rust: less set lookups in AncestorsIterator This uses the boolean return of `HashSet::insert()` to factor pairs of contains()/insert() into a single insert() On the mozilla-central repository (450k changesets), I get about a bit more than 10% better medians in perfancestors (taking the mean of three runs) Best run for parent changeset: ! wall 0.106474 comb 0.110000 user 0.110000 sys 0.000000 (median of 93) Best run for this changeset: ! wall 0.093191 comb 0.090000 user 0.090000 sys 0.000000 (median of 100) Differential Revision: https://phab.mercurial-scm.org/D5942

obsutil: don't assume leftctx and rightctx repo as same Backed out changeset 520514af2d93. hgsubversion can pass leftctx and rightctx which are instances of two different repositories. This was making tests fail on hgsubversion with 4.9. The two different instances are: (Pdb) p rightctx.repo() <filteredrepo:served <hgsubversion.svnrepo.svnlocalrepo object at 0x7fe29d296d10>> (Pdb) p leftctx.repo() <filteredrepo:visible <hgsubversion.svnrepo.svnlocalrepo object at 0x7fe29d494590>> Differential Revision: https://phab.mercurial-scm.org/D5968

tests: add more wildcards to test-extdiff.t The diff tool (which is `echo`) runs in the background and output order can therefore be non-deterministic. We need to glob over the file names to account for this. Differential Revision: https://phab.mercurial-scm.org/D5976

url: don't pass strict argument on Python 3 The argument was removed in Python 3.4. Differential Revision: https://phab.mercurial-scm.org/D5975

url: always use str for proxy configuration Previously, proxies didn't work on Python 3 for various reasons. First, the keys to the "proxies" dict are fed into a `setattr(self, "%s_open", ...)` call and passing bytestrings results in setting an oddly named attribute due to the b'' in %s formatting. This resulted in "http_open" and "https_open" not being properly overridden and proxies not being used. Second, the standard library was expecting proxy URLs to be str. And various operations (including our custom code in url.py) would fail to account for the str/bytes mismatch. This commit normalizes everything to str and adjusts our proxy code in url.py to account for the presence of str on Python 3. Differential Revision: https://phab.mercurial-scm.org/D5952

py3: port tinyproxy.py to work with Python 3 There were various str/bytes mismatches in the code. This caused the proxy server to misbehave at run-time. The manifestation was typically premature socket disconnect from the perspective of the client. Differential Revision: https://phab.mercurial-scm.org/D5951

url: always access req._tunnel_host The getattr() was there to handle Python versions before 2.6, which lacked this attribute. We /might/ be able to further delete some code here. However, the behavior here is extremely hard to follow because large parts of this code duplicate code from the Python standard library and it is difficult to understand what is actually needed. Differential Revision: https://phab.mercurial-scm.org/D5950

tests: double escape \ in test-import-eol.t and test-mq-eol.t The shell eats the \\. Differential Revision: https://phab.mercurial-scm.org/D5974

tests: use raw string in test-impexp-branch.t On first glance, the escaping of \s seems correct. However, the shell eats the escape and we're left with '\s` in the written file. Let's use a raw string so we don't have to double escape. Differential Revision: https://phab.mercurial-scm.org/D5973

phabricator: make user searches case-insensitive User names in conduit are case insensitive, but when looking for "FOO" it would return "foo" instead and we'd think the user didn't exist. So lower case both the query and the response when comparing them. Differential Revision: https://phab.mercurial-scm.org/D5934

exchange: raise error.Abort instead of ValueError Raising ValueError results in an uncaught exception and a traceback being printed. In the context of servers, it can result in an HTTP 500 and an exception being logged in the error log. I don't think this is proper behavior. The bundle2 code paths have a mechanism for translating an error.Abort into an error message reported to the clients. I think we should use that instead. This commit replaces some ValueError with Abort so that servers can error more gracefully. Differential Revision: https://phab.mercurial-scm.org/D5972

tests: remove -q from test-lfs-serve.t This will make it easier to observe a behavior change in the next commit. Differential Revision: https://phab.mercurial-scm.org/D5971