rust-dirstate: architecture independence fix Apparently, c_char is u8 on ppc64le and i8 on amd64 Differential Revision: https://phab.mercurial-scm.org/D6473

context: get filesadded() and filesremoved() from changeset if configured This adds the read side for getting the sets of added and removed files from the changeset extras. I timed this command on the hg repo: hg log -T '{rev}\n {files}\n %:{file_mods}\n +{file_adds}\n -{file_dels}\n' It took 1m21s before and 6.4s after. I also used that command to check that the result didn't change compared to calculating the values from the manifests on the fly (it didn't change). In the mozilla-unified repo, the same command run on FIREFOX_BETA_58_END::FIREFOX_BETA_59_END went from 29s to 0.67s. Differential Revision: https://phab.mercurial-scm.org/D6417

changelog: optionally store added and removed files in changeset extras As mentioned in an earlier patch, copies._chain() is used a lot in the changeset-centric version of pathcopies(). It is expensive because it needs to look at the manifest in order to filter out copies whose target file has since been removed. I want to store the sets of added and removed files in the changeset in order to speed that up. This patch does the writing part of that. It could easily be a separate config, but it's currently tied to experimental.copies.write-to since that's the only real use case (it will also make the {file_*} template keywords faster, but I doubt that anyone cares enough about those to write extra metadata for them). The new information is stored in the changeset extras. Since they're always subsets of the changeset's "files" list, they're stored as indexes into that list. I've stored the indexes as stringified ints separated by NUL bytes. The size of 00changelog.d for the hg repo increased in size by 0.28% percent (compared to the size with only copy information in the changesets, which in turn is 0.17% larger than without copy information). We could store only the delta between the indexes and we could store them in binary, but the chosen format is more readable. We could also have implemented this as a cache outside the changelog. One advantage of doing it that way is that we would get the speedups from the {file_*} template keywords also on old repos. Another advantage is that it we can rewrite the cache if we find a bug in how we calculate the set of files. A disadvantage is that it would be more complex. Another is that it would surely use more space. We already write the copy information to the changeset extras, so it seems like a small step to also write these file sets. Differential Revision: https://phab.mercurial-scm.org/D6416

templatekw: make {file_*} compare to both merge parents (issue4292) This redefines the {file_adds}, {file_dels}, {file_mods} template keywords by getting the lists from the recently introduced context methods instead of getting them from status compared to p1. As mentioned before, these are better defined on merge commits. The total number of files from the three lists now always add up to the number of files in {files}. I timed this command: hg log -r 4.0::5.0 -T '{rev}\n {file_mods}\n {file_adds}\n {file_dels}\n' It went from 7.6s to 5.6s with this patch. So it's actually faster than before. Note that the "files:" field in the bazaar test log output was using "{file_mods}" (not "{files}" as one might think based on the label). Differential Revision: https://phab.mercurial-scm.org/D6369

narrowspec: use vfs.tryread() instead of reimplementing Note that parseconfig() works well with empty strings. Differential Revision: https://phab.mercurial-scm.org/D6465

help: remove a superfluous "the" in revlogs text Differential Revision: https://phab.mercurial-scm.org/D6466

