record: prevent commits that don't pick up dirty subrepo changes (issue6102) This path covers interactive mode for commit, amend, and shelve, as well as the deprecated record extension. Since shelf creation uses commit without -S in the non-interactive case, aborting here should be OK. (I didn't check what happens to non interactive shelve creation if `ui.commitsubrepos=True` is set.) subrepoutil.precommit() will abort on a dirty subrepo if the config option isn't set, but the hint recommends using --subrepos to commit. Since only the commit command currently supports that option, the error has to be raised here to omit the hint. Doing the check before asking about all of the hunks in the MQ test seems like an improvement on its own. There's probably an additional check on this path that can be removed.

test-https: add some more known failure messages of client certs (issue6030) I don't think the exact error message is important here. On Debian sid, ECONNRESET is raised, and "[SSL] tlsv13 alert certificate required" on NetBSD.

test-https: turn off system OpenSSL configuration This mostly fixes the test failure on Debian sid where TLS 1.0 and 1.1 are disabled by default. https://sources.debian.org/patches/openssl/1.1.1a-1/Set-systemwide-default-settings-for-libssl-users.patch/ $OPENSSL_CONF could be set by run-tests.py, but the other tests should work without a "legacy" TLS, so I decided to not.

rebase: add missing dashes in help text

test: stabilize bundle generation for test-sparse-revlog.t To reduce the instability in the bundle binary content, we force it to contains delta against p1 in all cases. In the previous changeset, we already stabilized the processing of the bundle. So we don't see any output change in the test itself.

test: don't trust delta bases from the bundle in test-sparse-revlog.t The point of the test is to check the strategy sparse-revlog uses to pick delta base. If we trust the bases used in the bundle, we no longer fully test this logic. In order to force this computation we have to use the side effect of a legacy format configuration `format.generaldelta`. The lack of a more official way to do so will be fixed in a later changeset.

test: update test-sparse-revlog.t output This test is skipped unless a large artefact is pre-build. It seems like nobody ran it in a while. Changeset 3764330f76a6 changed the expected output but nobody noticed. This changeset focus on the first and simpler step: putting the expected output back to what one would get by running this test. However this test changes highlight a couple of deeper issues: 1) Even if the revision content did not changed, a change in the delta contained in the bundle affected the delta stored in the final revlog, changing the test result. Since we are testing the delta computation strategy with sparse, we should not blindly reuse the delta-base from the bundled delta. 2) A change in the format of the repository used to generate the bundle changed the delta stored in the bundle. We should get a more stable output to avoid future instabilities of this test. 3) The test is it not run by CI or developer. We'll try to address all this issues in the coming changesets.

tests: add a missing abort exit code on Windows in test-audit-subrepo.t

subrepo: add test for Windows relative-ish path with drive letter Matt Harbison pointed out that Windows had some weird path syntax. Fortunately it's rejected appropriately by pathauditor, so we're safe. Let's test the behavior as we have a special handling for Windows drive letters. This patch includes a basic example. Maybe we'll need to extend the test case further, but writing such tests on Linux isn't easy.

tests: fix regression tests failing on CentOS 7 Differential Revision: https://phab.mercurial-scm.org/D5877

packaging: modify rc detection to work with X.Yrc instead of X.Y-rc rc detection on CentOS failed without this change, resulting in upgrades from 4.9rc to 4.9 not working (4.9rc was considered more recent than 4.9). Differential Revision: https://phab.mercurial-scm.org/D5876

subrepo: avoid false unsafe path detection on Windows Subrepo paths are not normalized for the OS, so what was happening in the subsequent root path check was: root -> $TESTTMP\issue1852a\sub/repo util.expandpath(...) -> $TESTTMP\issue1852a\sub/repo os.path.realpath(...) -> $TESTTMP\issue1852a\sub\repo

Added signature for changeset 83377b4b4ae0

Added tag 4.9 for changeset 83377b4b4ae0

subrepo: reject potentially unsafe subrepo paths (BC) (SEC) In addition to the previous patch, this prohibits '~', '$nonexistent', etc. for any subrepo types. I think this is safer, and real-world subrepos wouldn't use such (local) paths.

subrepo: prohibit variable expansion on creation of hg subrepo (SEC) It's probably wrong to expand path at localrepo.*repository() layer, but fixing the layering issue would require careful inspection of call paths. So, this patch adds add a validation to the subrepo constructor. os.path.realpath(util.expandpath(root)) is what vfsmod.vfs() would do.

