repoview: move changelog.strip() override to filteredchangelog Differential Revision: https://phab.mercurial-scm.org/D7242

repoview: move changelog.headrevs() override to filteredchangelog Differential Revision: https://phab.mercurial-scm.org/D7241

repoview: move changelog.revs() override to filteredchangelog Differential Revision: https://phab.mercurial-scm.org/D7240

repoview: move changelog.__iter__() override to filteredchangelog Differential Revision: https://phab.mercurial-scm.org/D7239

repoview: move changelog.__contains__() override to filteredchangelog Differential Revision: https://phab.mercurial-scm.org/D7238

repoview: move changelog.tiprev() override to filteredchangelog Differential Revision: https://phab.mercurial-scm.org/D7237

repoview: wrap changelog class when filtering The class doesn't yet do anything. I'll move the filter-aware overrides from the changelog class over one by one to this class in coming patches. That will leave the changelog class simpler and will centralize more of the filtering logic to repoview. I could not measure any performance difference. Differential Revision: https://phab.mercurial-scm.org/D7236

repoview: extract a function for wrapping changelog I would like to clean up the changelog class by moving out knowledge of filtering. The filtering will instead be done in a class that wraps the changelog. This patch prepares for that. Differential Revision: https://phab.mercurial-scm.org/D7235

revlog: move tiprev() from changelog up to revlog This makes the changelog override simpler and it seems more consistent. Differential Revision: https://phab.mercurial-scm.org/D7250

tests: fix a few `(no-windows !)` conditionals that are really `(symlink !)` The CI for py3 is assuming that symlinks are possible (they are when running as Administrator or when Developer Mode is enabled), and these popped up. Differential Revision: https://phab.mercurial-scm.org/D7232

rust-matchers: add `Matcher` trait and implement `AlwaysMatcher` In our quest of a faster Mercurial, we have arrived at the point where we need to implement the matchers in Rust. This RFC mainly for the `Matcher` trait to see if the changes proposed feel fine to people with more experience on the matter. While the `AlwaysMatcher` implementation is here as a trivial example, it should be the first step towards matchers use in Rust as it is currently the only supported one. Notable changes: - `exact` is renamed to `exact_match` - enums for `visit*` methods with `Recursive` instead of `'all'`, etc. - a new `roots`, separate from `file_set` - no `bad`, `explicitdir` or `traversedir` functions as they can be passed to the high functions instead of the matchers Thanks to Martin for suggesting the last two (most important) changes and for reaching out to help a few weeks ago. Differential Revision: https://phab.mercurial-scm.org/D7178

merge with stable

perf: add a way to benchmark `dirstate.status` Getting more details about time spend in this specific internal bit is meaningful.

largefiles: delete obsolete and unused repo.push() The function was removed from localrepo in 4d52e6eb98ea (locarepo: remove the `push` method (API), 2014-09-25). Differential Revision: https://phab.mercurial-scm.org/D7174

censor: document that some commands simply ignore censored data I can't see a benefit for hg grep to ever error out early when it encounters censored data.

grep: warn on censored revisions instead of erroring out We need most of the grep logic to go through in case we encounter a censored revision, so we just return a None body for a censored node, and we stop just short of trying to record matches with the contents of that censored body. The other parts such as recording that the censored file has been considered at this revision needs to go into the proper dicts. I have also gotten weary of all the abbreviations, so while I did a small refactor to move the file-data-getting operation into a common function, I also expanded the abbreviations of the relevant variables within this little function. Hopefully some day this helps someone figure out what all the abbreviations mean. Although the censoring docs currently state that some commands error out or are ignored depending on the `censor.policy` config, I cannot see a benefit for grep to ever stop dead in its tracks when a censored revision is encountered. I will also amend the docs to indicate that some commands, such as grep, unconditionally ignore censored revisions.

py3: enable legacy fs encoding to fix filename compatibility on Windows This patch is untested. I just followed the instruction: https://docs.python.org/3/whatsnew/3.6.html#pep-529-change-windows-filesystem-encoding-to-utf-8

rust-cpython: run cargo fmt

rust-cpython: remove useless PyResult<> from leak_immutable() The caller should know if the shared data is mutably borrowed or not.

