sparse-revlog: implement algorithm to write sparse delta chains (
issue5480)
The classic behavior of revlog._isgooddeltainfo is to consider the span size
of the whole delta chain, and limit it to 4 * textlen.
Once sparse-revlog writing is allowed (and enforced with a requirement),
revlog._isgooddeltainfo considers the span of the largest chunk as the
distance used in the verification, instead of using the span of the whole
delta chain.
In order to compute the span of the largest chunk, we need to slice into
chunks a chain with the new revision at the top of the revlog, and take the
maximal span of these chunks. The sparse read density is a parameter to the
slicing, as it will stop when the global read density reaches this threshold.
For instance, a density of 50% means that 2 of 4 read bytes are actually used
for the reconstruction of the revision (the others are part of other chains).
This allows a new revision to be potentially stored with a diff against
another revision anywhere in the history, instead of forcing it in the last 4
* textlen. The result is a much better compression on repositories that have
many concurrent branches. Here are a comparison between using deltas from
current upstream (aggressive-merge-deltas on by default) and deltas from a
sparse-revlog
Comparison of `.hg/store/` size:
mercurial (6.74% merges):
before: 46,831,873 bytes
after: 46,795,992 bytes (no relevant change)
pypy (8.30% merges):
before: 333,524,651 bytes
after: 308,417,511 bytes -8%
netbeans (34.21% merges):
before: 1,141,847,554 bytes
after: 1,131,093,161 bytes -1%
mozilla-central (4.84% merges):
before: 2,344,248,850 bytes
after: 2,328,459,258 bytes -1%
large-private-repo-A (merge 19.73%)
before: 41,510,550,163 bytes
after: 8,121,763,428 bytes -80%
large-private-repo-B (23.77%)
before: 58,702,221,709 bytes
after: 8,351,588,828 bytes -76%
Comparison of `00manifest.d` size:
mercurial (6.74% merges):
before: 6,143,044 bytes
after: 6,107,163 bytes
pypy (8.30% merges):
before: 52,941,780 bytes
after: 27,834,082 bytes -48%
netbeans (34.21% merges):
before: 130,088,982 bytes
after: 119,337,636 bytes -10%
mozilla-central (4.84% merges):
before: 215,096,339 bytes
after: 199,496,863 bytes -8%
large-private-repo-A (merge 19.73%)
before: 33,725,285,081 bytes
after: 390,302,545 bytes -99%
large-private-repo-B (23.77%)
before: 49,457,701,645 bytes
after: 1,366,752,187 bytes -97%
The better delta chains provide a performance boost in relevant repositories:
pypy, bundling 1000 revisions:
before: 1.670s
after: 1.149s -31%
Unbundling got a bit slower. probably because the sparse algorithm is still
pure
python.
pypy, unbundling 1000 revisions:
before: 4.062s
after: 4.507s +10%
Performance of bundle/unbundle in repository with few concurrent branches (eg:
mercurial) are unaffected.
No significant differences have been noticed then timing `hg push` and `hg
pull` locally. More state timings are being gathered.
Same as for aggressive-merge-delta, better delta comes with longer delta
chains. Longer chains have a performance impact. For example. The length of
the chain needed to get the manifest of pypy's tip moves from 82 item to 1929
items. This moves the restore time from 3.88ms to 11.3ms.
Delta chain length is an independent issue that affects repository without
this changes. It will be dealt with independently.
No significant differences have been observed on repositories where
`sparse-revlog` have not much effect (mercurial, unity, netbeans). On pypy,
small differences have been observed on some operation affected by delta chain
building and retrieval.
pypy, perfmanifest
before: 0.006162s
after: 0.017899s +190%
pypy, commit:
before: 0.382
after: 0.376 -1%
pypy, status:
before: 0.157
after: 0.168 +7%
More comprehensive and stable timing comparisons are in progress.
# show.py - Extension implementing `hg show`
#
# Copyright 2017 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.
"""unified command to show various repository information (EXPERIMENTAL)
This extension provides the :hg:`show` command, which provides a central
command for displaying commonly-accessed repository data and views of that
data.
The following config options can influence operation.
``commands``
------------
``show.aliasprefix``
List of strings that will register aliases for views. e.g. ``s`` will
effectively set config options ``alias.s<view> = show <view>`` for all
views. i.e. `hg swork` would execute `hg show work`.
Aliases that would conflict with existing registrations will not be
performed.
