dirstate: stop using `entry.state()` for logic in `verify` Use the new API instead of the deprecated one. The next changeset will remove the use of `state` in this method altogether (currently still in the messages)

dirstate-entry: add `modified` property This was already done in the Rust implementation and is a useful primitive. The C implementation had this called `merged`, but wasn't used anywhere. It will be used in the next changeset.

debug: add debug-revlog-stats command Display statistics about revlogs in the store. Useful to get an approximate size of a repository, etc. More statistics will be added in the future.

typing: attempt to remove @overloads in the platform module for stdlib methods This is mostly successful, as examining util.pyi, posix.pyi, and windows.pyi after a pytype run shows that the type overloads for `oslink`, `readlink`, `removedirs`, `rename`, `split`, and `unlink` have been removed. (Some of these still have an @overload, but the differences are the variable names, not the types.) However, @overloads remain for `abspath` and `normpath` for some reason. It's useful to redefine these methods for the type checking phase because in addition to excluding str and PathLike variants, some of these functions have optional args in stdlib that aren't implemented in the custom implementation on Windows, and we want the type checking to flag that instead of assuming it's an allowable overload everywhere. One last quirk I noticed that I can't explain- `pycompat.TYPE_CHECKING` is always False, so the conditionals need to check `typing.TYPE_CHECKING` directly. I tried dropping the custom code for assigning `pycompat.TYPE_CHECKING` and simply did `from typing import TYPE_CHECKING` directly in pycompat.py, and used `pycompat.TYPE_CHECKING` for the conditional here... and pytype complained that `pycompat` doesn't have the `TYPE_CHECKING` variable.

typing: add trivial type hints to rest of the windows platform module Skipping the file wrappers for now because there's interplay with C code, and making them subclass `typing.BinaryIO_Proxy` confuses PyCharm a bit.

typing: add type hints to the rest of the posix module These methods either don't have an analog in the windows module, or are aliased in the windows module from something else (like os.path.xxx).

typing: add type hints to the platform `cachestat` classes

util: fix the signature of observedbufferedinputpipe._fillbuffer() Flagged by PyCharm, since it didn't match the signature of the method being overridden. The default value in the superclass is also `_chunksize`, and I suspect that the amount read from `osread` should be limited to what is passed in. Only one caller (`bufferedinputpipe.unbufferedread()`) passes this argument, and it passes the max of `_chunksize` and whatever it was passed.

tests: drop some obsolete py2 handling in util.py doctest Flagged by PyCharm while inspecting imports from the platform modules.

typing: add type hints to the common posix/windows platform functions These are done in sync because some platforms have empty implementations, and it isn't obvious what the types should be without examining the other. We want the types aligned, so @overload definitions that differ aren't generated. The only differences here are the few methods that unconditionally raise an error are marked as `NoReturn`, which doesn't seem to bother pytype. A couple of the posix module functions needed to be updated with a modern ternary operator, because pytype seems to want to use the type of the second object in the old `return x and y` style.

typing: add type hints to the posix platform module matching win32.py

typing: add type hints to mercurial/win32.py These are the low level functions that are imported by the mercurial.windows module, which is in turn imported by mercurial.utils as the platform module. Pretty straightforward, but pytype inferred very little of it, likely because of the heavy ctypes usage. It also seems to trigger a pytype bug in procutil, now that it has an idea of the underlying function type, so disable that warning to maintain a working test.

windows: drop some py2 registry module importing The comment was actually backwards- `winreg` is importable on py3, and is already imported by mercurial/windows.py.

typing: add type hints to the platform specific scm modules Surprisingly, pytype struggled to figure out the return types in the posix functions.

typing: add type hints to most mercurial/pycompat.py functions The `rapply` methods are left out because it's not `rapply(f, xs: _T0) -> _T0` as I first thought- it's used somewhere to walk a collection and convert between bytes and str. Also, the `open()` call is partially untyped because I'm not sure what its purpose is at this point- both the name and mode can be either bytes or str as it is currently constituted. It might make sense to assert that the file is being opened in binary mode (like `namedtempfile()`) and cast the result to `BinaryIO`, but that shouldn't be smuggled in with these other changes. The return is currently typed as `Any` because something suddenly got smarter and a few uses in util.py (like readfile()) suddenly think it returns `IO[str]` instead of `IO[bytes]` (BinaryIO), and it flags the type mismatch there.

statprof: don't pass str `sys.argv` to a function expecting bytes Found by typing the global functions in mercurial.pycompat.

