git: a little pycompat.bytestring() love to make this code work in py3 Differential Revision: https://phab.mercurial-scm.org/D5765

py3: have test-revset2.t write test scripts in a more portable way Fixes the test on Python 3. Differential Revision: https://phab.mercurial-scm.org/D5763

py3: fix up test-remotefilelog-cacheprocess.t to not depend on a repr It looks like the repr() of Exceptions is different from Python 2 to Python 3.7 (but not 3.5?), but the str() is still stable. Sigh. Differential Revision: https://phab.mercurial-scm.org/D5761

remotefilelog: cast division result to an int Otherwise mid is a float and this confuses __slice__ on Python 3. Differential Revision: https://phab.mercurial-scm.org/D5760

tests: cast division result to int Otherwise it is a float on Python 3 and code later compares about casting a float to an int. Differential Revision: https://phab.mercurial-scm.org/D5759

tests: various Python 3 ports for test-remotefilelog-datapack.py Use bytes I/O. Use byteschr(). Convert temporary path to bytes. Differential Revision: https://phab.mercurial-scm.org/D5758

tests: use items() in test-remotefilelog-datapack.py Performance doesn't matter in tests. iteritems() doesn't exist in Python 3. Differential Revision: https://phab.mercurial-scm.org/D5757

tests: use bytes and %d formatting in test-remotefilelog-datapack.py There were numerous failures on Python 3 due to str/bytes mismatch and '%s' not working for ints. Differential Revision: https://phab.mercurial-scm.org/D5756

perf: add a --[no-]clear-caches option to `perfnodemap` The option is useful to look at pure lookup performance on a warm data structure.

perf: add a perfnodemap command The command focus on timing of the nodemap object itself.

wireprotov1server: use binascii.unhexlify The "hex" codec doesn't exist in Python 3. We could use `codecs.decode(h, 'hex_codec')`. But `binascii.unhexlify()` exists and should work the same on Python 2 and 3. Differential Revision: https://phab.mercurial-scm.org/D5755

tests: conditionalize test-http-bad-server.t for Python 3.5 It appears that Python 3 introduced output buffering in the HTTP response stack. And Python 3.6 switched from sock.makefile().write() to sock.sendall(). So, we need to conditionalize test-http-bad-server.t to account for the difference in behavior between Python 3.5 and 3.6. Differential Revision: https://phab.mercurial-scm.org/D5754

tests: log sendall() operations and port test-http-bad-server.t Python 3's HTTP server layer buffers output and uses sendall() instead of write(). In order to make test-http-bad-server.t pass on Python 3, we needed to teach our socket proxy to log sendall() events and to abort future sends if we reached our send limit. The tests using `tail` were difficult to port with inline output conditionals since the number of lines varied. So we now use `#if py3` for these tests. test-http-bad-server.t now passes on Python 3.6 and 3.7 on at least Linux. However, it does not yet pass on Python 3.5 because of low-level differences to how the HTTP server is implemented. Differential Revision: https://phab.mercurial-scm.org/D5753

tests: glob away readline(-1) Most of these are readline(65537) on Python 3. I don't think it is worth the readability hit to use (re), as it would require escaping parenthesis. Differential Revision: https://phab.mercurial-scm.org/D5752

tests: change how sockets are closed Python 3 uses a different type to represent a socket file object than Python 2. We need to conditionalize how the socket is closed accordingly. While we're here, we switch to use socket.shutdown() to close the socket. This is because socket.close() may not actually close the socket until it is GCd. socket.shutdown() forces an immediate shutdown. I suspect Python 3 changed semantic behavior here, as I can't get test-http-bad-server.t to work with socket.close(). socket.shutdown() does appear to work, however. Differential Revision: https://phab.mercurial-scm.org/D5751

tests: add b'' prefixes to badserverext.py This avoids a handful of failures due to missing str and bytes. # skip-blame: just a bunch of b'' prefixes Differential Revision: https://phab.mercurial-scm.org/D5750

hgweb: log error before attempting I/O Previously, an uncaught exception during HTTP request serving would attempt to send an error response then log the exception. If an exception occurred during I/O, this exception would be raised and the original exception wouldn't be logged. This commit changes behavior so the original exception is logged first, before we attempt to do anything else. This ensures the exception is logged. This change resulted in new tracebacks appearing in various tests. Because tracebacks can vary between Python versions, we added a simple script to filter the stack part of traceback lines. This makes testing much simpler, as we don't need to glob over lines and make lines conditional. Differential Revision: https://phab.mercurial-scm.org/D5749

tests: write commit message using file I/O Python 2.7 will print() \x94\x5c\x0a whereas Python 3 will print() \xc2\x94\x5c\x0a. Why, I'm not sure. It probably has to do with print() being Unicode aware on Python 3 and Python attempting some kind of encoding before emitting the output. This difference results in a different bytes making it to the commit message and the JSON output varying. We work around this by writing bytes to a commit message file. Differential Revision: https://phab.mercurial-scm.org/D5741