rust-cpython: remove useless PyRefMut wrapper

rust-cpython: drop manual management of mutably_borrowed RefCell::borrow() should guarantee there's no mutable borrow. As a follow up, maybe PySharedState can be a pure data structure + function. Most ref-sharing business has already been moved to PySharedRef* and PyLeaked*.

rust-cpython: leverage RefCell::borrow() to guarantee there's no mutable ref Since the underlying value can't be mutably borrowed by PyLeaked, we don't have to manage yet another mutably-borrowed state. We can just rely on the RefCell implementation. Maybe we can add try_leak_immutable(), but this patch doesn't in order to keep the patch series not too long.

rust-cpython: remove useless Option<$leaked> from py_shared_iterator We no longer need to carefully drop the iterator when it's consumed. Mutation is allowed even if the iterator exists. There's a minor behavior change: next(iter) may return/raise something other than StopIteration if it's called after the iterator has been fully consumed, and if the Rust object isn't a FusedIterator.

rust-cpython: allow mutation unless leaked reference is borrowed In other words, mutation is allowed while a Python iterator holding PyLeaked exists. The iterator will be invalidated instead. We still need a borrow_count to prevent mutation while leaked data is dereferenced in Rust world, but most leak_count business is superseded by the generation counter. decrease_leak_count(py, true) will be removed soon.

rust-cpython: add generation counter to leaked reference This counter increments on borrow_mut() to invalidate existing leaked references. This is modeled after the iterator invalidation in Python. The other checks will be adjusted by the subsequent patches.

rust-cpython: add stub wrapper that'll prevent leaked data from being mutated In order to allow mutation of PySharedRefCell value while PyLeaked reference exists, we need yet another "borrow" scope where mutation is prohibited. try_borrow<'a> and try_borrow_mut<'a> defines the "borrow" scope <'a>. The subsequent patches will implement leak counter based on this scope. PyLeakedRef<T> and PyLeakedRefMut<T> could be unified to PyLeakedRef<&T> and PyLeakedRef<&mut T> respectively, but I didn't do that since it seemed a bit weird that deref_mut() would return a mutable reference to an immutable reference.

rust-cpython: rename PyLeakedRef to PyLeaked This series will make PyLeaked* behave more like a Python iterator, which means mutation of the owner object will be allowed and the leaked reference (i.e. the iterator) will be invalidated instead. I'll add PyLeakedRef/PyLeakedRefMut structs which will represent a "borrowed" state, and prevent the underlying value from being mutably borrowed while the leaked reference is in use: let shared = self.inner_shared(py); let leaked = shared.leak_immutable(); { let leaked_ref: PyLeakedRef<_> = leaked.borrow(py); shared.borrow_mut(); // panics since the underlying value is borrowed } shared.borrow_mut(); // allowed The relation between PyLeaked* structs is quite similar to RefCell/Ref/RefMut, but the implementation can't be reused because the borrowing state will have to be shared across objects having no lifetime relation. PyLeaked isn't named as PyLeakedCell since it isn't actually a cell in that leaked.borrow_mut() will require &mut self.

py3: don't use bytes with vars() or __dict__ Inspired by D7227. These were all the remaining instances I could find. Differential Revision: https://phab.mercurial-scm.org/D7230

Added signature for changeset ca3dca416f8d

Added tag 5.2 for changeset ca3dca416f8d

py3: add inline comment about encoding issue of str(Abort())

py3: do not reimplement Abort.__str__() on Python 2 It isn't necessary on Python 2, and the default implementation should be better than our BaseException_str() clone.

tests: write out file using bytes I/O The encoding of sys.stdout varies between Python versions. So using a one-liner to write a file from a Unicode string is not deterministic. This commit writes out the file using bytes I/O to ensure we have exactly the bytes we want in the file. This change fixes a test failure in Python 3.5/3.6. Differential Revision: https://phab.mercurial-scm.org/D7226

import-checker: open all source files as utf-8 Before, we opened in text mode and used the default encoding to interpret the bytes within. This caused problems interpreting some byte sequences in some files. This commit changes things to always open files as UTF-8, which makes the error go away. test-check-module-imports.t now passes on Python 3.5 and 3.6 with this change. Differential Revision: https://phab.mercurial-scm.org/D7225

localrepo: use str for lookup in vars() vars() returns a dict of str. So always use a native str for the key lookup. Differential Revision: https://phab.mercurial-scm.org/D7227