subrepo: extend path auditing test to include more weird patterns (SEC) While reviewing patches for the issue 5739, "$foo in repository path expanded", I realized that subrepo paths can also be cheated. This patch includes various subrepo paths which are potentially unsafe. Since an expanded subrepo path isn't audited, this bug allows symlink check bypass. As a result, a malicious subrepository could be checked out to a sub tree of e.g. $HOME directory. The good news is that the destination directory must be empty or nonexistent, so the existing ~/.bashrc wouldn't be overwritten. See the last part of the tests for details.

copyright: update to 2019 Differential Revision: https://phab.mercurial-scm.org/D5779

revlog: make sure we never use sparserevlog without general delta (issue6056) We are getting user report where the delta code tries to use `sparse-revlog` logic on repository where `generaldelta` is disabled. This can't work so we ensure the two booleans have a consistent value. Creating this kind of repository is not expected to be possible the current bug report point at a clonebundle related bug that is still to be properly isolated (Yuya Nishihara seems to a have done it). Corrupting a repository to reproduce the issue is possible. A test using this method is included in this fix.

sparserevlog: document the config option This was overlooked when this graduated from experimental.

rust-cpython: raising error.WdirUnsupported The Graph implementation of hg-cpython returns the appropriate error upon encounter with the working directory special revision number, and this gives us in particular a code path to test from test-rust-ancestors.py In the current implementation, the exception is actually raised from the iterator instantiation; we are nonetheless consuming the iterator in the test with `list()` in order not to depend on implementation details.

rust: error for WdirUnsupported with cpython conversion as exception This introduces WorkingDirectoryUnsupported as an enum variant of GraphError in the core and converts it to the expected `mercurial.error.WdirUnsupported`.

rust: working directory revision number constant This introduces the constant, but does not use it anywhere yet.

ui: remove unreachable branches and function calls from write() (issue6059) This is at least faster than ui.write() of 4.8.2. $ HGRCPATH=/dev/null hg files -R mozilla-central --time >/dev/null 4.8.2: time: real 2.340 secs (user 2.310+0.000 sys 0.020+0.000) 4.9rc0: time: real 2.580 secs (user 2.550+0.000 sys 0.020+0.000) this: time: real 2.230 secs (user 2.210+0.000 sys 0.020+0.000) Maybe the formatter should own a resolved write() function because it will just call dest.write(msg) most of the time, but that would be too much for stable.

ui: inline _writenobuf() into write() due to performance issue I'll remove redundant conditions later in this series.

ui: inline _write() into write() due to performance issue I'll remove redundant conditions later in this series.

ui: optimize buffered write with no label This was spotted while making fastannotate faster again after ditching its own formatter. Since I'm going to inline _write() into ui.write(), I decided to include this patch in this series. Here, the cost of '(self.label(a, label) for a in args)' was significant in hot loops.

partialdiscovery: avoid `undecided` related computation sooner than necessary Changeset 1d30be90c move the update of the `undecided` set within the `partialdiscovery` object in order to clarify the API. The update to the `undecided` set was unconditional in 1d30be90c and the first access to the `self.undecided` property triggered the initial computation of the set of undecided revisions. As a result, the set was computed much earlier, at a time where less information is available, immediately followed by an update of this set to remove common revisions. To fix this regression, we ignore the `undecided` related logic in `addcommons` when that `undecided` set has not been computed yet. Code that actually needs to know the `undecided` set will trigger its computation later. The change has no effects on semantic because the initial computation `undecided` set takes all knowns `common` into account. Example performance running `hg debugdiscovery` from a pypy repo missing 10 changesets: 870a89c6909d: 52.3ms (regression parent) 1d30be90c9dc: 72.0ms (regression) 5a5f504a7175: 64.8ms (this fix parent) this fix: 52.6ms

revlog: fix resolution of revlog version 0 This partially backs out cecf3f8bccd3, "revlog: always process opener options." My understanding is that if there's no "revlog1" nor "revlog2" in .hg/requires, the repository should stick to the v0 format. The reasoning is briefly described in 31a5973fcf96, "revlog: get rid of defversion." Maybe we can drop support for missing opener options, but I didn't do that in this patch.

rust: add comment about lack of wdirrev handling If hg is compiled with rust support, 'only(wdir())' crashed as "rustext.GraphError: ('ParentOutOfRange', 2147483647)", which should instead say "abort: working directory revision cannot be specified."