sapling/eden/scm/edenscm/mercurial/mutation.py

867 lines
30 KiB
Python
Raw Normal View History

# Copyright (c) Facebook, Inc. and its affiliates.
#
# This software may be used and distributed according to the terms of the
# GNU General Public License version 2.
# mutation.py - commit mutation tracking
from __future__ import absolute_import
from collections import defaultdict
from bindings import mutationstore
from . import error, node as nodemod, perftrace, phases, repoview, util
from .i18n import _
from .node import nullid
def identfromnode(node):
return "hg/%s" % nodemod.hex(node)
def nodefromident(ident):
if ident.startswith("hg/"):
return nodemod.bin(ident[3:])
raise error.Abort("Unrecognised commit identifier: %s" % ident)
def record(repo, extra, prednodes, op=None, splitting=None):
mutation: make the mutationstore the source of truth for mutation entries Summary: Previously the mutation commit extras were the source of truth for mutation information, and the mutation store served as a kind of cache. This turned out to be less useful than expected, as oftentimes commits are missing, and the store is better indexed, so in practice using the store as the source of truth is better. This change makes the mutationstore the (sole) source of truth for mutation data. The extras are kept, but they are now only useful as human-readable debug information, and to ensure the commit hash is unique. Collecting the mutation information during commit creation is now done through a new `mutinfo` object. This is a dict with the same keys as the mutation extras, for simplicity, but it is now passed through the `committablectx` and used to generate the mutation store entry directly. The `mutation.enabled` config option is now used to control all aspects of enabling mutation. The `mutation.record` config option is now only used to indicate whether the mutation extras should also added to the commit. Generally this should be set to `true`, however the option is retained so that mutation extras can be stripped by running `hg amend --config mutation.record=false`, which no longer has the side-effect of not recording mutation information to the store. The "remote commit" mutation record origin is now obsolete, and won't be generated anymore. Pushrebase now relies on the obsmarker information coming back from the server in order to correctly generate mutation information. We will need to change this so that the server returns mutation records before we can fully deprecate obsmarkers. Reviewed By: DurhamG Differential Revision: D19410650 fbshipit-source-id: 8d7094e4bfd8d8e97916898d899a8debd339485f
2020-01-16 12:01:10 +03:00
mutinfo = None
for key in "mutpred", "mutuser", "mutdate", "mutop", "mutsplit":
if key in extra:
del extra[key]
mutation: make the mutationstore the source of truth for mutation entries Summary: Previously the mutation commit extras were the source of truth for mutation information, and the mutation store served as a kind of cache. This turned out to be less useful than expected, as oftentimes commits are missing, and the store is better indexed, so in practice using the store as the source of truth is better. This change makes the mutationstore the (sole) source of truth for mutation data. The extras are kept, but they are now only useful as human-readable debug information, and to ensure the commit hash is unique. Collecting the mutation information during commit creation is now done through a new `mutinfo` object. This is a dict with the same keys as the mutation extras, for simplicity, but it is now passed through the `committablectx` and used to generate the mutation store entry directly. The `mutation.enabled` config option is now used to control all aspects of enabling mutation. The `mutation.record` config option is now only used to indicate whether the mutation extras should also added to the commit. Generally this should be set to `true`, however the option is retained so that mutation extras can be stripped by running `hg amend --config mutation.record=false`, which no longer has the side-effect of not recording mutation information to the store. The "remote commit" mutation record origin is now obsolete, and won't be generated anymore. Pushrebase now relies on the obsmarker information coming back from the server in order to correctly generate mutation information. We will need to change this so that the server returns mutation records before we can fully deprecate obsmarkers. Reviewed By: DurhamG Differential Revision: D19410650 fbshipit-source-id: 8d7094e4bfd8d8e97916898d899a8debd339485f
2020-01-16 12:01:10 +03:00
if enabled(repo):
mutinfo = {}
mutinfo["mutpred"] = ",".join(identfromnode(p) for p in prednodes)
mutinfo["mutuser"] = repo.ui.config("mutation", "user") or repo.ui.username()
date = repo.ui.config("mutation", "date")
if date is None:
date = util.makedate()
else:
date = util.parsedate(date)
mutation: make the mutationstore the source of truth for mutation entries Summary: Previously the mutation commit extras were the source of truth for mutation information, and the mutation store served as a kind of cache. This turned out to be less useful than expected, as oftentimes commits are missing, and the store is better indexed, so in practice using the store as the source of truth is better. This change makes the mutationstore the (sole) source of truth for mutation data. The extras are kept, but they are now only useful as human-readable debug information, and to ensure the commit hash is unique. Collecting the mutation information during commit creation is now done through a new `mutinfo` object. This is a dict with the same keys as the mutation extras, for simplicity, but it is now passed through the `committablectx` and used to generate the mutation store entry directly. The `mutation.enabled` config option is now used to control all aspects of enabling mutation. The `mutation.record` config option is now only used to indicate whether the mutation extras should also added to the commit. Generally this should be set to `true`, however the option is retained so that mutation extras can be stripped by running `hg amend --config mutation.record=false`, which no longer has the side-effect of not recording mutation information to the store. The "remote commit" mutation record origin is now obsolete, and won't be generated anymore. Pushrebase now relies on the obsmarker information coming back from the server in order to correctly generate mutation information. We will need to change this so that the server returns mutation records before we can fully deprecate obsmarkers. Reviewed By: DurhamG Differential Revision: D19410650 fbshipit-source-id: 8d7094e4bfd8d8e97916898d899a8debd339485f