"""
from __future__ import absolute_import
from mercurial.i18n import _
from mercurial.node import (
nullrev,
)
from mercurial import (
cmdutil,
commands,
destutil,
error,
formatter,
graphmod,
logcmdutil,
phases,
pycompat,
registrar,
revset,
revsetlang,
scmutil,
)
# Note for extension authors: ONLY specify testedwith = 'ships-with-hg-core' for
# extensions which SHIP WITH MERCURIAL. Non-mainline extensions should
# be specifying the version(s) of Mercurial they are tested with, or
# leave the attribute unspecified.
testedwith = 'ships-with-hg-core'
cmdtable = {}
command = registrar.command(cmdtable)
revsetpredicate = registrar.revsetpredicate()
class showcmdfunc(registrar._funcregistrarbase):
"""Register a function to be invoked for an `hg show <thing>`."""
# Used by _formatdoc().
_docformat = '%s -- %s'
def _extrasetup(self, name, func, fmtopic=None, csettopic=None):
"""Called with decorator arguments to register a show view.
``name`` is the sub-command name.
``func`` is the function being decorated.
``fmtopic`` is the topic in the style that will be rendered for
this view.
``csettopic`` is the topic in the style to be used for a changeset
printer.
If ``fmtopic`` is specified, the view function will receive a
formatter instance. If ``csettopic`` is specified, the view
function will receive a changeset printer.
"""
func._fmtopic = fmtopic
func._csettopic = csettopic
showview = showcmdfunc()
@command('show', [
# TODO: Switch this template flag to use cmdutil.formatteropts if
# 'hg show' becomes stable before --template/-T is stable. For now,
# we are putting it here without the '(EXPERIMENTAL)' flag because it
# is an important part of the 'hg show' user experience and the entire
# 'hg show' experience is experimental.
('T', 'template', '', ('display with template'), _('TEMPLATE')),
], _('VIEW'))
def show(ui, repo, view=None, template=None):
"""show various repository information
A requested view of repository data is displayed.
If no view is requested, the list of available views is shown and the
command aborts.
.. note::
There are no backwards compatibility guarantees for the output of this
command. Output may change in any future Mercurial release.
Consumers wanting stable command output should specify a template via
``-T/--template``.
List of available views:
"""
if ui.plain() and not template:
hint = _('invoke with -T/--template to control output format')
raise error.Abort(_('must specify a template in plain mode'), hint=hint)
views = showview._table
if not view:
ui.pager('show')
# TODO consider using formatter here so available views can be
# rendered to custom format.
ui.write(_('available views:\n'))
ui.write('\n')
for name, func in sorted(views.items()):
ui.write(('%s\n') % pycompat.sysbytes(func.__doc__))
ui.write('\n')
raise error.Abort(_('no view requested'),
hint=_('use "hg show VIEW" to choose a view'))
# TODO use same logic as dispatch to perform prefix matching.
if view not in views:
raise error.Abort(_('unknown view: %s') % view,
hint=_('run "hg show" to see available views'))
template = template or 'show'
fn = views[view]
ui.pager('show')
if fn._fmtopic:
fmtopic = 'show%s' % fn._fmtopic
with ui.formatter(fmtopic, {'template': template}) as fm:
return fn(ui, repo, fm)
elif fn._csettopic:
ref = 'show%s' % fn._csettopic
spec = formatter.lookuptemplate(ui, ref, template)
displayer = logcmdutil.changesettemplater(ui, repo, spec, buffered=True)
return fn(ui, repo, displayer)
else:
return fn(ui, repo)
@showview('bookmarks', fmtopic='bookmarks')
def showbookmarks(ui, repo, fm):
"""bookmarks and their associated changeset"""
marks = repo._bookmarks
if not len(marks):
# This is a bit hacky. Ideally, templates would have a way to
# specify an empty output, but we shouldn't corrupt JSON while
# waiting for this functionality.
if not isinstance(fm, formatter.jsonformatter):
ui.write(_('(no bookmarks set)\n'))
return
revs = [repo[node].rev() for node in marks.values()]
active = repo._activebookmark
longestname = max(len(b) for b in marks)
nodelen = longestshortest(repo, revs)
for bm, node in sorted(marks.items()):
fm.startitem()
fm.context(ctx=repo[node])
fm.write('bookmark', '%s', bm)
fm.write('node', fm.hexfunc(node), fm.hexfunc(node))
fm.data(active=bm == active,
longestbookmarklen=longestname,
nodelen=nodelen)
@showview('stack', csettopic='stack')
def showstack(ui, repo, displayer):
"""current line of work"""
wdirctx = repo['.']