typing: drop an unnecessary warning disabling comment in match.py This stopped being necessary in d2e1dcd4490d, when the exception stopped being subscripted.

scmposix: don't subscript IOError This warning disabling has been in place since late 2019 in 667f56d73ceb. We should have had some py3 support at the time, but both pytype complains and subscripting a real FileNotFoundError generated in `hg debugshell` crashed, so maybe this fixes a problem. It looks like all other instances of subscripting exceptions have been replaced (at least as far as greping for `== errno.` revealed).

typing: add type hints to pycompat.bytestr The problem with leaving pytype to its own devices here was that for functions that returned a bytestr, pytype inferred `Union[bytes, int]`. It now accepts that it can be treated as plain bytes. I wasn't able to figure out the arg type for `__getitem__`- `SupportsIndex` (which PyCharm indicated is how the superclass function is typed) got flagged: File "/mnt/c/Users/Matt/hg/mercurial/pycompat.py", line 236, in __getitem__: unsupported operand type(s) for item retrieval: bytestr and SupportsIndex [unsupported-operands] Function __getitem__ on bytestr expects int But some caller got flagged when I marked it as `int`. There's some minor spillover problems elsewhere- pytype doesn't seem to recognize that `bytes.startswith()` can optionally take a 3rd and 4th arg, so those few places have the warning disabled. It also flags where the tar API is being abused, but that would be a tricky refactor (and would require typing extensions until py3.7 is dropped), so disable those too.

pycompat: explicitly prefix builtin attr usage with `builtins.` It doesn't seem like this would fix any bug, because the wrapped functions that take bytes instead of str are defined after these calls. But PyCharm was flagging the second and third uses, saying "Type 'str' doesn't have expected attribute 'decode'". It wasn't flagging the first, but I changed it for consistency.

typing: add type hints to global variables in mercurial/pycompat.py The way `osaltsep` and `sysexecutable` were defined, pytype determined them to be `Union[bytes, str]`. This was a problem because that cascaded to all of the callers, and also because it couldn't be annotated as bytes on the initial assignment. Therefore, we use a ternary operator. The documentation says that `sys.executable` can either be None or an empty string if the value couldn't be determined. We opt for an empty string here because there are places that blindly pass it to `os.path.xxx()` functions, which crash if given None. Other places test `if pycompat.sysexecutable`, so empty string works for both.

windows: drop an unused method The only caller was removed in 563eb25e079b.

typing: add type hints to the prompt methods in mercurial/ui.py The @overloads allow for the callers that pass a non-None `default` to not have to worry about handling a None return to appease pytype.

ui: split the `default` arg out of **kwargs for the internal prompt method This arg was required anyway, based on how it was accessed. Having it separate allows it to be typed though, and this will simplify things for the callers- if a non-None `default` is passed, the return can never be None. That can be expressed with `@overload` when the arg can be typed, but that's not possible when it is rolled up in **kwargs. The default value is simply copied from the public `prompt()` above it.

typing: add trivial type hints to mercurial/ui.py There's not really a pattern here; it's mostly obvious return types and in a few cases, obvious parameter types. Some other "obvious" functions are left out because of quirks in how the return value for the various config() functions are inferred cause pytype to complain.

doc: don't pass str to ui methods in check-seclevel.py

typing: add type hints related to message output in mercurial/ui.py This will shake loose some bytes vs str issues in the doc checker.

typing: add type hints related to progress bars in mercurial/ui.py Pretty low hanging fruit while trying to deal with other more complicated parts of this module.

pytype: stop excluding mercurial/ui.py ui.extractchoices() is perhaps making assumptions that it shouldn't about the pattern always matching, but presumably we have test coverage for that. PyCharm flags the updated classes with a warning "Class xxx must implement all abstract methods", and suggests adding `abc.ABC` to the superclasses. I'm not sure why, unless it doesn't recognize the `__getattr__()` delegation pattern. Additionally, we can't unconditionally subclass `typing.BinaryIO` because that defeats the `__getattr__` delegation to the wrapped object at runtime. Instead, it has to only subclass during the type checking phase[1]. In any event, this fixes: File "/mnt/c/Users/Matt/hg/mercurial/ui.py", line 1518, in _runpager: Function subprocess.Popen.__new__ was called with the wrong arguments [wrong-arg-types] Expected: (cls, args, bufsize, executable, stdin, stdout: Optional[Union[IO, int]] = ..., ...) Actually passed: (cls, args, bufsize, stdin, stdout: Union[mercurial.utils.procutil.WriteAllWrapper, mercurial.windows.winstdout], ...) File "/mnt/c/Users/Matt/hg/mercurial/ui.py", line 1798, in extractchoices: No attribute 'group' on None [attribute-error] In Optional[Match[bytes]] File "/mnt/c/Users/Matt/hg/mercurial/ui.py", line 1799, in extractchoices: No attribute 'group' on None [attribute-error] In Optional[Match[bytes]] [1] https://stackoverflow.com/q/71365594

