view mercurial/bundlerepo.py @ 44950:f9734b2d59cc

py3: make stdout line-buffered if connected to a TTY

Status messages that are to be shown on the terminal should be written to the
file descriptor before anything further is done, to keep the user updated.

One common way to achieve this is to make stdout line-buffered if it is
connected to a TTY. This is done on Python 2 (except on Windows, where libc,
which the CPython 2 streams depend on, does not properly support this).

Python 3 rolls it own I/O streams. On Python 3, buffered binary streams can't be
set line-buffered. The previous code (added in 227ba1afcb65) incorrectly
assumed that on Python 3, pycompat.stdout (sys.stdout.buffer) is already
line-buffered. However the interpreter initializes it with a block-buffered
stream or an unbuffered stream (when the -u option or the PYTHONUNBUFFERED
environment variable is set), never with a line-buffered stream.

One example where the current behavior is unacceptable is when running
`hg pull https://www.mercurial-scm.org/repo/hg` on Python 3, where the line
"pulling from https://www.mercurial-scm.org/repo/hg" does not appear on the
terminal before the hg process blocks while waiting for the server.

Various approaches to fix this problem are possible, including:

1. Weaken the contract of procutil.stdout to not give any guarantees about
   buffering behavior. In this case, users of procutil.stdout need to be
   changed to do enough flushes. In particular,
  1. either ui must insert enough flushes for ui.write() and friends, or
  2. ui.write() and friends get split into flushing and fully buffered
     methods, or
  3. users of ui.write() and friends must flush explicitly.
2. Make stdout unbuffered.
3. Make stdout line-buffered. Since Python 3 does not natively support that for
   binary streams, we must implement it ourselves.

(2.) is problematic because using unbuffered I/O changes the performance
characteristics significantly compared to line-buffered (which is used on
Python 2) and this would be a regression.

(1.2.) and (1.3) are a substantial amount of work. It’s unclear whether the
added complexity would be justified, given that raw performance doesn’t matter
that much when writing to a terminal much faster than the user could read it.

(1.1.) pushes complexity into the ui class instead of separating the concern of
how stdout is buffered. Other users of procutil.stdout would still need to take
care of the flushes.

This patch implements (3.). The general performance considerations are very
similar to (1.1.). The extra method invocation and method forwarding add a
little more overhead if the class is used. In exchange, it doesn’t add overhead
if not used.


For the benchmarks, I compared the previous implementation (incorrect on Python
3), (1.1.), (3.) and (2.). The command was chosen so that the streams were
configured as if they were writing to a TTY, but actually write to a pager,
which is also the default:
HGRCPATH=/dev/null python3 ./hg --cwd ~/vcs/mozilla-central --time --pager yes --config pager.pager='cat > /dev/null' status --all

previous:
time: real 7.880 secs (user 7.290+0.050 sys 0.580+0.170)
time: real 7.830 secs (user 7.220+0.070 sys 0.590+0.140)
time: real 7.800 secs (user 7.210+0.050 sys 0.570+0.170)

(1.1.) using Yuya Nishihara’s patch:
time: real 9.860 secs (user 8.670+0.350 sys 1.160+0.830)
time: real 9.540 secs (user 8.430+0.370 sys 1.100+0.770)
time: real 9.830 secs (user 8.630+0.370 sys 1.180+0.840)

(3.) using this patch:
time: real 9.580 secs (user 8.480+0.350 sys 1.090+0.770)
time: real 9.670 secs (user 8.480+0.330 sys 1.170+0.860)
time: real 9.640 secs (user 8.500+0.350 sys 1.130+0.810)

(2.) using a previous patch by me:
time: real 10.480 secs (user 8.850+0.720 sys 1.590+1.500)
time: real 10.490 secs (user 8.750+0.750 sys 1.710+1.470)
time: real 10.240 secs (user 8.600+0.700 sys 1.590+1.510)


As expected, there’s no difference on Python 2, as exactly the same code paths
are used:

previous:
time: real 6.950 secs (user 5.870+0.330 sys 1.070+0.770)
time: real 7.040 secs (user 6.040+0.360 sys 0.980+0.750)
time: real 7.070 secs (user 5.950+0.360 sys 1.100+0.760)

this patch:
time: real 7.010 secs (user 5.900+0.390 sys 1.070+0.730)
time: real 7.000 secs (user 5.850+0.350 sys 1.120+0.760)
time: real 7.000 secs (user 5.790+0.380 sys 1.170+0.710)

# bundlerepo.py - repository class for viewing uncompressed bundles
#
# Copyright 2006, 2007 Benoit Boissinot <[e-mail]>
#
# This software may be used and distributed according to the terms of the
# GNU General Public License version 2 or any later version.

