view mercurial/profiling.py @ 35569:964212780daf

rust: implementation of `hg` This commit provides a mostly-working implementation of the `hg` script in Rust along with scaffolding to support Rust in the repository. If you are familiar with Rust, the contents of the added rust/ directory should be pretty straightforward. We create an "hgcli" package that implements a binary application to run Mercurial. The output of this package is an "hg" binary. Our Rust `hg` (henceforth "rhg") essentially is a port of the existing `hg` Python script. The main difference is the creation of the embedded CPython interpreter is handled by the binary itself instead of relying on the shebang. In that sense, rhg is more similar to the "exe wrapper" we currently use on Windows. However, unlike the exe wrapper, rhg does not call the `hg` Python script. Instead, it uses the CPython APIs to import mercurial modules and call appropriate functions. The amount of code here is surprisingly small. It is my intent to replace the existing C-based exe wrapper with rhg. Preferably in the next Mercurial release. This should be achievable - at least for some Mercurial distributions. The future/timeline for rhg on other platforms is less clear. We already ship a hg.exe on Windows. So if we get the quirks with Rust worked out, shipping a Rust-based hg.exe should hopefully not be too contentious. Now onto the implementation. We're using python27-sys and the cpython crates for talking to the CPython API. We currently don't use too much functionality of the cpython crate and could have probably cut it out. However, it does provide a reasonable abstraction over unsafe {} CPython function calls. While we still have our fair share of those, at least we're not dealing with too much refcounting, error checking, etc. So I think the use of the cpython crate is justified. Plus, there is not-yet-implemented functionality that could benefit from cpython. I see our use of this crate only increasing. The cpython and python27-sys crates are not without their issues. The cpython crate didn't seem to account for the embedding use case in its design. Instead, it seems to assume that you are building a Python extension. It is making some questionable decisions around certain CPython APIs. For example, it insists that PyEval_ThreadsInitialized() is called and that the Python code likely isn't the main thread in the underlying application. It is also missing some functionality that is important for embedded use cases (such as exporting the path to the Python interpreter from its build script). After spending several hours trying to wrangle python27-sys and cpython, I gave up and forked the project on GitHub. Our Cargo.toml tracks this fork. I'm optimistic that the upstream project will accept our contributions and we can eventually unfork. There is a non-trivial amount of code in our custom Cargo build script. Our build.rs (which is called as part of building the hgcli crate): * Validates that the Python interpreter that was detected by the python27-sys crate provides a shared library (we only support shared library linking at this time - although this restriction could be loosened). * Validates that the Python is built with UCS-4 support. This ensures maximum Unicode compatibility. * Exports variables to the crate build allowing the built crate to e.g. find the path to the Python interpreter. The produced rhg should be considered alpha quality. There are several known deficiencies. Many of these are documented with inline TODOs. Probably the biggest limitation of rhg is that it assumes it is running from the ./rust/target/<target> directory of a source distribution. So, rhg is currently not very practical for real-world use. But, if you can `cargo build` it, running the binary *should* yield a working Mercurial CLI. In order to support using rhg with the test harness, we needed to hack up run-tests.py so the path to Mercurial's Python files is set properly. The change is extremely hacky and is only intended to be a stop-gap until the test harness gains first-class support for installing rhg. This will likely occur after we support running rhg outside the source directory. Despite its officially alpha quality, rhg copes extremely well with the test harness (at least on Linux). Using `run-tests.py --with-hg ../rust/target/debug/hg`, I only encounter the following failures: * test-run-tests.t -- Warnings emitted about using an unexpected Mercurial library. This is due to the hacky nature of setting the Python directory when run-tests.py detected rhg. * test-devel-warnings.t -- Expected stack trace missing frame for `hg` (This is expected since we no longer have an `hg` script!) * test-convert.t -- Test running `$PYTHON "$BINDIR"/hg`, which obviously assumes `hg` is a Python script. * test-merge-tools.t -- Same assumption about `hg` being executable with Python. * test-http-bad-server.t -- Seeing exit code 255 instead of 1 around line 358. * test-blackbox.t -- Exit code 255 instead of 1. * test-basic.t -- Exit code 255 instead of 1. It certainly looks like we have a bug around exit code handling. I don't think it is severe enough to hold up review and landing of this initial implementation. Perfect is the enemy of good. Differential Revision: https://phab.mercurial-scm.org/D1581
author Gregory Szorc <gregory.szorc@gmail.com>
date Wed, 10 Jan 2018 08:53:22 -0800
parents 83dfbda40e67
children 7b86aa31b004
line wrap: on
line source

# profiling.py - profiling functions
#
# Copyright 2016 Gregory Szorc <gregory.szorc@gmail.com>
#
# This software may be used and distributed according to the terms of the
# GNU General Public License version 2 or any later version.

from __future__ import absolute_import, print_function

import contextlib

from .i18n import _
from . import (
    encoding,
    error,
    extensions,
    util,
)

def _loadprofiler(ui, profiler):
    """load profiler extension. return profile method, or None on failure"""
    extname = profiler
    extensions.loadall(ui, whitelist=[extname])
    try:
        mod = extensions.find(extname)
    except KeyError:
        return None
    else:
        return getattr(mod, 'profile', None)