2020-01-16 12:01:10 +03:00
mutinfo["mutdate"] = "%d %d" % date
if op is not None:
mutation: make the mutationstore the source of truth for mutation entries Summary: Previously the mutation commit extras were the source of truth for mutation information, and the mutation store served as a kind of cache. This turned out to be less useful than expected, as oftentimes commits are missing, and the store is better indexed, so in practice using the store as the source of truth is better. This change makes the mutationstore the (sole) source of truth for mutation data. The extras are kept, but they are now only useful as human-readable debug information, and to ensure the commit hash is unique. Collecting the mutation information during commit creation is now done through a new `mutinfo` object. This is a dict with the same keys as the mutation extras, for simplicity, but it is now passed through the `committablectx` and used to generate the mutation store entry directly. The `mutation.enabled` config option is now used to control all aspects of enabling mutation. The `mutation.record` config option is now only used to indicate whether the mutation extras should also added to the commit. Generally this should be set to `true`, however the option is retained so that mutation extras can be stripped by running `hg amend --config mutation.record=false`, which no longer has the side-effect of not recording mutation information to the store. The "remote commit" mutation record origin is now obsolete, and won't be generated anymore. Pushrebase now relies on the obsmarker information coming back from the server in order to correctly generate mutation information. We will need to change this so that the server returns mutation records before we can fully deprecate obsmarkers. Reviewed By: DurhamG Differential Revision: D19410650 fbshipit-source-id: 8d7094e4bfd8d8e97916898d899a8debd339485f
2020-01-16 12:01:10 +03:00
mutinfo["mutop"] = op
if splitting:
mutation: make the mutationstore the source of truth for mutation entries Summary: Previously the mutation commit extras were the source of truth for mutation information, and the mutation store served as a kind of cache. This turned out to be less useful than expected, as oftentimes commits are missing, and the store is better indexed, so in practice using the store as the source of truth is better. This change makes the mutationstore the (sole) source of truth for mutation data. The extras are kept, but they are now only useful as human-readable debug information, and to ensure the commit hash is unique. Collecting the mutation information during commit creation is now done through a new `mutinfo` object. This is a dict with the same keys as the mutation extras, for simplicity, but it is now passed through the `committablectx` and used to generate the mutation store entry directly. The `mutation.enabled` config option is now used to control all aspects of enabling mutation. The `mutation.record` config option is now only used to indicate whether the mutation extras should also added to the commit. Generally this should be set to `true`, however the option is retained so that mutation extras can be stripped by running `hg amend --config mutation.record=false`, which no longer has the side-effect of not recording mutation information to the store. The "remote commit" mutation record origin is now obsolete, and won't be generated anymore. Pushrebase now relies on the obsmarker information coming back from the server in order to correctly generate mutation information. We will need to change this so that the server returns mutation records before we can fully deprecate obsmarkers. Reviewed By: DurhamG Differential Revision: D19410650 fbshipit-source-id: 8d7094e4bfd8d8e97916898d899a8debd339485f
2020-01-16 12:01:10 +03:00
mutinfo["mutsplit"] = ",".join(identfromnode(n) for n in splitting)
if recording(repo):
extra.update(mutinfo)
return mutinfo
def recording(repo):
return repo.ui.configbool("mutation", "record")
def enabled(repo):
return repo.ui.configbool("mutation", "enabled")
def makemutationstore(repo):
return mutationstore.mutationstore(repo.svfs.join("mutation"))
class bundlemutationstore(object):
def __init__(self, bundlerepo):
self._entries = {}
self._splitheads = {}
self._successorssets = {}
self._mutationstore = makemutationstore(bundlerepo)
def addbundleentries(self, entries):
for entry in entries:
self._entries[entry.succ()] = entry
succs = []
for split in entry.split() or ():
self._splitheads[split] = entry.succ()
succs.append(split)
succs.append(entry.succ())
for pred in entry.preds():
self._successorssets.setdefault(pred, []).append(succs)
def get(self, node):
return self._mutationstore.get(node) or self._entries.get(node)
def getsplithead(self, node):
return self._mutationstore.getsplithead(node) or self._splitheads.get(node)
def getsuccessorssets(self, node):
storesets = self._mutationstore.getsuccessorssets(node)
bundlesets = self._successorssets.get(node, [])
if bundlesets and storesets:
return util.removeduplicates(storesets + bundlesets)
else:
return storesets or bundlesets
def has(self, node):
return self._mutationstore.has(node) or node in self._entries
def add(self, entry):