"""Repository class for viewing uncompressed bundles.

This provides a read-only repository interface to bundles as if they
were part of the actual repository.
"""

from __future__ import absolute_import

import os
import shutil

from .i18n import _
from .node import nullid, nullrev

from . import (
    bundle2,
    changegroup,
    changelog,
    cmdutil,
    discovery,
    encoding,
    error,
    exchange,
    filelog,
    localrepo,
    manifest,
    mdiff,
    node as nodemod,
    pathutil,
    phases,
    pycompat,
    revlog,
    util,
    vfs as vfsmod,
)


class bundlerevlog(revlog.revlog):
    def __init__(self, opener, indexfile, cgunpacker, linkmapper):
        # How it works:
        # To retrieve a revision, we need to know the offset of the revision in
        # the bundle (an unbundle object). We store this offset in the index
        # (start). The base of the delta is stored in the base field.
        #
        # To differentiate a rev in the bundle from a rev in the revlog, we
        # check revision against repotiprev.
        opener = vfsmod.readonlyvfs(opener)
        revlog.revlog.__init__(self, opener, indexfile)
        self.bundle = cgunpacker
        n = len(self)
        self.repotiprev = n - 1
        self.bundlerevs = set()  # used by 'bundle()' revset expression
        for deltadata in cgunpacker.deltaiter():
            node, p1, p2, cs, deltabase, delta, flags = deltadata

            size = len(delta)
            start = cgunpacker.tell() - size

            link = linkmapper(cs)
            if self.index.has_node(node):
                # this can happen if two branches make the same change
                self.bundlerevs.add(self.index.rev(node))
                continue

            for p in (p1, p2):
                if not self.index.has_node(p):
                    raise error.LookupError(
                        p, self.indexfile, _(b"unknown parent")
                    )

            if not self.index.has_node(deltabase):
                raise LookupError(
                    deltabase, self.indexfile, _(b'unknown delta base')
                )

            baserev = self.rev(deltabase)
            # start, size, full unc. size, base (unused), link, p1, p2, node
            e = (
                revlog.offset_type(start, flags),
                size,
                -1,
                baserev,
                link,
                self.rev(p1),
                self.rev(p2),
                node,
            )
            self.index.append(e)
            self.bundlerevs.add(n)
            n += 1

    def _chunk(self, rev, df=None):
        # Warning: in case of bundle, the diff is against what we stored as
        # delta base, not against rev - 1
        # XXX: could use some caching
        if rev <= self.repotiprev:
            return revlog.revlog._chunk(self, rev)
        self.bundle.seek(self.start(rev))
        return self.bundle.read(self.length(rev))

    def revdiff(self, rev1, rev2):
        """return or calculate a delta between two revisions"""
        if rev1 > self.repotiprev and rev2 > self.repotiprev:
            # hot path for bundle
            revb = self.index[rev2][3]
            if revb == rev1:
                return self._chunk(rev2)
        elif rev1 <= self.repotiprev and rev2 <= self.repotiprev:
            return revlog.revlog.revdiff(self, rev1, rev2)

        return mdiff.textdiff(self.rawdata(rev1), self.rawdata(rev2))

    def _rawtext(self, node, rev, _df=None):
        if rev is None:
            rev = self.rev(node)
        validated = False
        rawtext = None
        chain = []
        iterrev = rev
        # reconstruct the revision if it is from a changegroup
        while iterrev > self.repotiprev:
            if self._revisioncache and self._revisioncache[1] == iterrev:
                rawtext = self._revisioncache[2]
                break
            chain.append(iterrev)
            iterrev = self.index[iterrev][3]
        if iterrev == nullrev:
            rawtext = b''
        elif rawtext is None:
            r = super(bundlerevlog, self)._rawtext(
                self.node(iterrev), iterrev, _df=_df
            )
            __, rawtext, validated = r
        if chain:
            validated = False
        while chain:
            delta = self._chunk(chain.pop())
            rawtext = mdiff.patches(rawtext, [delta])
        return rev, rawtext, validated