setdiscovery: make progress on most connected groups each roundtrip Consider history like this: o | o | | | o | | | o |/ o | o | | | o | | | o |/ o | o | | | o | | | o |/ o ~ Assume the left mainline is available in the remote repo and the other commits are only in the local repo. Also imagine that instead of 3 local branches with 3 commits on each, there are 1000 branches (the number of commits on each doesn't matter much here). In such a scenario, the current setdiscovery code will pick a sample size of 200 among these branches and ask the remote which of them it has. However, the discovery for each such branch is completely independent of the discovery for the others -- knowing whether the remote has a commit in one branch doesn't give us any information about the other branches. The discovery will therefore take at least 5 roundtrips (maybe more depending on which commit in each linear chain was sampled). Since the discovery for each branch is independent, there is no reason to let one branch wait for another, so this patch makes it so we sample at least as many commits as there are branches. It may still happen (it's very likely, even) that we get multiple samples from one branch and none from another, but that will even out over a few rounds and I think this is still a big improvement. Because of http header size limits, we still use the old behavior unless experimental.httppostargs=true. I've timed this by running `hg debugdiscovery mozilla-unified --debug` in the mozilla-try repo. Both repos were local. Before this patch, last part of the output was: 2249 total queries in 5276.4859s elapsed time: 5276.652634 seconds heads summary: total common heads: 13 also local heads: 4 also remote heads: 8 both: 4 local heads: 28317 common: 4 missing: 28313 remote heads: 12 common: 8 unknown: 4 local changesets: 2014901 common: 530373 missing: 1484528 common heads: 1dad417c28ad 4a108e94d3e2 4d7ef530fffb 5350524bb654 777e60ca8853 7d97fafba271 9cd2ab4d0029 a55ce37217da d38398e5144e dcc6d7a0dc00 e09297892ada e24ec6070d7b fd559328eaf3 After this patch, the output was (including all the samples, since there were so few now): taking initial sample query 2; still undecided: 1599476, sample size is: 108195 sampling from both directions query 3; still undecided: 810922, sample size is: 194158 sampling from both directions query 4; still undecided: 325882, sample size is: 137302 sampling from both directions query 5; still undecided: 111459, sample size is: 74586 sampling from both directions query 6; still undecided: 26805, sample size is: 23960 sampling from both directions query 7; still undecided: 2549, sample size is: 2528 sampling from both directions query 8; still undecided: 21, sample size is: 21 8 total queries in 24.5064s elapsed time: 24.670051 seconds heads summary: total common heads: 13 also local heads: 4 also remote heads: 8 both: 4 local heads: 28317 common: 4 missing: 28313 remote heads: 12 common: 8 unknown: 4 local changesets: 2014901 common: 530373 missing: 1484528 common heads: 1dad417c28ad 4a108e94d3e2 4d7ef530fffb 5350524bb654 777e60ca8853 7d97fafba271 9cd2ab4d0029 a55ce37217da d38398e5144e dcc6d7a0dc00 e09297892ada e24ec6070d7b fd559328eaf3 Differential Revision: https://phab.mercurial-scm.org/D2647

help: clarify overlap of revlog header and first revlog entry Differential Revision: https://phab.mercurial-scm.org/D6449

py3: fix test-convert-svn-sink.t In cases where the root commit is empty commit, None will be returned as parents. This was implemented by 2c13e91ede6e. This breaks test on py3 because `b'%s' % None` does not work. It does not matter whether we return `None` or `'None'` as we skipped converting to svn step by doing an early return. So let's return `'None'`. I tried to patch all the users to convert `None` to `'None'`, but there were more users than I expected. I hit 3 of them and decided to fix it this way around. Differential Revision: https://phab.mercurial-scm.org/D6458

commit: respect --no-edit in combination with --amend Differential Revision: https://phab.mercurial-scm.org/D6464

commit: add test showing that commit --amend --no-edit still shows editor Differential Revision: https://phab.mercurial-scm.org/D6463

githelp: translate git stash show and clear actions and --patch flag Differential Revision: https://phab.mercurial-scm.org/D6461

githelp: add --dry-run for mv Differential Revision: https://phab.mercurial-scm.org/D6460

githelp: translate --directory of git apply to --prefix According to the help pages, these flags do the same. Differential Revision: https://phab.mercurial-scm.org/D6459

help: include subtopic in error message if passed Differential Revision: https://phab.mercurial-scm.org/D6442

help: check if a subtopic exists and raise an error if it doesn't (issue6145) Differential Revision: https://phab.mercurial-scm.org/D6441

perf: fix some missing b prefixes # skip-blame just b prefixes Differential Revision: https://phab.mercurial-scm.org/D6457

testparseutil: fix doctest to use str instead of bytes Differential Revision: https://phab.mercurial-scm.org/D6456

testparseutil: stop extracting using std* streams as bytes on py3 This is no longer required due to other cleanups in our linting tools. Differential Revision: https://phab.mercurial-scm.org/D6455

