inhibit hacky extension The user of this extension has forked their own version. We lighten our maintenance burden and drop it.

compat: drop 'lru.get' work-around for 3.9 We no longer support this version.

compat: stop working around 3.8 file cache limitation We no longer support this version.

compat: simply the wrapping of '_destmergebranch' All the version we support now use the same interface.

compat: drop support for 3.8's '_destrebase' function We no longer support this version.

compat: mark version 4.1 as the minimum version required We dropped compatibility for version 3.8, 3.9 and 4.0.

compat: drop compatibility for Mercurial <= 4.0 We are dropping compatibility with older mercurial version. This will help us to clean up the code base.

packaging: mark as development version This helps avoid confusion.

branching: merge with stable Prepare the 7.0.0 line.

packaging: mark version as development version This helps to avoid confusion.

branching: close mercurial-4.0 branch We are dropping support for older Mercurial version.

branching: close mercurial-3.9 branch We are dropping support for older Mercurial version.

branching: close mercurial-3.8 branch We are dropping support for older Mercurial version.

util: add a small script to help with the merging of branches We have multiple test-compat branches merging them can be a bit tedious. We add a nice script doing so for us.

test-compat: merge mercurial-3.9 into mercurial-3.8

test-compat: merge mercurial-4.0 into mercurial-3.9

test-compat: merge mercurial-4.1 into mercurial-4.0

test-compat: merge mercurial-4.2 into mercurial-4.1

test-compat: merge mercurial-4.3 into mercurial-4.2

test-compat: merge stable into mercurial-4.3

Added tag 6.8.0 for changeset ec0bbf26ce7f

packaging: prepare version 6.8.0

flake8: rename some ambiguous identifier Apparently, pyflakes declared war on 'l'.

topic: instroduce a fast path when computing stack We do some stack related computation during transaction now. The current inefficiency of the revset is hurting us too much. We pre-compute a subset using native set operation to speed things up.

branching: merge stable back into default

test-compat: merge future 6.7.0 in mercurial-3.8 branch

test-compat: merge future 6.7.0 in mercurial-3.9 branch

test-compat: merge future 6.7.0 in mercurial-4.0 branch

test-compat: merge future 6.7.0 in mercurial-4.1 branch

test-compat: merge future 6.7.0 in mercurial-4.2 branch

test-compat: merge future 6.7.0 in mercurial-4.3 branch

obsfate: re-enables operation display in obsfate Now that the compatibility checks are in a good shape we can re-enable more noise in it.

changelog: update to reflect latest change

test-compat: merge with mercurial-3.9 branch

test-compat: merge with mercurial-4.0 branch

compat-test: remove topic related tests for 3.9 The extension is no longer supported for 3.9

test-compat: merge with mercurial-4.1 branch

test-compat: merge with mercurial-4.2 branch

test-compat: merge with mercurial-4.3 branch

branching: merge with mercurial-4.3 branch creation The other changeset marks the point were 'stable' is no longer about Mercurial version 4.3.

test-compat: revert output changes from phase renaming We keep using the old name for the older version. test has to reflect that.

test-compat: revert output changes related to temporary amend commit This backed out changeset 06844693bb21 and other related changes.

test-compat: revert ouput change to blackbox This backed out changeset 86be70ae38b9.

test-compat: revert ouput related to operation This effectively backs out 4b951a9a6895.

test-compat: revert output change in rebase This backed out changeset d12c82fd1143 and a couple of other changes. The change are not relevant in Mercurial < 4.4/

test-compat: revert output change from bundle2 capability This backout changeset dd3eda2f7a89 not relevant for verison < 4.4.

test-compat: revert output change on pull report The revert the changeset from 8feb2cae7eae. Not relevant for verison < 4.4.

test-compat: revert output change from 'phase' part The output change were introduced in changeset from 87e349f617f9 and it not relevant for version lower than 4.4.

test-compat: revert output change from using core obsfate The output update arrives in e91ca8b5ecf7. This output is only relevant for Mercurial 4.4

topic: do to check to topic while amending Right know the 'topic-mode' mess with amend. Especially in case of amending merge where the merge is not properly detected.

topic-mode: add missing end line in topic-mode warning

compat: work around the lack of configoverride in mercurial 4.0 The phase preservation will stay broken on 4.0 and below

compat: add an abstraction for 'scmutil.cleanupnodes' The usage of the function was added in 103244e34a9c but is not compatible with Mercurial <= 4.2.