    def addrevision(self, *args, **kwargs):
        raise NotImplementedError

    def addgroup(self, *args, **kwargs):
        raise NotImplementedError

    def strip(self, *args, **kwargs):
        raise NotImplementedError

    def checksize(self):
        raise NotImplementedError


class bundlechangelog(bundlerevlog, changelog.changelog):
    def __init__(self, opener, cgunpacker):
        changelog.changelog.__init__(self, opener)
        linkmapper = lambda x: x
        bundlerevlog.__init__(
            self, opener, self.indexfile, cgunpacker, linkmapper
        )


class bundlemanifest(bundlerevlog, manifest.manifestrevlog):
    def __init__(
        self, opener, cgunpacker, linkmapper, dirlogstarts=None, dir=b''
    ):
        manifest.manifestrevlog.__init__(self, opener, tree=dir)
        bundlerevlog.__init__(
            self, opener, self.indexfile, cgunpacker, linkmapper
        )
        if dirlogstarts is None:
            dirlogstarts = {}
            if self.bundle.version == b"03":
                dirlogstarts = _getfilestarts(self.bundle)
        self._dirlogstarts = dirlogstarts
        self._linkmapper = linkmapper

    def dirlog(self, d):
        if d in self._dirlogstarts:
            self.bundle.seek(self._dirlogstarts[d])
            return bundlemanifest(
                self.opener,
                self.bundle,
                self._linkmapper,
                self._dirlogstarts,
                dir=d,
            )
        return super(bundlemanifest, self).dirlog(d)


class bundlefilelog(filelog.filelog):
    def __init__(self, opener, path, cgunpacker, linkmapper):
        filelog.filelog.__init__(self, opener, path)
        self._revlog = bundlerevlog(
            opener, self.indexfile, cgunpacker, linkmapper
        )


class bundlepeer(localrepo.localpeer):
    def canpush(self):
        return False


class bundlephasecache(phases.phasecache):
    def __init__(self, *args, **kwargs):
        super(bundlephasecache, self).__init__(*args, **kwargs)
        if util.safehasattr(self, 'opener'):
            self.opener = vfsmod.readonlyvfs(self.opener)

    def write(self):
        raise NotImplementedError

    def _write(self, fp):
        raise NotImplementedError

    def _updateroots(self, phase, newroots, tr):
        self.phaseroots[phase] = newroots
        self.invalidate()
        self.dirty = True


def _getfilestarts(cgunpacker):
    filespos = {}
    for chunkdata in iter(cgunpacker.filelogheader, {}):
        fname = chunkdata[b'filename']
        filespos[fname] = cgunpacker.tell()
        for chunk in iter(lambda: cgunpacker.deltachunk(None), {}):
            pass
    return filespos


class bundlerepository(object):
    """A repository instance that is a union of a local repo and a bundle.

    Instances represent a read-only repository composed of a local repository
    with the contents of a bundle file applied. The repository instance is
    conceptually similar to the state of a repository after an
    ``hg unbundle`` operation. However, the contents of the bundle are never
    applied to the actual base repository.

    Instances constructed directly are not usable as repository objects.
    Use instance() or makebundlerepository() to create instances.
    """

    def __init__(self, bundlepath, url, tempparent):
        self._tempparent = tempparent
        self._url = url

        self.ui.setconfig(b'phases', b'publish', False, b'bundlerepo')

        self.tempfile = None
        f = util.posixfile(bundlepath, b"rb")
        bundle = exchange.readbundle(self.ui, f, bundlepath)

        if isinstance(bundle, bundle2.unbundle20):
            self._bundlefile = bundle
            self._cgunpacker = None

            cgpart = None
            for part in bundle.iterparts(seekable=True):
                if part.type == b'changegroup':
                    if cgpart:
                        raise NotImplementedError(
                            b"can't process multiple changegroups"
                        )
                    cgpart = part

                self._handlebundle2part(bundle, part)

            if not cgpart:
                raise error.Abort(_(b"No changegroups found"))