automation: install python3-venv Debian package Debian's python install has a crippled venv by default, as it is lacking ensurepip. When you try to run `python3 -m venv` it tells you to install `python3-venv`. So this commit does that in our automation environment so we can fully test installing Mercurial using venv+pip with the system Python. Differential Revision: https://phab.mercurial-scm.org/D7229

tests: look for ensurepip before using venv Debian appears to cripple the venv module by default by removing the associated ensurepip functionality. (The module isn't present at all.) This caused test-install.t to fail when using the Debian python3 unless the python3-venv package was installed. This commit introduces a new hghave requirement for detecting ensurepip and makes the Python 3 install variant conditional on its presence. This should make test-install.t pass when using an incomplete Debian Python. Differential Revision: https://phab.mercurial-scm.org/D7228

automation: avoid '~' in the temp directory on Windows If a long-ish username is used, the environment variable ends up with a '~' to be 8.3 path compatible. That in turn causes a handful of tests (mostly ssh related) to add quotes around $TESTMP. I have no AWS experience, so I have no idea if this is the proper way to do it. But I've hit this problem locally, and redirecting the directory is a workaround. I don't recall if the directory is created on demand by the test harness, but presumably if this is configured before the machine boots, Windows will do it for us. Differential Revision: https://phab.mercurial-scm.org/D7130

tests: use venv on Python 3 This test was failing in some Python 3 environments because `$PYTHON -m virtualenv` was somehow resulting in Python 2 being used. Why, I'm not sure. Python 3 includes virtualenv in the standard library as the `venv` module. This commit changes test-install.t to use `$PYTHON -m venv` on Python 3 and `$PYTHON -m virtualenv` on Python 2 (if available). I chose to make some test output duplicated because we can't have nested conditionals and there is no easy way to express ORing of hghave checks. Differential Revision: https://phab.mercurial-scm.org/D7224

tests: remove HGALLOWPYTHON3 reference This variable was removed from setup.py in c3e10f705a6c. Differential Revision: https://phab.mercurial-scm.org/D7223

run-tests: use byte strings for inserted output We were inserting str on Python 3 which resulted in mixed str/bytes types on the list. This would later blow up when trying to write str to the .err file opened in bytes mode. Differential Revision: https://phab.mercurial-scm.org/D7222

contrib: require Python 3.7 for byteify-strings.py bb509f39d387 made an error, it's actually 3.7 that introduced token.COMMENT. Differential Revision: https://phab.mercurial-scm.org/D7220

hghave: fix bytes/string issue on Python 3

packaging: add support for CentOS 8 The resulting executable has not been tested in detail yet. I ran 'hg version' and 'hg clone', which worked fine (except for extensions acting up due to Python 3). Differential Revision: https://phab.mercurial-scm.org/D7216

packaging: allow choosing python version depending on centos version Differential Revision: https://phab.mercurial-scm.org/D7217

fsmonitor: use stringutil.forcebytestr() instead of str() on an exception Similar to 5fa8ac91190e / D7206, should get test-install.t passing on py3. Differential Revision: https://phab.mercurial-scm.org/D7218

py3: add a __str__ method to Abort This improves the rendering of some exceptions by avoiding raw bytestrings, especially when using --traceback option.

py3: add Python 3 exception output to test-lfs-serve-access.t Similar to a973a75e92bf or 3e9c6cef949b.

automation: install black This should unblock us from running the code formatting test in our automated environment. Differential Revision: https://phab.mercurial-scm.org/D7197

py3: use %d to format an int Avoids a TypeError under py3. Fortunately this is very much an edge case since it requires the user to have deliberately created a local tag of the form 'D\d+' that isn't truthful. Differential Revision: https://phab.mercurial-scm.org/D7215

fsmonitor: normalize exception types to bytes Unavailable.msg should now always be bytes. We also rename Unavailable.__str__ to __bytes__ as it always returns bytes. We make __str__ a simple wrapper that decodes that result to str. There's probably some excessive strutil.forcebytestr() in fsmonitor/__init__.py now. But at least the exceptions around type coercion should now be gone. Differential Revision: https://phab.mercurial-scm.org/D7214

