py3: make stdout line-buffered if connected to a TTY Status messages that are to be shown on the terminal should be written to the file descriptor before anything further is done, to keep the user updated. One common way to achieve this is to make stdout line-buffered if it is connected to a TTY. This is done on Python 2 (except on Windows, where libc, which the CPython 2 streams depend on, does not properly support this). Python 3 rolls it own I/O streams. On Python 3, buffered binary streams can't be set line-buffered. The previous code (added in 227ba1afcb65) incorrectly assumed that on Python 3, pycompat.stdout (sys.stdout.buffer) is already line-buffered. However the interpreter initializes it with a block-buffered stream or an unbuffered stream (when the -u option or the PYTHONUNBUFFERED environment variable is set), never with a line-buffered stream. One example where the current behavior is unacceptable is when running `hg pull https://www.mercurial-scm.org/repo/hg` on Python 3, where the line "pulling from https://www.mercurial-scm.org/repo/hg" does not appear on the terminal before the hg process blocks while waiting for the server. Various approaches to fix this problem are possible, including: 1. Weaken the contract of procutil.stdout to not give any guarantees about buffering behavior. In this case, users of procutil.stdout need to be changed to do enough flushes. In particular, 1. either ui must insert enough flushes for ui.write() and friends, or 2. ui.write() and friends get split into flushing and fully buffered methods, or 3. users of ui.write() and friends must flush explicitly. 2. Make stdout unbuffered. 3. Make stdout line-buffered. Since Python 3 does not natively support that for binary streams, we must implement it ourselves. (2.) is problematic because using unbuffered I/O changes the performance characteristics significantly compared to line-buffered (which is used on Python 2) and this would be a regression. (1.2.) and (1.3) are a substantial amount of work. It’s unclear whether the added complexity would be justified, given that raw performance doesn’t matter that much when writing to a terminal much faster than the user could read it. (1.1.) pushes complexity into the ui class instead of separating the concern of how stdout is buffered. Other users of procutil.stdout would still need to take care of the flushes. This patch implements (3.). The general performance considerations are very similar to (1.1.). The extra method invocation and method forwarding add a little more overhead if the class is used. In exchange, it doesn’t add overhead if not used. For the benchmarks, I compared the previous implementation (incorrect on Python 3), (1.1.), (3.) and (2.). The command was chosen so that the streams were configured as if they were writing to a TTY, but actually write to a pager, which is also the default: HGRCPATH=/dev/null python3 ./hg --cwd ~/vcs/mozilla-central --time --pager yes --config pager.pager='cat > /dev/null' status --all previous: time: real 7.880 secs (user 7.290+0.050 sys 0.580+0.170) time: real 7.830 secs (user 7.220+0.070 sys 0.590+0.140) time: real 7.800 secs (user 7.210+0.050 sys 0.570+0.170) (1.1.) using Yuya Nishihara’s patch: time: real 9.860 secs (user 8.670+0.350 sys 1.160+0.830) time: real 9.540 secs (user 8.430+0.370 sys 1.100+0.770) time: real 9.830 secs (user 8.630+0.370 sys 1.180+0.840) (3.) using this patch: time: real 9.580 secs (user 8.480+0.350 sys 1.090+0.770) time: real 9.670 secs (user 8.480+0.330 sys 1.170+0.860) time: real 9.640 secs (user 8.500+0.350 sys 1.130+0.810) (2.) using a previous patch by me: time: real 10.480 secs (user 8.850+0.720 sys 1.590+1.500) time: real 10.490 secs (user 8.750+0.750 sys 1.710+1.470) time: real 10.240 secs (user 8.600+0.700 sys 1.590+1.510) As expected, there’s no difference on Python 2, as exactly the same code paths are used: previous: time: real 6.950 secs (user 5.870+0.330 sys 1.070+0.770) time: real 7.040 secs (user 6.040+0.360 sys 0.980+0.750) time: real 7.070 secs (user 5.950+0.360 sys 1.100+0.760) this patch: time: real 7.010 secs (user 5.900+0.390 sys 1.070+0.730) time: real 7.000 secs (user 5.850+0.350 sys 1.120+0.760) time: real 7.000 secs (user 5.790+0.380 sys 1.170+0.710)

simplemerge: rewrite flag merging loop as expression I feel binary operations are more readable.

