dispatch: use detailed exit code 250 for keyboard interrupt Among our users at Google, we're still seeing several percent of commands fail with exit code 255. I suspect keyboard interrupts is an important remaining reason. This is a resend of D10086 with some fixes for pager handling added ahead of it. Differential Revision: https://phab.mercurial-scm.org/D11628

dispatch: ignore failure to flush ui When the pager dies, we get a `SIGPIPE`. That causes `error.SignalInterrupt` to be raised ` (from `ui._catchterm()`). Any further writes or flushes will cause further `SIGPIPE`s and furhter `error.SignalInterrupt`. If we write or flush outside of the try/except that handle `KeyboardInterrupt` (which `error.SignalInterrupt` is a subclass of), then control will escape from the `dispatch` module. Let's fix that by ignoring errors from flushing the ui. I would have rather fixed this by restoring the stdout and stderr streams when the pager dies, but it gets complicated because of multiple ui instances (ui/lui) and different pager setups between regular hg and chg. This changes a test in `test-pager.t`, but I don't understand why. I would have thought that all the output from the command should have gone to the broken pager. Differential Revision: https://phab.mercurial-scm.org/D11627

dispatch: don't change error status if flushing stdio fails If we already have a non-zero exit code, I don't think we should change it to 255 because we fail to flush stdio. This may not matter yet, but it will matter when I make a killed pager result in exit code 250 (it's currently 255). Differential Revision: https://phab.mercurial-scm.org/D11626

dirstate-v2: Use "byte sequence" in docs The patch originally sent as https://phab.mercurial-scm.org/D11546 used "byte string" but that was changed during review to avoid suggesting Unicode or character encodings. However "byte range" sounds to be like a range of *indices* within a byte string/sequence elsewhere. This changes to "byte sequence". Python docs use "sequence" a lot when discussing the `bytes` type: https://docs.python.org/3/library/stdtypes.html Differential Revision: https://phab.mercurial-scm.org/D11623

rust: Make the hg-cpython crate default to Python 3 This default is used when running `cargo` manually such as for `cargo test`. `setup.py` and `Makefile` both configure the Python major version explicitly. Differential Revision: https://phab.mercurial-scm.org/D11618

packaging: update the certifi dependency Not sure if this helps with recent certificate issues[1], but we might as well keep this modern. Note that certifi no longer claims py2 support, and there's a PR to add it back in[2]. Py2 support was dropped in 2020.04.05.2 (which predates what's being updated here). None of the *.py files have changed since the 2020.6.20 release, and I was able to call `certifi.where()` in both a virtualenv and with `hg debugshell` from an MSI install with this version, so we shouldn't be any worse off than before. [1] https://foss.heptapod.net/mercurial/tortoisehg/thg/-/issues/5745 [2] https://github.com/certifi/python-certifi/pull/154 Differential Revision: https://phab.mercurial-scm.org/D11609

setup: stop packaging python3.dll and python3X.dll in the wheel distribution Now that exewrapper is smart enough to find the DLLs it needs without help from the build script, backout ed286d150aa8 and 2960b7fac966. Note that this will require deleting the build/lib.win-amd64-3.X directory to actually remove it from the final wheel. Differential Revision: https://phab.mercurial-scm.org/D11455

exewrapper: find the proper python3X.dll in the registry Previously, we relied on the default library lookup[1], which for us is essentially to look on `PATH`. That has issues- the Python installations are not necessarily on `PATH`, so I started copying the DLLs locally in 2960b7fac966 and ed286d150aa8 during the build to work around that. However, it's been discovered that causes `python3.dll` and `python3X.dll` to get slipped into the wheel that gets distributed on PyPI. Additionally, Mercurial would fail to run in a venv if the Python environment that created it isn't on `PATH`, because venv creation doesn't copy the DLLs locally. The logic here is inspired by the `py.exe` launcher[2], though this is simpler because we don't care about the architecture- if this is a 32 bit process running on Win64, the registry reflection will redirect to where the 32 bit Python process wrote its keys. A nice unintended side effect is to also make venvs that don't have their root Python on `PATH` work without all of the code required to read `pyvenv.cfg`[3]. I don't see any reasonable way to create a venv without Python being installed (other than maybe building Python from source?), so punt on trying to read that file for now and save a bunch of string manipulation code. I somehow managed to corrupt my Windows user profile, and that makes the Microsoft Store python not run (even loading the DLL gives an access error), so I'm giving priority to both global and user specific python.org installations. Loading python3.dll is new, but when I went down the rabbit hole of implementing `pyvenv.cfg` support, I saw a comment[4] that led me to think we could have trouble if we don't. The comment in ed286d150aa8 confirms this, so we should probably bail out completely if Python3 can't be loaded from the registry, rather than getting something random on `PATH`. But I'll leave that for the default branch. [1] https://docs.microsoft.com/en-us/windows/win32/Dlls/dynamic-link-library-search-order#standard-search-order-for-desktop-applications [2] https://github.com/python/cpython/blob/adcd2205565f91c6719f4141ab4e1da6d7086126/PC/launcher.c#L249 [3] https://github.com/python/cpython/blob/bb3e0c240bc60fe08d332ff5955d54197f79751c/PC/getpathp.c#L707 [4] https://github.com/python/cpython/blob/bb3e0c240bc60fe08d332ff5955d54197f79751c/PC/getpathp.c#L1098 Differential Revision: https://phab.mercurial-scm.org/D11454

fix: reduce number of tool executions By grouping together (path, ctx) pairs according to the inputs they would provide to fixer tools, we can deduplicate executions of fixer tools to significantly reduce the amount of time spent running slow tools. This change does not handle clean files in the working copy, which could still be deduplicated against the files in the checked out commit. It's a little harder to do that because the filerev is not available in the workingfilectx (and it doesn't exist for added files). Anecdotally, this change makes some real uses cases at Google 10x faster. I think we were originally hesitant to do this because the benefits weren't obvious, and implementing it efficiently is kind of tricky. If we simply memoized the formatter execution function, we would be keeping tons of file content in memory. Also included is a regression test for a corner case that I broke with my first attempt at optimizing this code. Differential Revision: https://phab.mercurial-scm.org/D11280

fix: add test to demonstrate how many times tools are executed The current implementation wastes a lot of effort invoking fixer tools more than once on the same file name+content. It is good to track changes in this behavior. Differential Revision: https://phab.mercurial-scm.org/D11279