fsmonitor: normalize clock value to bytes We normalize the value returned by watchman because we perform a number of compares with this value in code. So the easiest path forward is to normalize to bytes so we don't have to update many call sites. With this commit, the fsmonitor extension appears to be working with Python 3! Although there are still some failures in edge cases... Differential Revision: https://phab.mercurial-scm.org/D7213

fsmonitor: use next() instead of .next() This is needed for Python 3 compatibility. Differential Revision: https://phab.mercurial-scm.org/D7212

fsmonitor: normalize Watchman paths to bytes Otherwise it will be a str on Python 3 and operations below which operate in the bytes domain will fail. Differential Revision: https://phab.mercurial-scm.org/D7211

fsmonitor: handle unicode keys in tuples In Python 3, keys in the bset tuple are typically str, not bytes. PyBytes_AsString() would return NULL. But we weren't checking the return value and this would lead to a segfault. This commit makes the code type and Python version aware. The Python version specific code is to allow us to utilize a modern API for converting str -> char* without having to allocate an extra PyObject. FWIW I wanted to assume that keys were always str. However, there appear to be some bytes keys in some cases. I haven't debugged this further. Differential Revision: https://phab.mercurial-scm.org/D7210

fsmonitor: make _hashignore compatible with Python 3 The Hasher wants a bytes but we were feeding it a str. Let's use our repr() implementation to return bytes. In addition, the hexdigest() would return a str, which would be compared against a bytes and would always fail. Normalize to bytes so the compare works. Differential Revision: https://phab.mercurial-scm.org/D7209

fsmonitor: normalize hostname to bytes Without this, we get a str/bytes mismatching when using % formatting a few lines below. Differential Revision: https://phab.mercurial-scm.org/D7208

fsmonitor: access repo.root There is no repo._root. It looks like fsmonitor has been busted since this access was introduced in ab1900323b1 in July 2019! Differential Revision: https://phab.mercurial-scm.org/D7207

fsmonitor: coerce watchman exception to bytes Without this, we get errors due to passing str to a function which expects bytes. Differential Revision: https://phab.mercurial-scm.org/D7206

fsmonitor: fix str/bytes mismatch when accessing watchman version There were 2 bugs here. First, keys in the tuple are always str. Second, we needed to normalize the value to bytes to prevent a str/bytes mismatch on Python 3. With this commit, `hg debuginstall` with fsmonitor enabled now works on Python 3. Differential Revision: https://phab.mercurial-scm.org/D7205

fsmonitor: reapply b1f62cd39b5c The recent revendoring of pywatchman undid this changeset. Let's reapply it. This commit was generated by running `hg graft -f b1f62cd39b5c`. It applied cleanly. Differential Revision: https://phab.mercurial-scm.org/D7204

fsmonitor: reapply dd35abc409ee The recent revendoring of pywatchman undid this bug fix. Let's reapply it. Differential Revision: https://phab.mercurial-scm.org/D7203

fsmonitor: remove pywatchman from exclusion rule The recently vendored pywatchman code base is now formatted with black. We can now remove pywatchman from our black exclusion rule. Differential Revision: https://phab.mercurial-scm.org/D7202

fsmonitor: refresh pywatchman with upstream This commit vendors pywatchman commit 259dc66dc9591f9b7ce76d0275bb1065f390c9b1 from upstream without modifications. The previously vendored pywatchman from changeset 16f4b341288d was from Git commit c77452. This commit effectively undoes the following Mercurial changesets: * dd35abc409ee fsmonitor: correct an error message * b1f62cd39b5c fsmonitor: layer on another hack in bser.c for os.stat() compat (issue5811) * c31ce080eb75 py3: convert arguments, cwd and env to native strings when spawning subprocess * 876494fd967d cleanup: delete lots of unused local variables * 57264906a996 watchman: add the possibility to set the exact watchman binary location The newly-vendored code has support for specifying the binary location, so 57264906a996 does not need applied. But we do need to modify our code to specify a proper argument name. 876494fd967d is not important, so it will be ignored. c31ce080eb75 globally changed the code base to always pass str to subprocess. But pywatchman's code is Python 3 clean, so we don't need to do this. This leaves dd35abc409ee and b1f62cd39b5c, which will be re-applied in subsequent commits. Differential Revision: https://phab.mercurial-scm.org/D7201