bundle: emit full snapshot as is, without doing a redelta With the new `forced` delta-reused policy, it become important to be able to send full snapshot where full snapshot are needed. Otherwise, the fallback delta will simply be used on the client side… creating monstrous delta chain, since revision that are meant as a reset of delta-chain chain becoming too complex are simply adding a new full delta-tree on the leaf of another one. In the `non-forced` cases, client process full snapshot from the bundle differently from deltas, so client will still try to convert the full snapshot into a delta if possible. So this will no lead to pathological storage explosion. I have considered making this configurable, but the impact seems limited enough that it does not seems to be worth it. Especially with the current sparse-revlog format that use "delta-tree" with multiple level snapshots, full snapshot are much less frequent and not that different from other intermediate snapshot that we are already sending over the wire anyway. CPU wise, this will help the bundling side a little as it will not need to reconstruct revisions and compute deltas. The unbundling side might save a tiny amount of CPU as it won't need to reconstruct the delta-base to reconstruct the revision full text. This only slightly visible in some of the benchmarks. And have no real impact on most of them. ### data-env-vars.name = pypy-2018-08-01-zstd-sparse-revlog # benchmark.name = perf-bundle # benchmark.variants.revs = last-40000 before: 11.467186 seconds just-emit-full: 11.190576 seconds (-2.41%) with-pull-force: 11.041091 seconds (-3.72%) # benchmark.name = perf-unbundle # benchmark.variants.revs = last-40000 before: 16.744862 just-emit-full:: 16.561036 seconds (-1.10%) with-pull-force: 16.389344 seconds (-2.12%) # benchmark.name = pull # benchmark.variants.revs = last-40000 before: 26.870569 just-emit-full: 26.391188 seconds (-1.78%) with-pull-force: 25.633184 seconds (-4.60%) Space wise (so network-wise) the impact is fairly small. When taking compression into account. Below are tests the size of `hg bundle --all` for a handful of benchmark repositories (with bzip, zstd compression and without it) This show a small increase in the bundle size, but nothing really significant except maybe for mozilla-try (+12%) that nobody really pulls large chunk of anyway. Mozilla-try is also the repository that benefit the most for not having to recompute deltas client size. ### mercurial: bzip-before: 26 406 342 bytes bzip-after: 26 691 543 bytes +1.08% zstd-before: 27 918 645 bytes zstd-after: 28 075 896 bytes +0.56% none-before: 98 675 601 bytes none-after: 100 411 237 bytes +1.76% ### pypy bzip-before: 201 295 752 bytes bzip-after: 209 780 282 bytes +4.21% zstd-before: 202 974 795 bytes zstd-after: 205 165 780 bytes +1.08% none-before: 871 070 261 bytes none-after: 993 595 057 bytes +14.07% ### netbeans bzip-before: 601 314 330 bytes bzip-after: 614 246 241 bytes +2.15% zstd-before: 604 745 136 bytes zstd-after: 615 497 705 bytes +1.78% none-before: 3 338 238 571 bytes none-after: 3 439 422 535 bytes +3.03% ### mozilla-central bzip-before: 1 493 006 921 bytes bzip-after: 1 549 650 570 bytes +3.79% zstd-before: 1 481 910 102 bytes zstd-after: 1 513 052 415 bytes +2.10% none-before: 6 535 929 910 bytes none-after: 7 010 191 342 bytes +7.26% ### mozilla-try bzip-before: 6 583 425 999 bytes bzip-after: 7 423 536 928 bytes +12.76% zstd-before: 6 021 009 212 bytes zstd-after: 6 674 922 420 bytes +10.86% none-before: 22 954 739 558 bytes none-after: 26 013 854 771 bytes +13.32%

bundle: when forcing acceptance of incoming delta also accept snapshot Snapshot where never considered reusable and the unbundling side always tried to find a delta from them. In the `forced` mode this is counter-productive because it will either connect two delta-tree that should not be connected or it will spend potentially a lot of time because creating a full snapshot anyway. So in this mode, we accept the full snapshot as is. This changeset is benchmarked with its children so please do not split them apart when landing.