simplemerge: leverage pycompat function to convert byte string to set

simplemerge: fix function name that tests if ctx is not null revision

git: decode node IDs back into Python strings (issue6349) db.text_factory = bytes, so the database contains only strings. The object IDs we get from pygit2 are Python strings. b'foo' != 'foo' This change allows the "don't reindex" optimization to work by allowing the "cur_cache_heads == cache_heads" comparison a few lines down to succeed. Differential Revision: https://phab.mercurial-scm.org/D8622

phabricator: make it clear what happen when no response Differential Revision: https://phab.mercurial-scm.org/D8621

tests: make it clear what happen when no response entered Differential Revision: https://phab.mercurial-scm.org/D8620

localrepo: handle ValueError during repository opening Python 3.8 can raise ValueError on attempt of an I/O operation against an illegal path. This was causing test-remotefilelog-gc.t to fail on Python 3.8. This commit teaches repository opening to handle ValueError and re-raise an Abort on failure. An arguably better solution would be to implement this logic in the vfs layer. But that seems like a bag of worms and I don't want to go down that rabbit hole. Until users report uncaught ValueError exceptions in the wild, I think it is fine to patch this at the only occurrence our test harness is finding it. Differential Revision: https://phab.mercurial-scm.org/D7944

metadata: filter the `removed` set to only contains relevant data The `files` entry can be bogus and contains too many entries. This can badly combines with the computation of `removed` inflating the set size. The can lead to the changesets centric rename computation to process much more data than needed, slowing it down (and increasing space taken by data storage). In practice newer commits already that reduced set, this applies this "fix" to older changeset. Differential Revision: https://phab.mercurial-scm.org/D8589

files: extract code for extra filtering of the `removed` entry into copies We want to reduce the set of `removed` files that to the set of files actually removed. That `removed` set is used as of the changeset centric algorithm, having smaller sets means less processing and faster computation. In this changeset we extract the code to be a function of it own. We will make use of it in the next changesets. Differential Revision: https://phab.mercurial-scm.org/D8588

metadata: move computation related to files touched in a dedicated module This was suggested by Yuya Nishihara a while back. Since I am about to add more metadata related computation, lets create a new repositories. Differential Revision: https://phab.mercurial-scm.org/D8587

merge: move an inspection of the dirstate from record to calculate phase The intent is clearly to have `calculateupdates()` figure out what actions need to be taken and `recordupdates()` to make necessary modifications to the dirstate. However, in the `ACTION_PATH_CONFLICT_RESOLVE` case, there was one little inspection of copy information done in `recordupdates()`. This patch moves that to `calculateupdates()`. That will help with the next patch, which makes `merge.update()` work better with `overlayworkingctx` (copies should be recorded there too, even though we skip the `recordupdates()` step). Differential Revision: https://phab.mercurial-scm.org/D8615

tests: adjust to the new format in pyflakes output According to the pyflakes' NEWS.rst, the default output format changed recently: 2.2.0 (2020-04-08) - Include column information in error messages So the lines now read: contrib/perf.py:149:15 undefined name 'xrange' mercurial/hgweb/server.py:427:13 undefined name 'reload' mercurial/util.py:2862:24 undefined name 'file' Differential Revision: https://phab.mercurial-scm.org/D8618

tests: consistently use pyflakes as a Python module We check availability of pyflakes as a module, and also running it for real as a module. Only fair to test filterpyflakes.py working correctly when using pyflakes as a module too. Differential Revision: https://phab.mercurial-scm.org/D8617

heptapod-ci: also run tests for chg on python 2 I am not aware of any regular effort to run test with chg. So let's at least do it here. Differential Revision: https://phab.mercurial-scm.org/D8178

rust-dependencies: upgrade `micro-timer` dependency I wanted to to a tour of dependencies to upgrade, but only `micro-timer` has a new release which does not print when the function panics, which should be less misleading. Differential Revision: https://phab.mercurial-scm.org/D8605

context: fix creation of ProgrammingError to not use non-existent field Differential Revision: https://phab.mercurial-scm.org/D8606

help: explain in `hg help flags` that unambiguous prefixes are allowed I used `hg commit --amend` as an example because that's the most frequently used flag I could think of that doesn't yet have a short form. Differential Revision: https://phab.mercurial-scm.org/D8607

git: add coverage for manifest.diff() so we don't regress

git: implement diff manifest method This makes 'hg diff' work.