if wdirctx.rev() == nullrev:
raise error.Abort(_('stack view only available when there is a '
'working directory'))
if wdirctx.phase() == phases.public:
ui.write(_('(empty stack; working directory parent is a published '
'changeset)\n'))
return
# TODO extract "find stack" into a function to facilitate
# customization and reuse.
baserev = destutil.stackbase(ui, repo)
basectx = None
if baserev is None:
baserev = wdirctx.rev()
stackrevs = {wdirctx.rev()}
else:
stackrevs = set(repo.revs('%d::.', baserev))
ctx = repo[baserev]
if ctx.p1().rev() != nullrev:
basectx = ctx.p1()
# And relevant descendants.
branchpointattip = False
cl = repo.changelog
for rev in cl.descendants([wdirctx.rev()]):
ctx = repo[rev]
# Will only happen if . is public.
if ctx.phase() == phases.public:
break
stackrevs.add(ctx.rev())
# ctx.children() within a function iterating on descandants
# potentially has severe performance concerns because revlog.children()
# iterates over all revisions after ctx's node. However, the number of
# draft changesets should be a reasonably small number. So even if
# this is quadratic, the perf impact should be minimal.
if len(ctx.children()) > 1:
branchpointattip = True
break
stackrevs = list(sorted(stackrevs, reverse=True))
# Find likely target heads for the current stack. These are likely
# merge or rebase targets.
if basectx:
# TODO make this customizable?
newheads = set(repo.revs('heads(%d::) - %ld - not public()',
basectx.rev(), stackrevs))
else:
newheads = set()
allrevs = set(stackrevs) | newheads | set([baserev])
nodelen = longestshortest(repo, allrevs)
try:
cmdutil.findcmd('rebase', commands.table)
haverebase = True
except (error.AmbiguousCommand, error.UnknownCommand):
haverebase = False
# TODO use templating.
# TODO consider using graphmod. But it may not be necessary given
# our simplicity and the customizations required.
# TODO use proper graph symbols from graphmod
tres = formatter.templateresources(ui, repo)
shortesttmpl = formatter.maketemplater(ui, '{shortest(node, %d)}' % nodelen,
resources=tres)
def shortest(ctx):
return shortesttmpl.renderdefault({'ctx': ctx, 'node': ctx.hex()})
# We write out new heads to aid in DAG awareness and to help with decision
# making on how the stack should be reconciled with commits made since the
# branch point.
if newheads:
# Calculate distance from base so we can render the count and so we can
# sort display order by commit distance.
revdistance = {}
for head in newheads:
# There is some redundancy in DAG traversal here and therefore
# room to optimize.
ancestors = cl.ancestors([head], stoprev=basectx.rev())
revdistance[head] = len(list(ancestors))
sourcectx = repo[stackrevs[-1]]
sortedheads = sorted(newheads, key=lambda x: revdistance[x],
reverse=True)
for i, rev in enumerate(sortedheads):
ctx = repo[rev]
if i:
ui.write(': ')
else:
ui.write(' ')
ui.write(('o '))
displayer.show(ctx, nodelen=nodelen)
displayer.flush(ctx)
ui.write('\n')
if i:
ui.write(':/')
else:
ui.write(' /')
ui.write(' (')
ui.write(_('%d commits ahead') % revdistance[rev],
label='stack.commitdistance')
if haverebase:
# TODO may be able to omit --source in some scenarios
ui.write('; ')
ui.write(('hg rebase --source %s --dest %s' % (
shortest(sourcectx), shortest(ctx))),
label='stack.rebasehint')
ui.write(')\n')
ui.write(':\n: ')
ui.write(_('(stack head)\n'), label='stack.label')
if branchpointattip:
ui.write(' \\ / ')
ui.write(_('(multiple children)\n'), label='stack.label')
ui.write(' |\n')
for rev in stackrevs:
ctx = repo[rev]
symbol = '@' if rev == wdirctx.rev() else 'o'
if newheads:
ui.write(': ')
else:
ui.write(' ')
ui.write(symbol, ' ')
displayer.show(ctx, nodelen=nodelen)
displayer.flush(ctx)
ui.write('\n')
