packaging: modernize (compat PEP 517) with less distutils and setup.py calls - setup.py: less distutils imports and setuptools required distutils is deprecated and one should import commands from setuptools to support modern workflows depending on PEP 517 and 518. Moreover, for Python >=3.12, distutils comes from setuptools. It corresponds to old and unmaintain code that do not support PEP 517. The PEP 517 frontends (pip, build, pipx, PDM, UV, etc.) are responsible for creating a venv just for the build. The build dependencies (currently only setuptools) are specified in the pyproject.toml file. Therefore, there is no reason to support building without setuptools. Calling directly setup.py is deprecated and we have to use a PEP 517 frontend. For this commit we use pip with venv. - run-tests.py: install with pip instead of direct call of setup.py Mercurial is then built in an isolated environment. - Makefile: use venv+pip instead of setup.py

setup-py: remove unused and deprecated setup argument datafiles

setup: skip building rust extensions on Windows Without this, the build process in the next commit would have started trying (and failing) to copy the associated module. Not sure why this worked before. I'd like to think there's a better way to conditionalize this inside the extension declaration. [snip] done updating hg.exe's manifest copying build\lib.win-amd64-cpython-39\mercurial\cext\base85.cp39-win_amd64.pyd -> mercurial\cext copying build\lib.win-amd64-cpython-39\mercurial\cext\bdiff.cp39-win_amd64.pyd -> mercurial\cext copying build\lib.win-amd64-cpython-39\mercurial\cext\mpatch.cp39-win_amd64.pyd -> mercurial\cext copying build\lib.win-amd64-cpython-39\mercurial\cext\parsers.cp39-win_amd64.pyd -> mercurial\cext copying build\lib.win-amd64-cpython-39\mercurial\cext\osutil.cp39-win_amd64.pyd -> mercurial\cext copying build\lib.win-amd64-cpython-39\mercurial\thirdparty\sha1dc.cp39-win_amd64.pyd -> mercurial\thirdparty copying build\lib.win-amd64-cpython-39\hgext\fsmonitor\pywatchman\bser.cp39-win_amd64.pyd -> hgext\fsmonitor\pywatchman error: can't copy 'build\lib.win-amd64-cpython-39\mercurial\rustext.cp39-win_amd64.pyd': doesn't exist or not a regular file

setup-py: weaken build isolation when calling hg See https://github.com/pypa/setuptools-scm/blob/5d3a1f88382c7106d009a0731df97330e4cf5f7c/src/setuptools_scm/_run_cmd.py#L105

git: skip recursing into manifest subtrees that are the same If during manifest comparison we are presented with identical tree objects, then we are guaranteed that: 1. the contents of these tree objects are the same 2. the contents of any of their children are the same This speeds up 'hg diff -c <hash>' considerably since unchanged subtrees are are skipped. For example, a commit that modified the README in the linux kernel repo (~84k tracked files) now takes 1.1s instead of 15.7s.

tests: obfuscate Python version checks to subvert `pyupgrade` When specifying `--py38-plus`, it reasonably deletes conditionals that check for that version of python or earlier. But we actually want the check to remain, so use a simple obfuscation to preserve it.

tests: drop py2 support for `fsmonitor-run-tests.py` There's some py2 compat code still in the fsmonitor extension, but I don't care about that and this doesn't affect that.

pyupgrade: drop redundant `''.encode("utf-8")` Again, I'd prefer explicit. But without a way to selectively choose what fixes are applied by `_fix_tokens()`, I'll accept this for the sake of reducing the nuisance output when running `pyupgrade`.

pyupgrade: drop redundant `open()` modes On one hand, this gets rid of some py2 cruft (like `U` and `t` modes). OTOH, I like being explict. But there's no way to disable individual transformations (and this change is part of `_fix_tokens()`, which is even harder to selectively use locally), and I'd rather have less noise when using `pyupgrade`. Warning: `contrib/synthrepo.py` appears not to have been updated to py3, as it's passing str to `error.Abort`. But since it was opening the file in unicode on py2, this change is a no-op.

pyupgrade: convert `str.format()` to f-strings This is part of the `_fix_tokens()` changes made by `pyupgrade` once py36 is the minimum.

pyupgrade: drop `coding=UTF-8` comments PEP-3120[1] (Python 3.0 in 2007) says that UTF-8 is the default encoding. That should be long enough ago that no reasonable editor would trip over this, and certainly any supported version of Python won't. The comments were probably harmless, but as `pyupgrade` has no mechanism to disable this change, omitting this change makes it unusable as a code checking tool, and makes it a pain to use occasionally to upgrade the source (since these changes would need to be manually reverted). [1] https://peps.python.org/pep-3120/

pyupgrade: primitive types directly where possible This was rewritten by the `type_of_primitive` fixer in `pyupgrade`. Interesting that 9db77d46de79 missed a bunch of these `u''` prefixes.

pyupgrade: rewrite `' '.join(shlex.quote(x) for x in ...)` with `shlex.join()` This was rewritten by the `shlex_join` fixer in `pyupgrade`. This is new in Python 3.8.

pyupgrade: use set literals where possible This was rewritten by the `set_literals` fixer in `pyupgrade`. I thought there was a past change around using set literals, but I can't find it now.

pyupgrade: replace `io.open()` with `open()` This was rewritten by the `io_open` fixer in `pyupgrade`.

pyupgrade: replace `dict(...)` with dict comprehensions This was rewritten by the `dict_literals` fixer in `pyupgrade`. The last pass over the code for this that I could find was a61ed1c2d7a7.

pyupgrade: replace unnecessary lambdas in `collections.defaultdict()` calls This was rewritten by the `defaultdict_lambda` fixer in `pyupgrade`.

pyupgrade: drop a few redundant types in `isinstance` lists This was rewritten by the `constant_fold` fixer in `pyupgrade`. I suspect that one of these `int`s were `long` in the py2 days, and some automated py2 -> py3 tool wasn't smart enough to elide these.

pyupgrade: drop a usage of `typing.Text` This is the `typing_text` fixer in `pyupgrade`. This type was meant to be for py2/py3 interoperability, being bytes on py2 and str on py3.

pyupgrade: drop the quoting around type annotations This is the `typing_pep563` fixer in `pyupgrade`. Quoting to delay evaluation hasn't been necessary since adding `from __future__ import annotations` in 1c5810ce737e.