            # This is required to placate a later consumer, which expects
            # the payload offset to be at the beginning of the changegroup.
            # We need to do this after the iterparts() generator advances
            # because iterparts() will seek to end of payload after the
            # generator returns control to iterparts().
            cgpart.seek(0, os.SEEK_SET)

        elif isinstance(bundle, changegroup.cg1unpacker):
            if bundle.compressed():
                f = self._writetempbundle(
                    bundle.read, b'.hg10un', header=b'HG10UN'
                )
                bundle = exchange.readbundle(self.ui, f, bundlepath, self.vfs)

            self._bundlefile = bundle
            self._cgunpacker = bundle
        else:
            raise error.Abort(
                _(b'bundle type %s cannot be read') % type(bundle)
            )

        # dict with the mapping 'filename' -> position in the changegroup.
        self._cgfilespos = {}

        self.firstnewrev = self.changelog.repotiprev + 1
        phases.retractboundary(
            self,
            None,
            phases.draft,
            [ctx.node() for ctx in self[self.firstnewrev :]],
        )

    def _handlebundle2part(self, bundle, part):
        if part.type != b'changegroup':
            return

        cgstream = part
        version = part.params.get(b'version', b'01')
        legalcgvers = changegroup.supportedincomingversions(self)
        if version not in legalcgvers:
            msg = _(b'Unsupported changegroup version: %s')
            raise error.Abort(msg % version)
        if bundle.compressed():
            cgstream = self._writetempbundle(part.read, b'.cg%sun' % version)

        self._cgunpacker = changegroup.getunbundler(version, cgstream, b'UN')

    def _writetempbundle(self, readfn, suffix, header=b''):
        """Write a temporary file to disk
        """
        fdtemp, temp = self.vfs.mkstemp(prefix=b"hg-bundle-", suffix=suffix)
        self.tempfile = temp

        with os.fdopen(fdtemp, 'wb') as fptemp:
            fptemp.write(header)
            while True:
                chunk = readfn(2 ** 18)
                if not chunk:
                    break
                fptemp.write(chunk)

        return self.vfs.open(self.tempfile, mode=b"rb")

    @localrepo.unfilteredpropertycache
    def _phasecache(self):
        return bundlephasecache(self, self._phasedefaults)

    @localrepo.unfilteredpropertycache
    def changelog(self):
        # consume the header if it exists
        self._cgunpacker.changelogheader()
        c = bundlechangelog(self.svfs, self._cgunpacker)
        self.manstart = self._cgunpacker.tell()
        return c

    def _refreshchangelog(self):
        # changelog for bundle repo are not filecache, this method is not
        # applicable.
        pass

    @localrepo.unfilteredpropertycache
    def manifestlog(self):
        self._cgunpacker.seek(self.manstart)
        # consume the header if it exists
        self._cgunpacker.manifestheader()
        linkmapper = self.unfiltered().changelog.rev
        rootstore = bundlemanifest(self.svfs, self._cgunpacker, linkmapper)
        self.filestart = self._cgunpacker.tell()

        return manifest.manifestlog(
            self.svfs, self, rootstore, self.narrowmatch()
        )

    def _consumemanifest(self):
        """Consumes the manifest portion of the bundle, setting filestart so the
        file portion can be read."""
        self._cgunpacker.seek(self.manstart)
        self._cgunpacker.manifestheader()
        for delta in self._cgunpacker.deltaiter():
            pass
        self.filestart = self._cgunpacker.tell()

    @localrepo.unfilteredpropertycache
    def manstart(self):
        self.changelog
        return self.manstart

    @localrepo.unfilteredpropertycache
    def filestart(self):
        self.manifestlog

