perf: offer full details in perfrevlogwrite This will be useful for people who want to study the timing pattern more closely.

perf: introduce a perfrevlogwrite command The command record times taken by adding many revisions to a revlog. Timing each addition, individually. The "added revision" are recreations of the original ones. To time each addition individually, we have to handle the timing and the reporting ourselves. This command is introduced to track the impact of sparse-revlog format on delta computations at initial storage time. It starts with the full text, a situation similar to the "commit". Additions from an existing delta are better timed with bundles. The complaints from `check-perf-code.py` are not relevant. We are accessing and "revlog" opener, not a repository opener.

tests: fix config knob in test-narrow-clone-stream.t Two patches landed in parallel and had a semantic conflict. This resolves the mess and leaves us with passing tests. Differential Revision: https://phab.mercurial-scm.org/D5231

remotefilelog: fix various whitespace issues in docstring Differential Revision: https://phab.mercurial-scm.org/D5230

ui: add config knob to redirect status messages to stderr (API) This option can be used to isolate structured output from status messages. For now, "stdio" (stdout/err pair) and "stderr" are supported. In future patches, I'll add the "channel" option which will send status messages to a separate command-server channel with some metadata attached, maybe in CBOR encoding. This is a part of the generic templating plan: https://www.mercurial-scm.org/wiki/GenericTemplatingPlan#Sanity_check_output .. api:: Status messages may be sent to a dedicated stream depending on configuration. Don't use ``ui.status()``, etc. as a shorthand for conditional writes. Use ``ui.write()`` for data output.

phabricator: ensure the command summaries are available in extension help Previously, `hg help phabricator` listed the 3 supported commands at the bottom of the extension help, but said "no help text available".

hgweb: cast bytearray to bytes PEP-3333 seems to indicate that bytes is the only allowed type that can be used to express the output of a WSGI application. And some WSGI environments seem to enforce this (mod_wsgi does). This commit universally casts bytearray instances to bytes to appease the WSGI specification. I found this because wireprotov2 is emitting bytearray instances. I'd like to keep things that way because the way it builds a data structure, bytearray is more efficient. I'd rather keep the low-level code efficient (and using bytearray) and cast at the edges than impose a performance penalty on code that may run outside WSGI contexts.

help: unjumble the list of default config values for `internals.config`

ui: hide fin/fout/ferr attributes behind @property functions This allows keeping references to fout/ferr/fin which are updated when these properties are changed. See the next patch.

ui: label prompt and echo messages I'm going to add a dedicated command-server channel for status messages, which carries metadata alongside a message text. 'ui.*' label provides a hint how message text should be processed.

ui: add inner function to select write destination I'm going to add a config knob to redirect any status messages to stderr. This function helps to switch underlying file objects. # no-check-commit because of existing write_err() function

ui: remove _write() and _write_err() functions

ui: move pre/post processes from low-level write()s to _writenobuf() This helps adding a dedicated stream for status/error messages. I don't want to add _write*() function per stream.

ui: pass in file object to _writenobuf() See the subsequent patches for why. The "if" block in _writenobuf() will be removed soon.

perf: fix perfrevlogrevisions --reverse Currently, 'endrev' equals `len(revlog)`, a revision that does not exist. When asking for the reverse order, the arguments passed to xrange are `xrange(len(revlog), startrev)` which then crash. We need to offset 'endrev' by one so we don't crash anymore. Also, we offset 'startrev' to ensure we get the same number of revisions with and without the `--reverse` option. Differential Revision: https://phab.mercurial-scm.org/D5228

procutil: import concerns about creationflags on Windows from D1701 I don't have the need anymore for the change in D1701 nor the time to investigate the changes on all supported Windows platforms. I import the stuff I learned on D1701 in the `runbgcommand` so the next people working on it can starts from there. Differential Revision: https://phab.mercurial-scm.org/D5229

localrepo: extract loading of hgrc files to standalone function Various 3rd party extensions supplement where per-repo config data lives. Looking at their sources, they resort to unorthodox means to inject the config data. And the way they do it is susceptible to corner cases. e.g. not processing automatic extension loads, not reacting to new or disabled extensions in configs, etc. This commit extracts the core logic of loading hgrc files into a standalone function so there is a clear function that can be monkeypatched to inject per-repo config data at repository open time. Differential Revision: https://phab.mercurial-scm.org/D5221

revsets: make bookmark/named('re:nonexistent') not abort (issue6018) (BC) Foozy documented the differences between revsets branch(), tag(), bookmark(), and named() in eeb5d5ab14a6 (revset: raise RepoLookupError to make present() predicate continue the query, 2015-01-31). He seemed to want tag() to change behavior to not error out on non-matching regular expressions. I think it's instead bookmark() and named() that should not error out. So that's what this patch does. Differential Revision: https://phab.mercurial-scm.org/D5220

fix: rename :fileset subconfig to :pattern This name was always inaccurate, since the config accepts any pattern. Hopefully so few people use this right now that it won't matter, but there will now be a warning if the old config name is used. Differential Revision: https://phab.mercurial-scm.org/D5226

fix: add a config to abort when a fixer tool fails This allows users to stop and address tool failures before proceeding, instead of the default behavior of continuing to apply any tools that didn't fail. For example, a code formatting tool could fail if you have syntax errors, and you might want your repo to stay in its current state while you fix the syntax error before re-running 'hg fix'. It's conceivable that this would even be necessary for the correctness of some fixer tools across a chain of revisions. Differential Revision: https://phab.mercurial-scm.org/D5200

perf: measure slicing time in perfrevlogrevision Slicing a sparse delta chain can be expensive. We now benchmark the associated time.

perf: teach perfrevlogrevision about sparse reading Before this change, chunks were always read in a single block. Even in the sparse-read/sparse-revlog case. This gave a false view of the performance and could lead to memory consumption issue.

perf: use the same timer for all section of perfrevlogrevision Otherwise the -T json output is invalid.

obsutil: clarify the access to "repo" We use the variable multiple times and we might use it even more in the future. We use a temporary variable instead.

pycompat: adding Linux detection and fixing Mac Python 3 recommends detecting OSs with the prefix of the platform, but we were comparing the full string for macOS. We also didn't have Linux detection, which is convenient for extensions to use (rather than have some OSs detected by hg and some by the extension). Reference: https://docs.python.org/3/library/sys.html#sys.platform Differential Revision: https://phab.mercurial-scm.org/D5227

remotefilelog: fix various minor py3 problems # skip-blame b prefixes and pycompat.long, nothing remotely interesting Differential Revision: https://phab.mercurial-scm.org/D5223

remotefilelog: rip out a missed mention of lz4 Differential Revision: https://phab.mercurial-scm.org/D5222

ui: wrap whole _write() block with timeblockedsection I think the cost of color labeling is negligible compared to the I/O syscalls. Let's simply wrap the whole write() function so that we can eliminate _write() and _write_err() in later changeset.

ui: indent _writenobuf() to prepare moving bits from _write() functions

ui: simply concatenate messages before applying color labels This should be cheaper in space than applying labels for each message.