git: properly visit child tree objects when resolving a path

git: don't yield paths for directories when walking

git: correctly check for type of object when walking

git: implement stub prefetch_parents dirstate method A recent change (35b255e474d9) introduced this new required dirstate method but didn't update the git extension.

rust: leverage .expect() in place of .unwrap() + inline comment For a better error indication in case we made a mistake.

rust: fix false comment about mpsc::Sender We need Sync to share the Sender reference across threads.

merge with stable

relnotes: advertize the possibility to use rust I think the rust work may have been mentioned in the release notes, but if so only in passing, and not as an invitation to try it out. I think the next version is a decent time to do this, because the rust doesn't come with performance regressions AFAIK, speeds up status noticeably when it applies, which is the case for most invocations of status, and doesn't have the undesirable restriction of regex around empty patterns anymore. I am cheating a bit, because I'm giving numbers for `hg status` in mozilla-central, but they have one hgignore pattern that uses lookaround, ".vscode/(?!extensions\.json|tasks\.json", which I took out as it would cause a fallback to python when unknown files are requested. But it seems that they could express their hgignore differently if they were so inclined. Not sure if there are limitation other than linux-only that I am not thinking of but would be worth mentioning upfront, to avoid disappointing users? Differential Revision: https://phab.mercurial-scm.org/D8604

rust: add a pointer for profiling to the README As figuring out how to get useful profiles is not obvious. Differential Revision: https://phab.mercurial-scm.org/D8603

rust: update the mention of hgcli in rust/README.rst This may not be exactly right, but it's better than before. Differential Revision: https://phab.mercurial-scm.org/D8602

sslutil: fix comment to use inclusive or instead of exclusive or The incorrect "either" was introduced by one of my recent patches.

sslutil: propagate return value ssl.PROTOCOL_SSLv23 from protocolsettings() Also, protocolsettings() was renamed to commonssloptions() to reflect that only the options are returned.

sslutil: stop storing protocol and options for SSLContext in settings dict Call protocolsettings() where its return values are needed.

sslutil: rename 'minimumprotocolui' -> 'minimumprotocol' Before, both 'minimumprotocolui' and 'minimumprotocol' were used, but meaning the same.

sslutil: properly detect which TLS versions are supported by the ssl module For the record, I contacted the CPython developers to remark that unconditionally defining ssl.PROTOCOL_TLSv1_1 / ssl.PROTOCOL_TLSv1_2 is problematic: https://github.com/python/cpython/commit/6e8cda91d92da72800d891b2fc2073ecbc134d98#r39569316

sslutil: remove dead code (that failed if only TLS 1.0 is available) We ensure in setup.py that TLS 1.1 or TLS 1.2 is present.

config: remove unused hostsecurity.disabletls10warning config

sslutil: remove dead code (that downgraded default minimum TLS version) We ensure in setup.py that TLS 1.1 or TLS 1.2 is present.

sslutil: remove comment referring to unsupported legacy stacks

setup: require that Python has TLS 1.1 or TLS 1.2 This ensures that Mercurial never downgrades the minimum TLS version from TLS 1.1+ to TLS 1.0+ and enables us to remove that compatibility code. It is reasonable to expect that distributions having Python 2.7.9+ or having backported modern features to the ssl module (which we require) have a OpenSSL version supporting TLS 1.1 or TLS 1.2, as this is the main reason why distributions would want to backport these features. TLS 1.1 and TLS 1.2 are often either both enabled or both not enabled. However, both can be disabled independently, at least on current Python / OpenSSL versions. For the record, I contacted the CPython developers to remark that unconditionally defining ssl.PROTOCOL_TLSv1_1 / ssl.PROTOCOL_TLSv1_2 is problematic: https://github.com/python/cpython/commit/6e8cda91d92da72800d891b2fc2073ecbc134d98#r39569316

sslutil: check for OpenSSL without TLS 1.0 support in one case It can only happen if supportedprotocols gets fixed to contain only correct items (see the FIXME above in the file).