compat: conditionally access config item Version 4.2 and below do not have config items support. So the extensions helper do not register anything.

compat: handle version where obsutil is not available Version older than 4.3 is do not have the obsutil option. We need to handle that.

test: use node instead of rev in topic push tests This makes the test more stable when revision number changes.

test: use node instead of rev in uncommit -i test This makes the test more stable when revision number changes.

branching: create the mercurial-4.3 test compatibility branch This branch will contains the expected test output with 4.3.

branch: merge with stable to prepare 6.7 release In addition, Mercurial 4.4-rc is out, so we need to update the expected output.

changeloge: add details about topic-mode

effect-flag: remove wrapping of createmarkers for Mercurial 4.4 Now that effect-flag has landed in Mercurial 4.4, we don't need to wrap createmarkers for this version anymore.

stack: order the adjective of changeset The new order give more interesting result: * base current * base unstable * current unstable It seems more harmonious this way. Base seems the most core and immutable adjective describing the element. Then current is a volatile but important one. Finally, unstable is less information than current, so it goes last.

stack: show current and unstable also for t0 and bases

obsfate: use core version of obsfate if available Let us use the upstream version whenever available. The operation display is currently disabled to ease with the test changes.

topic: register config items There is a couple more hoops to jump through but all config items used are now registered.

directaccess: register config item if possible Mercurial 4.4 warns about unregistered config.

evolve: registed configitems if available The config are registered as 'dynamicdefault' because we still need to explicitly pass the config option for older version of Mercurial. (4.2 and before).

inhibit: compatibility with 4.4-rc

branching: merge with stable Merge with the inhibit fix.

inhibit: respect '--keep' when handling rebase clean up

test: adapt to change from 537de0b14030 If we use phase for pull/push this impact some debug output. CORE-TEST-OUTPUT-UPDATE: 537de0b14030

test: adapt to output change from eb586ed5d8ce Core have a more verbose output on pull. We have to update tests with it. CORE-TEST-OUTPUT-UPDATE: eb586ed5d8ce

compat: fix compatibility with new duplicatecopies Changeset 754b5117622f updated the signature of 'copies.duplicatecopies'. We can now handle the multiple version of that code for compatiblity.

topic: remove 'ass' from the random animal list As much as I love donkey… this generate inappropriate names.

config: stop passing useless default value to config

branching: merge with stable

changelog: add about fixed bookmark and phase handling while changing topics

changelog: add about pager support

pager: add support to `hg evolve --list`

pager: add support to `hg obslog`

pager: add support to `hg topic`

pager: add support to `hg stack`

pager: add a function in compats to start pager This will help us enabling pager when ui.pager API exists. Also we can try plugging in the old pager API in this function and we won't have to change anything.

metadata: cleanup metadata file I'm not sure how this conflict marker survived the pre-publish test run...

branching: merge with stable

topic: remove a silly duplicate in a revset Spotted by Pulkit Goyal. More performance win seems to exists here, but we starts with the obvious.

topics: show changesetcount, troubledcount and headscount by default

topics: use stack.stack() instead of stack.stackdata() stack.stackdata() also calls stack.stack() internally. Moreover if we just want to calculate changesetcount, troubled count, it will be cheap to calculate using stack.stack rather than stackdata().

packaging: mark version as bugfix dev version This avoid confusion when installing non tagged version.

branching: merge with stable

Added tag 6.7.1 for changeset 430ad68292d7

obslog: remove the word "yet" from "No patch available yet" "yet" was a bit confusing as it let me think that patch is currently not available but can be available sometimes later. This is true for cases when it says "context not local" but we have more messages like "no successors", "too many successors", "changesets rebased" where we can't have a patch.

obslog: spell out successor completely If the successor is not known locally, obslog says `succ is unknown locally` which is not a good message. This patch makes sure succ is spelled out completely as successor.

readme: update readme with latest fix (also more the stack one in the right location).

obsfate: make the user list more readable Separate each user in the obsfate user list by a space for readability.

obsfate: fix user filtering in obsfate-printer Only hide current user only if it's the only one that create obs-markers.

obsfate: add a test with user both current and other This will help highlight change in the next changesets.

obsfate: update test The obsfate templatekw and the obsfateprinter behavior was different when displaying or not the current user in obsfate users. Update the test to makes the behavior fix more visible.

packaging: prepare version 6.7.1 The are an important fix on evolve and another important fix for topic performance, let us make a release.