# TODO display histedit hint?
if basectx:
# Vertically and horizontally separate stack base from parent
# to reinforce stack boundary.
if newheads:
ui.write(':/ ')
else:
ui.write(' / ')
ui.write(_('(stack base)'), '\n', label='stack.label')
ui.write(('o '))
displayer.show(basectx, nodelen=nodelen)
displayer.flush(basectx)
ui.write('\n')
@revsetpredicate('_underway([commitage[, headage]])')
def underwayrevset(repo, subset, x):
args = revset.getargsdict(x, 'underway', 'commitage headage')
if 'commitage' not in args:
args['commitage'] = None
if 'headage' not in args:
args['headage'] = None
# We assume callers of this revset add a topographical sort on the
# result. This means there is no benefit to making the revset lazy
# since the topographical sort needs to consume all revs.
#
# With this in mind, we build up the set manually instead of constructing
# a complex revset. This enables faster execution.
# Mutable changesets (non-public) are the most important changesets
# to return. ``not public()`` will also pull in obsolete changesets if
# there is a non-obsolete changeset with obsolete ancestors. This is
# why we exclude obsolete changesets from this query.
rs = 'not public() and not obsolete()'
rsargs = []
if args['commitage']:
rs += ' and date(%s)'
rsargs.append(revsetlang.getstring(args['commitage'],
_('commitage requires a string')))
mutable = repo.revs(rs, *rsargs)
relevant = revset.baseset(mutable)
# Add parents of mutable changesets to provide context.
relevant += repo.revs('parents(%ld)', mutable)
# We also pull in (public) heads if they a) aren't closing a branch
# b) are recent.
rs = 'head() and not closed()'
rsargs = []
if args['headage']:
rs += ' and date(%s)'
rsargs.append(revsetlang.getstring(args['headage'],
_('headage requires a string')))
relevant += repo.revs(rs, *rsargs)
# Add working directory parent.
wdirrev = repo['.'].rev()
if wdirrev != nullrev:
relevant += revset.baseset({wdirrev})
return subset & relevant
@showview('work', csettopic='work')
def showwork(ui, repo, displayer):
"""changesets that aren't finished"""
# TODO support date-based limiting when calling revset.
revs = repo.revs('sort(_underway(), topo)')
nodelen = longestshortest(repo, revs)
revdag = graphmod.dagwalker(repo, revs)
ui.setconfig('experimental', 'graphshorten', True)
logcmdutil.displaygraph(ui, repo, revdag, displayer, graphmod.asciiedges,
props={'nodelen': nodelen})
def extsetup(ui):
# Alias `hg <prefix><view>` to `hg show <view>`.
for prefix in ui.configlist('commands', 'show.aliasprefix'):
for view in showview._table:
name = '%s%s' % (prefix, view)
choice, allcommands = cmdutil.findpossible(name, commands.table,
strict=True)
# This alias is already a command name. Don't set it.
if name in choice:
continue
# Same for aliases.
if ui.config('alias', name, None):
continue
ui.setconfig('alias', name, 'show %s' % view, source='show')
def longestshortest(repo, revs, minlen=4):
"""Return the length of the longest shortest node to identify revisions.
The result of this function can be used with the ``shortest()`` template
function to ensure that a value is unique and unambiguous for a given
set of nodes.
The number of revisions in the repo is taken into account to prevent
a numeric node prefix from conflicting with an integer revision number.
If we fail to do this, a value of e.g. ``10023`` could mean either
revision 10023 or node ``10023abc...``.
"""
if not revs:
return minlen
cl = repo.changelog
return max(len(scmutil.shortesthexnodeidprefix(repo, cl.node(r), minlen))
for r in revs)
# Adjust the docstring of the show command so it shows all registered views.
# This is a bit hacky because it runs at the end of module load. When moved
# into core or when another extension wants to provide a view, we'll need
# to do this more robustly.
# TODO make this more robust.
def _updatedocstring():
longest = max(map(len, showview._table.keys()))
entries = []
for key in sorted(showview._table.keys()):
entries.append(pycompat.sysstr(' %s %s' % (
key.ljust(longest), showview._table[key]._origdoc)))
cmdtable['show'][0].__doc__ = pycompat.sysstr('%s\n\n%s\n ') % (
cmdtable['show'][0].__doc__.rstrip(),
pycompat.sysstr('\n\n').join(entries))
_updatedocstring()