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

patch: handle 0 context lines (diff.unified=0) when parsing patches Previously, if there were no context lines, we would just keep updating the ranges and the hunk, but not actually storing the hunk (just overwriting it each time). Thus a diff like this: $ hg diff --config diff.unified=0 diff --git a/bar b/bar --- a/bar +++ b/bar @@ -1,0 +2,1 @@ 1 +change1 @@ -3,0 +5,1 @@ 3 +change2 would come out of the parser like this (change1 is lost): bar: @@ -3,0 +5,1 @@ 3 +change2 This had some really weird side effects for things like commit --interactive, split, etc. Differential Revision: https://phab.mercurial-scm.org/D5743

py3: pass str into RuntimeError() to prevent b'' in output # skip-blame as just r'' prefixes Differential Revision: https://phab.mercurial-scm.org/D5740

py3: use '%d' instead of '%s' for integers Differential Revision: https://phab.mercurial-scm.org/D5738

tests: port test-hgweb-auth.py to Python 3 Differential Revision: https://phab.mercurial-scm.org/D5736

url: convert some variables back to bytes Differential Revision: https://phab.mercurial-scm.org/D5735

url: add some defensive asserts on expected incoming types Our type handling is a nightmare here, and we're loading passwords to do network IO, so we can afford to be potentially-slow but pedantic here. Differential Revision: https://phab.mercurial-scm.org/D5734

changegroup: don't try to build changelog chunks if not required When we extend a narrow clone without ellipsis, we don't download changelog information because that's already present with the client. However we still try to build that chunk stream. Building that chunk stream involves calling a lookup function and store.emitrevisions() API. The lookup function is called len(cl) number of times. On large repositories, where len(cl) is in millions, calling that lookup function is not a good idea. Also it's not required to use the store.emitrevisons() API because we already have nodes present which we can use. This patch short-circuits state building logic if we are processing a non-ellipsis case and changelog is not required. This saves up ~20 seconds on our internal repo for a single extend call. Differential Revision: https://phab.mercurial-scm.org/D5733

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.

changegroup: initialize the state variable a bit earlier This will make the next patch much easier. Differential Revision: https://phab.mercurial-scm.org/D5732

tests: conditionalize test output on Python 3.7 Python 3.7 changed behavior of urllib.parse.quote() from RFC 2396 to RFC 3986 and ~ is now in the set of reserved characters and isn't escaped. We conditioanlize test output accordingly. Differential Revision: https://phab.mercurial-scm.org/D5717

hghave: add pyXY features for Python version numbers This will allow us to sniff for Python >= versions in tests. Differential Revision: https://phab.mercurial-scm.org/D5088

py3: whitelist couple more passing tests found by buildbot Differential Revision: https://phab.mercurial-scm.org/D5731

keepalive: implement _close_conn() so closes are known Keepalives were not working on Python 3 because http.client.HTTPResponse was refactored to call _close_conn() instead of close(). Our custom close() is what returns inactive connections to the available state. We better support Python 3 by implementing a _close_conn(). Differential Revision: https://phab.mercurial-scm.org/D5720