tests: sort some imports that were previously missed I'm a little unclear why the import checker didn't catch this before, but when I fixed it to work in Python 3 this failure started showing up. Sigh. Differential Revision: https://phab.mercurial-scm.org/D6454

contrib: fix import-checker to operate on str instead of bytes I believe this is fallout from other Python 3 cleanups, and our code linting tools are now leaning towards operating on str and not bytes. I don't feel strongly, so I've just restored this tool to working on Python 3. Differential Revision: https://phab.mercurial-scm.org/D6453

verify: use self._err not self.err, it changed in 7eaf4b1ac2a3 Differential Revision: https://phab.mercurial-scm.org/D6451

tests: make run-tests exit non-zero if there are "errors" Previously, if there was an error such as a broken .t file that caused run-tests.py to encounter an exception during parsing, the test would be considered in an "errored" state, which is separate from "failed". The check for whether to exit non-zero or not was based entirely on whether there were any tests in a "failed" state, so if there was only an error, run-tests would exit with 0. Our test infrastructure would then consider the test as passing, causing us to have some tests with false negatives that have gone undetected for a few weeks now. Differential Revision: https://phab.mercurial-scm.org/D6452

perf: add a `perfhelper-mergecopies` command This command gather data that are useful to pick argument for `perfmergecopies`.

perf: add a new `perfmergecopies` command This command benchmark calls to `mercurial.copies.mergecopies`

perf: factor selection of revisions involved in the merge out We will introduce more performance command around merge. As a first step we factor out pieces of `perfmergecalculate` that can be reused.

perf: allow to specify the base of the merge in perfmergecalculate We can now test the rebase case.

perf: add a --from flag to perfmergecalculate Before this change, `perfmergecalculate` was always benchmarking the merge of the working copy with another revision. We can now benchmark the `mergecalculate` call for any arbitrary pair of revision.

merge with stable

py3: fix test-narrow* which started failing because of recent changes #skip-blame because just r'' prefix Differential Revision: https://phab.mercurial-scm.org/D6447

tests: add test for {file_mods}, {file_adds}, {file_dels} on merge commit Differential Revision: https://phab.mercurial-scm.org/D6368

context: add ctx.files{modified,added,removed}() methods Changeset-centric copy tracing is currently very slow because it often reads manifests. One place it needs the manifest is in _chain(), where it removes a copy X->Y if Y has subsequently gotten removed. I want to speed that up by keeping track directly in the changeset of which files are removed in the changeset. These methods will be similar to ctx.p[12]copies() in that way: they will either read from the changeset or calculate the information from the manifests otherwise. Note that these are different from ctx.{modified,added,removed}() on merge commits. Those functions always compare to p1, but the new ones compare to both parents. filesadded() means "file does not exist in either parent but exists now", filesremoved() means "file existed in either parent but does not exist now", and filesmodified() means "file existed in either parent and still exists". The set of files in ctx.files() is the union of the files from the three new functions (and the three new ones are all disjoint sets). Also note that uncommitted merges are weird as usual. The invariant mentioned above still holds, but the functions compare to p1 (and are thus identical to the existing methods). Differential Revision: https://phab.mercurial-scm.org/D6367

copies: split up _chain() in naive chaining and filtering steps The function now has two clearly defined steps. The first step is the actual chaining. This step is very cheap. The second step is filtering out invalid copies. This step is expensive. For changeset-centric copy tracing, I want to do the filtering step only at the end. This patch prepares for that. Differential Revision: https://phab.mercurial-scm.org/D6418

relnotes: document changed behavior of ui.origbackuppath pointing to file Differential Revision: https://phab.mercurial-scm.org/D6446

templatekw: move showfileadds() close to showfile{mods,dels}() Differential Revision: https://phab.mercurial-scm.org/D6370

py3: use range() instead of xrange() The latter does not exist on Python 3. This makes test-contrib-perf.t pass on Python 3 again. Differential Revision: https://phab.mercurial-scm.org/D6443

narrow: move heads close to common as they are closely related Differential Revision: https://phab.mercurial-scm.org/D6445

narrow: pass binary nodeids to generateellipsesbundle2() We generally work with binary nodeids and it's should be expected that new function gets the nodeids in binary form already. Differential Revision: https://phab.mercurial-scm.org/D6444

match: stabilize _rootsdirsandparents doctest Changeset c4b8f8637d7a tried to stabilize some matcher test by using a set. This did not work because the set order is not stable. To fix it, we post process the result to display a sorted version of the set.