sslutil: don't set minimum TLS version to 1.0 if 1.2 but not 1.1 is available This case isn't very likely, but possible, especially if supportedprotocols gets fixed to contain only correct items (see the FIXME above in the file).

sslutil: add FIXME about supportedprotocols possibly containing too many items

sslutil: fix names of variables containing minimum protocol strings When working in this module, I found it very confusing that "protocol" as a variable name could mean either "minimum protocol string" or an exact version (as a string or ssl.PROTOCOL_* value). This patch prefixes variables of the former type with "minimum".

sslutil: stop returning argument as third return value of protocolsettings() The third return value was always the same as the argument.

relnotes: note that we now require modern SSL/TLS features in Python

tests: stop checking for optional, now impossible output In 7dd63a8cb1ee, the code that could output that line was removed.

rust: remove one more occurrence of re2 Differential Revision: https://phab.mercurial-scm.org/D8601

scmutil: clarify getuipathfn comment Differential Revision: https://phab.mercurial-scm.org/D8600

githelp: add some minimal help for pickaxe functionality Due to a conversation in work chat, I realized this is actually pretty well-hidden in Mercurial. Differential Revision: https://phab.mercurial-scm.org/D8590

rust: remove duplicate import Differential Revision: https://phab.mercurial-scm.org/D8456

tests: remove "sslcontext" check Now that we require the presence of ssl.SSLContext in setup.py, the check would always return `True`.

sslutil: eliminate `_canloaddefaultcerts` by constant-folding code using it

tests: remove "defaultcacerts" check `sslutil._canloaddefaultcerts` is always true (and will be removed).

sslutil: eliminate `modernssl` by constant-folding code using it

hgweb: avoid using `sslutil.modernssl` `sslutil.modernssl` is going to be removed. Since the point of using this attribute was to check the importability of the `sslutil`, a different attribute can be used. `sslutil.wrapserversocket` is used because it’s anyway used a few lines below.

sslutil: remove comments referring to removed SSLContext emulation class

sslutil: remove code checking for presence of ssl.SSLContext Now that we require the presence of ssl.SSLContext in setup.py, we can remove this code.

setup: require a Python version with modern SSL features This increases the minimum security baseline of Mercurial and enables us to remove compatibility code for supporting older, less secure Python versions.

sslutil: set `_canloaddefaultcerts` to `True` if `ssl.SSLContext` is present The `load_default_certs()` method was already present when `ssl.SSLContext` was backported to Python 2.7 (https://hg.python.org/cpython/rev/221a1f9155e2).

filemerge: add __bytes__ for absentfilectx This will at _least_ aid some upcoming debugging. Differential Revision: https://phab.mercurial-scm.org/D8592

mergestate: move staticmethod _filectxorabsent to module level I suspect this was a static method just because it made merge.py feel less messy, but now we have a mergestate package so we can do better. Differential Revision: https://phab.mercurial-scm.org/D8591

rust: remove support for `re2` With the performance issues with `regex` figured out and fixed in previous patches and `regex` newly gaining support for empty alternations, there is no reason to keep `re2` around anymore. It's only *marginally* faster at creating the regex which saves at most a couple of ms, but gets beaten by `regex` in every other aspect. This removes the Rust/C/C++ bridge (hooray!), the `with-re2` feature, the conditional code that goes with it, the documentation and relevant part of the debug/module output. Differential Revision: https://phab.mercurial-scm.org/D8594

rust-dependencies: update `regex` to 1.3.9 Version `1.3.8` introduces support for empty alternations, which makes previously disallowed patterns usable in `regex`. From a user's perspective, this means that glob patterns like `*.py{,c}` will no longer generate an "invalid" regex and will use the Rust path. `1.3.9` is a bugfix release, might as well update to the latest one. Differential Revision: https://phab.mercurial-scm.org/D8593