lfs: explicitly add the Content-Length header when uploading blobs, for py3 This was the reason for test-lfs-test-server.t#git-server complaining about an "invalid byte in chunk length". For some reason if this isn't explicitly added, py3.7.1 is adding `transfer-encoding: chunked` as well as `Content-length: x`. Wireshark flagged this as malformed. However, if this is set, it doesn't bother with `transfer-encoding`. Before this patch with py3: PUT /objects/31cf46fbc4ecd458a0943c5b4881f1f5a6dd36c53d6167d5b69ac45149b38e5b HTTP/1.1 Accept-Encoding: identity Content-length: 12 accept: application/vnd.git-lfs content-type: application/octet-stream host: localhost:20062 transfer-encoding: chunked user-agent: git-lfs/2.3.4 (Mercurial 4.9rc0+149-7eb7637e34bf) Before this patch with py27: PUT /objects/31cf46fbc4ecd458a0943c5b4881f1f5a6dd36c53d6167d5b69ac45149b38e5b HTTP/1.1 Accept-Encoding: identity accept: application/vnd.git-lfs content-type: application/octet-stream content-length: 12 host: localhost:20062 user-agent: git-lfs/2.3.4 (Mercurial 4.9rc0+149-7eb7637e34bf+20190128) With this patch and py3, the content is the same as the py27 example. RFC2616 says to ignore `Content-Length` if `Transfer-Encoding` is present, so maybe there's nothing to do in the hg-server side (though I'm not sure which it is using if presented both). Maybe chunked encoding is better to do? If someone knows how to suppress the `Content-Length`, we can try that instead.

perf: support looking up multiple revisions The nodemap code has optimisations around the number of lookup we actually made. As a result, being able to specify multiple revisions to look up is important when measuring performances. One can now specify full revspecs with the --rev arguments.

perf: add a no-lookup variant to perfindex It is useful to check how long it takes to create a index object without doing anything with it. We add a new flag dedicated to that.

perf: add some documentation to perfindex It seems useful to document how the arguments can affect the benchmark.

perf: move cache clearing in the `setup` step of `perfheads` The cache clearing is pretty fast, but this seems more "correct".

revlog: document cext oddities in terms of object/caches This clarify why we just call clearcaches on a single object.

perf: document perfheads

py3: stabilize the output of lfs commandserver tests The print() statements were being output at the very end, so write to the same stdout sink as runcommand, and explicitly flush.

py3: conditionalize some LFS test output I'm not sure why the one stackframe is py2 only, but that seems harmless. The remaining failure is LfsCorruptionError printing the fully qualified name, as well as b'' around its message.

lfs: strip the response headers from the Batch API before printing For reasons unknown, py3 is adding an extra '\n' before the headers print out. This makes the output the same as py2.

py3: force hgweb.server error log to internally write unicode Otherwise, there's a lot of py2/py3 divergence in the LFS tests because of the "HG error" lines picking up a b'' prefix. wsgicgi.py uses procutil.stderr, so I assume the input was meant to be bytes.

py3: byteify the decoded JSON responses upon receipt in the LFS blobstore It got too confusing juggling r'' vs b'' across several functions.

hgweb: ensure Content-Length and Content-Type are not promoted to HTTP_ on py3 In stabilizing test-lfs-serve-access.t for py3, the server started asserting on blob upload: Environment should not have the key: HTTP_CONTENT_LENGTH (use CONTENT_LENGTH instead) It could be avoided by explicitly setting the Content-Length header on the client side. I didn't go back to py2, but printing the original header here in py37 revealed 'Content-length' when sent to the error log.

py3: raw stringify various JSON and HTTP headers in the LFS blobstore module This is (almost?) entirely from Augie's work. I'm a bit surprised that the JSON data is being encoded with ASCII via `pycompat.bytesurl()`- I would have thought UTF-8.

py3: byteify the LFS blobstore module This is almost entirely b'' prefixing, with a couple of exceptions forced to bytes. Much of this is also borrowed from Augie's code. There's an HTTPError.read() that I flagged that I assume needs to be converted to bytes, but I can't find confirmation. Handling the deserialized JSON object over several functions made r'' vs b'' accesses confusing, so this assumes that the JSON object will be converted to bytes immediately. That will be done in the following commits, so it's not buried in these trivial changes.

py3: raw stringify various things in the LFS server module Some of this is based on code written by Augie. I'm slightly unsure if these are the correct pycompat bytes <-> str conversion methods.

py3: add b'' prefixes to the LFS server module There are a ton of these changes in the blobstore, as well as r'' prefixing. So separating these out hopefully makes review easier. # skip-blame for b'' prefixing

py3: add b'' and r'' prefixes to LFS tests

py3: more tests from the ratchet Differential Revision: https://phab.mercurial-scm.org/D5730

dispatch: unify handler of IOError and OSError These exceptions were merged in Python 3.

dispatch: quote filename in IOError as well It's explicitly added for OSError at 720308f741cb "dispatch: show empty filename in OSError aborts". Let's do the same for IOError.

dispatch: add inline comment about possible IOError subtypes It's hard to tell which "if" would handle which exception.