narrow: factor out logic to build ellipses related b2parts in separate fn This will help us switch more cleanly to using wireprotocol commands instead of using exchange.pull() which exchanges more things then required. Differential Revision: https://phab.mercurial-scm.org/D6435

narrow: remove unrequired compat code for old versions of hg As the comment says, that if is only required for servers having hg version 3.1 and 3.2. Any client connecting having hg 3.1 or 3.2 locally and trying to use narrow should already be broken taking in account the changes which have been done since narrow moved to core. Differential Revision: https://phab.mercurial-scm.org/D6434

perf: make sure to explicitly disable any profiler after the first iteration The current code work, because of some edge behavior of the `profile` class. We make it explicit that the profiler is not in effect more than once.

test: add missing 'cd ..' to test case Differential Revision: https://phab.mercurial-scm.org/D6439

match: remove an obsolete comment about util.finddirs() Obsolete since 8e55c0c642c (util: make util.dirs() and util.finddirs() include root directory (API), 2017-05-16). Differential Revision: https://phab.mercurial-scm.org/D6433

match: de-flake test-doctest.py by not depending on util.dirs() order util.dirs() yields directories in arbitrary order, which has made test-doctest.py flaky. I think they have been flaky since d8e55c0c642c (util: make util.dirs() and util.finddirs() include root directory (API), 2017-05-16). Before that commit, I think util.dirs() would return at most one entry, so there was only one iteration order. This patch fixes the problem by making _rootsdirsandparents() return a set (whose __str__() is defined to be in sorted order, I believe). The only caller wanted a set anyway. Differential Revision: https://phab.mercurial-scm.org/D6432

perf: add an option to profile the benchmark section Running a perf command with --profile gather data for the whole command execution, including setup and cleanup. This can significantly alter the data. To work around this we introduce a new option, it trigger the profiling of only one iteration of the benchmarked section.

perf: add a `pre-run` option sometimes, the initial run is necessary to warm some cache that are not relevant for the current measurement. We add a new `perf.pre-run` option to specify a number of run of the benchmark logic that will happens before measurement are taken.

narrow: consider empty commits to be "inside the narrow spec" for templates It doesn't seem useful to exclude them, or harmful to include them. Users writing log templates using outsidenarrow as a predicate might consider it unexpected if their locally created empty drafts are treated as if they contained something outside the clone. Differential Revision: https://phab.mercurial-scm.org/D6414

rust-python3: useless python2 specific import This python27_sys import prevents building with python3, it had been previously removed in a5fa9140ce4c, but that has been since pruned Differential Revision: https://phab.mercurial-scm.org/D6415

rust-python3: compatibility fix for incoming PyLong On Python3, PyInt is PyLong and it doesn't have the `value()` method. Re upcasting to PythonObj as done here works, but we might prefer taking a PythonObj from the onset (would require more testing) Differential Revision: https://phab.mercurial-scm.org/D6397

py3: add one new passing test found by buildbot Differential Revision: https://phab.mercurial-scm.org/D6412