        # If filestart was not set by self.manifestlog, that means the
        # manifestlog implementation did not consume the manifests from the
        # changegroup (ex: it might be consuming trees from a separate bundle2
        # part instead). So we need to manually consume it.
        if 'filestart' not in self.__dict__:
            self._consumemanifest()

        return self.filestart

    def url(self):
        return self._url

    def file(self, f):
        if not self._cgfilespos:
            self._cgunpacker.seek(self.filestart)
            self._cgfilespos = _getfilestarts(self._cgunpacker)

        if f in self._cgfilespos:
            self._cgunpacker.seek(self._cgfilespos[f])
            linkmapper = self.unfiltered().changelog.rev
            return bundlefilelog(self.svfs, f, self._cgunpacker, linkmapper)
        else:
            return super(bundlerepository, self).file(f)

    def close(self):
        """Close assigned bundle file immediately."""
        self._bundlefile.close()
        if self.tempfile is not None:
            self.vfs.unlink(self.tempfile)
        if self._tempparent:
            shutil.rmtree(self._tempparent, True)

    def cancopy(self):
        return False

    def peer(self):
        return bundlepeer(self)

    def getcwd(self):
        return encoding.getcwd()  # always outside the repo

    # Check if parents exist in localrepo before setting
    def setparents(self, p1, p2=nullid):
        p1rev = self.changelog.rev(p1)
        p2rev = self.changelog.rev(p2)
        msg = _(b"setting parent to node %s that only exists in the bundle\n")
        if self.changelog.repotiprev < p1rev:
            self.ui.warn(msg % nodemod.hex(p1))
        if self.changelog.repotiprev < p2rev:
            self.ui.warn(msg % nodemod.hex(p2))
        return super(bundlerepository, self).setparents(p1, p2)


def instance(ui, path, create, intents=None, createopts=None):
    if create:
        raise error.Abort(_(b'cannot create new bundle repository'))
    # internal config: bundle.mainreporoot
    parentpath = ui.config(b"bundle", b"mainreporoot")
    if not parentpath:
        # try to find the correct path to the working directory repo
        parentpath = cmdutil.findrepo(encoding.getcwd())
        if parentpath is None:
            parentpath = b''
    if parentpath:
        # Try to make the full path relative so we get a nice, short URL.
        # In particular, we don't want temp dir names in test outputs.
        cwd = encoding.getcwd()
        if parentpath == cwd:
            parentpath = b''
        else:
            cwd = pathutil.normasprefix(cwd)
            if parentpath.startswith(cwd):
                parentpath = parentpath[len(cwd) :]
    u = util.url(path)
    path = u.localpath()
    if u.scheme == b'bundle':
        s = path.split(b"+", 1)
        if len(s) == 1:
            repopath, bundlename = parentpath, s[0]
        else:
            repopath, bundlename = s
    else:
        repopath, bundlename = parentpath, path

    return makebundlerepository(ui, repopath, bundlename)


def makebundlerepository(ui, repopath, bundlepath):
    """Make a bundle repository object based on repo and bundle paths."""
    if repopath:
        url = b'bundle:%s+%s' % (util.expandpath(repopath), bundlepath)
    else:
        url = b'bundle:%s' % bundlepath

    # Because we can't make any guarantees about the type of the base
    # repository, we can't have a static class representing the bundle
    # repository. We also can't make any guarantees about how to even
    # call the base repository's constructor!
    #
    # So, our strategy is to go through ``localrepo.instance()`` to construct
    # a repo instance. Then, we dynamically create a new type derived from
    # both it and our ``bundlerepository`` class which overrides some
    # functionality. We then change the type of the constructed repository
    # to this new type and initialize the bundle-specific bits of it.

    try:
        repo = localrepo.instance(ui, repopath, create=False)
        tempparent = None
    except error.RepoError:
        tempparent = pycompat.mkdtemp()
        try:
            repo = localrepo.instance(ui, tempparent, create=True)
        except Exception:
            shutil.rmtree(tempparent)
            raise

    class derivedbundlerepository(bundlerepository, repo.__class__):
        pass

    repo.__class__ = derivedbundlerepository
    bundlerepository.__init__(repo, bundlepath, url, tempparent)

    return repo


class bundletransactionmanager(object):
    def transaction(self):
        return None

    def close(self):
        raise NotImplementedError

    def release(self):
        raise NotImplementedError


def getremotechanges(
    ui, repo, peer, onlyheads=None, bundlename=None, force=False
):
    '''obtains a bundle of changes incoming from peer

    "onlyheads" restricts the returned changes to those reachable from the
      specified heads.
    "bundlename", if given, stores the bundle to this file path permanently;
      otherwise it's stored to a temp file and gets deleted again when you call
      the returned "cleanupfn".
    "force" indicates whether to proceed on unrelated repos.