@contextlib.contextmanager
def lsprofile(ui, fp):
    format = ui.config('profiling', 'format')
    field = ui.config('profiling', 'sort')
    limit = ui.configint('profiling', 'limit')
    climit = ui.configint('profiling', 'nested')

    if format not in ['text', 'kcachegrind']:
        ui.warn(_("unrecognized profiling format '%s'"
                    " - Ignored\n") % format)
        format = 'text'

    try:
        from . import lsprof
    except ImportError:
        raise error.Abort(_(
            'lsprof not available - install from '
            'http://codespeak.net/svn/user/arigo/hack/misc/lsprof/'))
    p = lsprof.Profiler()
    p.enable(subcalls=True)
    try:
        yield
    finally:
        p.disable()

        if format == 'kcachegrind':
            from . import lsprofcalltree
            calltree = lsprofcalltree.KCacheGrind(p)
            calltree.output(fp)
        else:
            # format == 'text'
            stats = lsprof.Stats(p.getstats())
            stats.sort(field)
            stats.pprint(limit=limit, file=fp, climit=climit)

@contextlib.contextmanager
def flameprofile(ui, fp):
    try:
        from flamegraph import flamegraph
    except ImportError:
        raise error.Abort(_(
            'flamegraph not available - install from '
            'https://github.com/evanhempel/python-flamegraph'))
    # developer config: profiling.freq
    freq = ui.configint('profiling', 'freq')
    filter_ = None
    collapse_recursion = True
    thread = flamegraph.ProfileThread(fp, 1.0 / freq,
                                      filter_, collapse_recursion)
    start_time = util.timer()
    try:
        thread.start()
        yield
    finally:
        thread.stop()
        thread.join()
        print('Collected %d stack frames (%d unique) in %2.2f seconds.' % (
            util.timer() - start_time, thread.num_frames(),
            thread.num_frames(unique=True)))

@contextlib.contextmanager
def statprofile(ui, fp):
    from . import statprof

    freq = ui.configint('profiling', 'freq')
    if freq > 0:
        # Cannot reset when profiler is already active. So silently no-op.
        if statprof.state.profile_level == 0:
            statprof.reset(freq)
    else:
        ui.warn(_("invalid sampling frequency '%s' - ignoring\n") % freq)

    statprof.start(mechanism='thread')

    try:
        yield
    finally:
        data = statprof.stop()

        profformat = ui.config('profiling', 'statformat')

        formats = {
            'byline': statprof.DisplayFormats.ByLine,
            'bymethod': statprof.DisplayFormats.ByMethod,
            'hotpath': statprof.DisplayFormats.Hotpath,
            'json': statprof.DisplayFormats.Json,
            'chrome': statprof.DisplayFormats.Chrome,
        }

        if profformat in formats:
            displayformat = formats[profformat]
        else:
            ui.warn(_('unknown profiler output format: %s\n') % profformat)
            displayformat = statprof.DisplayFormats.Hotpath

        kwargs = {}

        def fraction(s):
            if isinstance(s, (float, int)):
                return float(s)
            if s.endswith('%'):
                v = float(s[:-1]) / 100
            else:
                v = float(s)
            if 0 <= v <= 1:
                return v
            raise ValueError(s)

        if profformat == 'chrome':
            showmin = ui.configwith(fraction, 'profiling', 'showmin', 0.005)
            showmax = ui.configwith(fraction, 'profiling', 'showmax')
            kwargs.update(minthreshold=showmin, maxthreshold=showmax)
        elif profformat == 'hotpath':
            # inconsistent config: profiling.showmin
            limit = ui.configwith(fraction, 'profiling', 'showmin', 0.05)
            kwargs['limit'] = limit

        statprof.display(fp, data=data, format=displayformat, **kwargs)

class profile(object):
    """Start profiling.

    Profiling is active when the context manager is active. When the context
    manager exits, profiling results will be written to the configured output.
    """
    def __init__(self, ui, enabled=True):
        self._ui = ui
        self._output = None
        self._fp = None
        self._fpdoclose = True
        self._profiler = None
        self._enabled = enabled
        self._entered = False
        self._started = False

    def __enter__(self):
        self._entered = True
        if self._enabled:
            self.start()
        return self

    def start(self):
        """Start profiling.

        The profiling will stop at the context exit.

        If the profiler was already started, this has no effect."""
        if not self._entered:
            raise error.ProgrammingError()
        if self._started:
            return
        self._started = True
        profiler = encoding.environ.get('HGPROF')
        proffn = None
        if profiler is None:
            profiler = self._ui.config('profiling', 'type')
        if profiler not in ('ls', 'stat', 'flame'):
            # try load profiler from extension with the same name
            proffn = _loadprofiler(self._ui, profiler)
            if proffn is None:
                self._ui.warn(_("unrecognized profiler '%s' - ignored\n")
                              % profiler)
                profiler = 'stat'

        self._output = self._ui.config('profiling', 'output')

        try:
            if self._output == 'blackbox':
                self._fp = util.stringio()
            elif self._output:
                path = self._ui.expandpath(self._output)
                self._fp = open(path, 'wb')
            else:
                self._fpdoclose = False
                self._fp = self._ui.ferr

            if proffn is not None:
                pass
            elif profiler == 'ls':
                proffn = lsprofile
            elif profiler == 'flame':
                proffn = flameprofile
            else:
                proffn = statprofile

            self._profiler = proffn(self._ui, self._fp)
            self._profiler.__enter__()
        except: # re-raises
            self._closefp()
            raise

    def __exit__(self, exception_type, exception_value, traceback):
        propagate = None
        if self._profiler is not None:
            propagate = self._profiler.__exit__(exception_type, exception_value,
                                                traceback)
            if self._output == 'blackbox':
                val = 'Profile:\n%s' % self._fp.getvalue()
                # ui.log treats the input as a format string,
                # so we need to escape any % signs.
                val = val.replace('%', '%%')
                self._ui.log('profile', val)
        self._closefp()
        return propagate

    def _closefp(self):
        if self._fpdoclose and self._fp is not None:
            self._fp.close()