discovery: slowly increase sampling size Some pathological discovery runs can requires many roundtrip. When this happens things can get very slow. To make the algorithm more resilience again such pathological case. We slowly increase the sample size with each roundtrip (+5%). This will have a negligible impact on "normal" discovery with few roundtrips, but a large positive impact of case with many roundtrips. Asking more question per roundtrip helps to reduce the undecided set faster. Instead of reducing the undecided set a linear speed (in the worst case), we reduce it as a guaranteed (small) exponential rate. The data below show this slow ramp up in sample size: round trip | 1 | 5 | 10 | 20 | 50 | 100 | 130 | sample size | 200 | 254 | 321 | 517 | 2 199 | 25 123 | 108 549 | covered nodes | 200 | 1 357 | 2 821 | 7 031 | 42 658 | 524 530 | 2 276 755 | To be a bit more concrete, lets take a very pathological case as an example. We are doing discovery from a copy of Mozilla-try to a more recent version of mozilla-unified. Mozilla-unified heads are unknown to the mozilla-try repo and there are over 1 million "missing" changesets. (the discovery is "local" to avoid network interference) Without this change, the discovery: - last 1858 seconds (31 minutes), - does 1700 round trip, - asking about 340 000 nodes. With this change, the discovery: - last 218 seconds (3 minutes, 38 seconds a -88% improvement), - does 94 round trip (-94%), - asking about 344 211 nodes (+1%). Of course, this is an extreme case (and 3 minutes is still slow). However this give a good example of how this sample size increase act as a safety net catching any bad situations. We could image a steeper increase than 5%. For example 10% would give the following number: round trip | 1 | 5 | 10 | 20 | 50 | 75 | 100 | sample size | 200 | 321 | 514 | 1 326 | 23 060 | 249 812 | 2 706 594 | covered nodes | 200 | 1 541 | 3 690 | 12 671 | 251 871 | 2 746 254 | 29 770 966 | In parallel, it is useful to understand these pathological cases and improve them. However the current change provides a general purpose safety net to smooth the impact of pathological cases. To avoid issue with older http server, the increase in sample size only occurs if the protocol has not limit on command argument size.

tests: make the grep pattern in remotefilelog-gcrepack portable (issue6122) test-remotefilelog-gcrepack was using "\" to escape "|" in the grep pattern. The most of implementations ignore "\" when it is followed by "|", so the regex works. However, OpenBSD doesn't ignore "\" and considers "|" part of the text instead of create two branches. Neither of both behaviors violate POSIX. This change removes the unnecessary escape character and changes grep to egrep, so the extended regular expression works on every unix. This is part of the bug 6122. Tested on OpenBSD, GNU, FreeBSD, NetBSD, Solaris 11 and BusyBox. Credits to Todd C. Miller, Paul de Weerd and Ingo Schwarze for helping me with it.

help: document new "bookmarksinstore" requirement in internals.requirements Differential Revision: https://phab.mercurial-scm.org/D6413

absorb: fix interactive mode I didn't know existed While investigating a bug in `hg absorb -e` I unintentionally discovered `hg absorb --interactive` and its brokenness. This adds a test and restores the functionality. Note that this interface is still marked experimental, so we can change this to be more sophisticated in the future. Differential Revision: https://phab.mercurial-scm.org/D6411

tests: work around libressl being different about error strings (issue6122) As far as I can tell, this is the right behavior. Thanks to Alex Gaynor for checking what the string means by looking at libressl sources for me. Differential Revision: https://phab.mercurial-scm.org/D6410

merge with stable

templatekw: change default value of 'requires' to () Since we dropped support for the old-style template keywords, we no longer have to distinguish None (old-style) and an empty requirement (new-style).

commit: move sorting of added and removed files list to lower level localrepo.commitctx() has lists of all changed files, as well as lists of added and removed files. The list of all files is unsorted and changelog.add() will sort it. Let's also sort the lists of added and removed files at a lower level (manifestrevlog.add()) for consistency. It also seems safer to do it there, just before we write them to the store. That way other callers won't be able to create invalid commits (or whatever the consequence is) by passing in unsorted lists. Also, alternative storages may not care that the lists are sorted. I don't think this will be a performance problem (someone should have fixed the sorting in changelog.add() if it were). Differential Revision: https://phab.mercurial-scm.org/D6390

match: drop unnecessary adding of '' to set of dirs This breaks some tests for "rootfilesin:" in a pattern matcher even more, but that just shows how broken that case is. Differential Revision: https://phab.mercurial-scm.org/D6406