    Returns a tuple (local, csets, cleanupfn):

    "local" is a local repo from which to obtain the actual incoming
      changesets; it is a bundlerepo for the obtained bundle when the
      original "peer" is remote.
    "csets" lists the incoming changeset node ids.
    "cleanupfn" must be called without arguments when you're done processing
      the changes; it closes both the original "peer" and the one returned
      here.
    '''
    tmp = discovery.findcommonincoming(repo, peer, heads=onlyheads, force=force)
    common, incoming, rheads = tmp
    if not incoming:
        try:
            if bundlename:
                os.unlink(bundlename)
        except OSError:
            pass
        return repo, [], peer.close

    commonset = set(common)
    rheads = [x for x in rheads if x not in commonset]

    bundle = None
    bundlerepo = None
    localrepo = peer.local()
    if bundlename or not localrepo:
        # create a bundle (uncompressed if peer repo is not local)

        # developer config: devel.legacy.exchange
        legexc = ui.configlist(b'devel', b'legacy.exchange')
        forcebundle1 = b'bundle2' not in legexc and b'bundle1' in legexc
        canbundle2 = (
            not forcebundle1
            and peer.capable(b'getbundle')
            and peer.capable(b'bundle2')
        )
        if canbundle2:
            with peer.commandexecutor() as e:
                b2 = e.callcommand(
                    b'getbundle',
                    {
                        b'source': b'incoming',
                        b'common': common,
                        b'heads': rheads,
                        b'bundlecaps': exchange.caps20to10(
                            repo, role=b'client'
                        ),
                        b'cg': True,
                    },
                ).result()

                fname = bundle = changegroup.writechunks(
                    ui, b2._forwardchunks(), bundlename
                )
        else:
            if peer.capable(b'getbundle'):
                with peer.commandexecutor() as e:
                    cg = e.callcommand(
                        b'getbundle',
                        {
                            b'source': b'incoming',
                            b'common': common,
                            b'heads': rheads,
                        },
                    ).result()
            elif onlyheads is None and not peer.capable(b'changegroupsubset'):
                # compat with older servers when pulling all remote heads

                with peer.commandexecutor() as e:
                    cg = e.callcommand(
                        b'changegroup',
                        {b'nodes': incoming, b'source': b'incoming',},
                    ).result()

                rheads = None
            else:
                with peer.commandexecutor() as e:
                    cg = e.callcommand(
                        b'changegroupsubset',
                        {
                            b'bases': incoming,
                            b'heads': rheads,
                            b'source': b'incoming',
                        },
                    ).result()

            if localrepo:
                bundletype = b"HG10BZ"
            else:
                bundletype = b"HG10UN"
            fname = bundle = bundle2.writebundle(ui, cg, bundlename, bundletype)
        # keep written bundle?
        if bundlename:
            bundle = None
        if not localrepo:
            # use the created uncompressed bundlerepo
            localrepo = bundlerepo = makebundlerepository(
                repo.baseui, repo.root, fname
            )

            # this repo contains local and peer now, so filter out local again
            common = repo.heads()
    if localrepo:
        # Part of common may be remotely filtered
        # So use an unfiltered version
        # The discovery process probably need cleanup to avoid that
        localrepo = localrepo.unfiltered()

    csets = localrepo.changelog.findmissing(common, rheads)

    if bundlerepo:
        reponodes = [ctx.node() for ctx in bundlerepo[bundlerepo.firstnewrev :]]

        with peer.commandexecutor() as e:
            remotephases = e.callcommand(
                b'listkeys', {b'namespace': b'phases',}
            ).result()

        pullop = exchange.pulloperation(bundlerepo, peer, heads=reponodes)
        pullop.trmanager = bundletransactionmanager()
        exchange._pullapplyphases(pullop, remotephases)

    def cleanup():
        if bundlerepo:
            bundlerepo.close()
        if bundle:
            os.unlink(bundle)
        peer.close()

    return (localrepo, csets, cleanup)