delta-find: properly report full snapshot used from cache as such The number of tries and the delta base is reported differently so we missed there detection initially.

test-acl: glob the payload size again This size of bundle-2 payload are irrelevant for this test and only appears in its output because other pieces of the debug output are important. We glob it these number before they get in our way again.

amend: add a --draft option to set phase to draft Some users create commits in secret phase by default and then want to make them draft so they can be uploaded. This patch adds a --draft option for that. We already have a flag for changing the phase to secret, so it seems consistent to have one for draft.

commit: add --draft option to use draft phase

tests: use graph log in test-phases.t It's hard to tell that the phases are ordered correctly without seeing the graph.

commit: move check for incompatible args earlier I think it makes sense to check the command line arguments as early as possible, so we don't have to wait for a repo lock to tell the user that they passed invalid arguments.

delta-find: add a delta-reuse policy that blindly accepts incoming deltas When this policy is set, incoming deltas are blindly accepted without regard for the validity of the chain they build.

delta-find: add a `delta-reuse-policy` on configuration `path` That option allows to control the behavior on a per-path basis, opening the way to treating pulls from central servers differently than other operations.

changegroup: add `delta_base_reuse_policy` argument The argument available through function from changegroup.apply to `revlog.apply` allow to override the revlog configuration in terms of delta-base-reuse policy when searching for a delta to store a revision. It will be put to use in the next changesets.

bundleoperation: optionnaly record the `remote` that produced the bundle We have the information at hand, and the peer now have knownledge of its `path` object, which constaints useful behavior configuration. So the simpler seems to be to pass that object around so it can be used if needed.

delta-find: add a test checking various simple behavior There are enough work happening in this area that it is worth having a dedicated test for it. So far we add to small test checking that the "best" parent is picked as the delta base and that this behavior can be controlled during commit and unbundle.

peer: pass the `path` to the statichttp peer We now have all peer able to receive and store a `path` object. Hooray.

peer: get the `path` object down to the sshpeer Same logic as the other peers.

logexchange: use the proper accessors to get the remote url There is an official method, let us use it. this will prevent a crash when the private attribute disappear.

peer: get the `path` object down to the httppeer One more peer with a path stored.

path: fix `url.copy` dropping the port The copy method have been wrong for a while, but the new code reveals it.

peer: pass the `path` object to `make_peer` We don't do anything with it yet, but we can start implementing it for each peer type starting now.

path: allow to copy a path while adjusting the url This will be used by `scheme` in the next changesets.

peer: store the path object used to build a peer from a repo This is the simplest case, we so starts with it.

peer: build a `path` object on the fly when needed So now, we always have a `path` object around when building the peer

peer: have `repo.peer` take an optional `path` argument We are ready to start to actually set the value now.

peer: add a `path` attribute to peer It will start being set in the coming changesets.

peer: have a common constructor and use it For now it does not do much, but we will extend it to also store a path object soon.

peer: use a dedicated name for the `peer` constructor We want to change the argument it takes, so we rather make them different function.

peer: dissolve `_peerlookup` into its last two callers This is about to need more changes and the function won't be useful. We do it early to clarify later changes.

peer: stop having a `peer()` method on `peer()` This is already a peer, why do you want a peer if you already have one.

clone: explicitly detect the need to fetch a peer Instead of having `peer()` method on all `peer()` for this usecase, we could simply handle it explicitly.

addbranchrevs: explicitly detect the need to fetch a peer Instead of having `peer()` method on all `peer()` for this usecase, we could simply handle it explicitly.

path: pass `path` to `peer` in `hg clone` We directly use the `path` object to build the `peer` object.

path: use `get_clone_path_obj` in share The return is simpler to use, and this mean less user for the old function. We directly use the `path` object to build the `peer` object.

path: pass `path` to `peer` in mq We directly use the `path` object to build the `peer` object. There is one case where we don't. We should fix that at the same time as we fix the sub-repo cases.

path: use `get_clone_path_obj` in _getlocal We don't need to feed the `path` object to a `peer` object, but using an higher level function is simpler here (e.g. no subscript access, explicit attribute access).

path: pass `path` to `peer` in `hg init` We directly use the `path` object to build the `peer` object.