readme: add changelog for topic 0.3.1

add mention of the topic 0.3.0 version

readme: fix evolve 6.7.1 entry

topic: reorder revset for faster evaluation When running the revset on unfiltered repository, the mutable set can get huge. So removing the obsolete changeset first provide a large speedup. This resolve the performance regression we were seeing since the 6.7.0 release.

topics: use stack rather than stackdata when one only wants the changeset count There might be a significant performance impact. Might resolve #309.

log: update obsfate output order Always display successors after the verb. So instead of "rewritten by boris as X", display "rewritten as X by boris". 07b9fcf8b6d3 did update template-based obsfate but forget to update obsfatelineprinter.

test: remove bashism in test-topic-change.t The test was broken on various machine using shell not supporting the `{start..end}` construct.

topics: handle phase correctly while changing topics

tests: add test showing change of phase while changing topic on secret commit

topics: use scmutil.cleanupnodes to create markers while changing topics Earlier if we change topics on changesets with bookmark, the bookmark is not moved to the new changeset as we didn't add that logic and we were creating obsmarkers directly. This patch calls scmutil.cleanupnodes to create obsmarkes which also handles the bookmark movement.

tests: add a new test for testing changing or clearing topics test-topic.t is getting bigger and bigger. This patch introduces a new test which includes test for clearing and changing topics. Since the tests are moved to a new file, this patch, removes those tests from test-topic.t also.

obscache: do not check of argument count of wrapped function After talking with the original author of this check, it appears that it never worked properly, since _readmarkers was always wrapped with @util.nogc.

legacy: rename lookup_errors to not be in all caps (flake8 warning) flake8 has a plugin, pep8-naming, which is not installed by default, but is used if available, no extra config required. This plugin checks names of variables, classes, functions and so on against PEP-8, and it complained about this variable: N806 variable in function should be lowercase Since this isn't a module-level constant, but just a helper variable used only in one function, it's fine to just rename it. With this error gone, flake8 output (used plugins: mccabe, naming, pycodestyle, pyflakes) is totally clean.

randomname: fix the header of the file

topics: move README-topic as hgext3rd/topic/README Before improving the docs about topics, lets move them into the topics folder.

README: move CHANGELOG to a new file

test: small cleanup in test-topic-mode.t remove the mention of "new" (this won't be new soon) and use the official config name.

test: add support for random-all topic mode The 'random' mode no longer generate a random topic for merge.

test: rename test-topicmode to test-topic-mode This is easier to read.

topicmode: add 'random' topic mode That mode will generate a random topic name when the user is about to commit an untopiced changeset.

topic: update the topic-mode hint The hint is now shorter and reference the documentation. That should provide more concise output with more details for user who seeks it.

topic: rename the 'topic-mode' help subtopic This will help making reference to it in warning.

topic: add a 'enforce-all' mode The mode abort for all commit without a topic, even merges.

topicmode: 'enforce' topic mode, no longer warn about untopiced merge merging a topic back in a branch is common case, it seems sensible to not pester the user about it.

topicmode: add new warning topicmode

topic: add documentation for the 'topic-mode' option This should help people to discover it.

test: update tests to use topic-mode enforce

topic: migrate experimental.enforce-topic to experimental.topic-mode

test: adapt to output change from 10e162bb9bf5 The changeset 10e162bb9bf5 in code extend the list of capabilities to include "phases=['heads']" this has a minor impact on test output CORE-TEST-OUTPUT-UPDATE: 10e162bb9bf5

merge with stable

stack: handle basic case of splitting with crash hg stack used to crash horribly whenever a split was encountered. This is no longer the case. If the result of the split (or there successors) produce a sensible linear result with on final head, we can just assume that head is the evolution destination.

tests: update test output with commit 4f969b9e0cf5 in core

tests: update test output with commit ecc96506c49d in core

packaging: update version to match development version This helps avoid confusion when installing from source.

packaging: merge stable back into default 6.7.0 has been released.

packaging: mark stable branch as dev bugfix version This help avoiding confusion when installing evolve from sources.

Added tag 6.7.0 for changeset 3a4f75c6619c

packaging: update debian changelog

packaging: update version number and testedwith version

packaging: mark version 6.7.0 in the source Version ready to be released.

packaging: update compatible version This has been tested with 4.3 too.

readme: prepare readme The feature that only work with 4.3+ are flagged as such.

compat-test: merge future 6.7.0 into 3.8 compat branch (merging through 3.9 compat branch)