# bundlerepo mutation stores are immutable. This should never be called.
raise NotImplementedError
def flush(self):
pass
def createentry(node, mutinfo):
mutation: make the mutationstore the source of truth for mutation entries Summary: Previously the mutation commit extras were the source of truth for mutation information, and the mutation store served as a kind of cache. This turned out to be less useful than expected, as oftentimes commits are missing, and the store is better indexed, so in practice using the store as the source of truth is better. This change makes the mutationstore the (sole) source of truth for mutation data. The extras are kept, but they are now only useful as human-readable debug information, and to ensure the commit hash is unique. Collecting the mutation information during commit creation is now done through a new `mutinfo` object. This is a dict with the same keys as the mutation extras, for simplicity, but it is now passed through the `committablectx` and used to generate the mutation store entry directly. The `mutation.enabled` config option is now used to control all aspects of enabling mutation. The `mutation.record` config option is now only used to indicate whether the mutation extras should also added to the commit. Generally this should be set to `true`, however the option is retained so that mutation extras can be stripped by running `hg amend --config mutation.record=false`, which no longer has the side-effect of not recording mutation information to the store. The "remote commit" mutation record origin is now obsolete, and won't be generated anymore. Pushrebase now relies on the obsmarker information coming back from the server in order to correctly generate mutation information. We will need to change this so that the server returns mutation records before we can fully deprecate obsmarkers. Reviewed By: DurhamG Differential Revision: D19410650 fbshipit-source-id: 8d7094e4bfd8d8e97916898d899a8debd339485f
2020-01-16 12:01:10 +03:00
def nodesfrominfo(info):
if info is not None:
return [nodefromident(x) for x in info.split(",")]
if mutinfo is not None:
try:
time, tz = mutinfo["mutdate"].split()
time = int(time)
mutation: make the mutationstore the source of truth for mutation entries Summary: Previously the mutation commit extras were the source of truth for mutation information, and the mutation store served as a kind of cache. This turned out to be less useful than expected, as oftentimes commits are missing, and the store is better indexed, so in practice using the store as the source of truth is better. This change makes the mutationstore the (sole) source of truth for mutation data. The extras are kept, but they are now only useful as human-readable debug information, and to ensure the commit hash is unique. Collecting the mutation information during commit creation is now done through a new `mutinfo` object. This is a dict with the same keys as the mutation extras, for simplicity, but it is now passed through the `committablectx` and used to generate the mutation store entry directly. The `mutation.enabled` config option is now used to control all aspects of enabling mutation. The `mutation.record` config option is now only used to indicate whether the mutation extras should also added to the commit. Generally this should be set to `true`, however the option is retained so that mutation extras can be stripped by running `hg amend --config mutation.record=false`, which no longer has the side-effect of not recording mutation information to the store. The "remote commit" mutation record origin is now obsolete, and won't be generated anymore. Pushrebase now relies on the obsmarker information coming back from the server in order to correctly generate mutation information. We will need to change this so that the server returns mutation records before we can fully deprecate obsmarkers. Reviewed By: DurhamG Differential Revision: D19410650 fbshipit-source-id: 8d7094e4bfd8d8e97916898d899a8debd339485f
2020-01-16 12:01:10 +03:00
tz = int(tz)
except (IndexError, ValueError):
time, tz = 0, 0
mutation: make the mutationstore the source of truth for mutation entries Summary: Previously the mutation commit extras were the source of truth for mutation information, and the mutation store served as a kind of cache. This turned out to be less useful than expected, as oftentimes commits are missing, and the store is better indexed, so in practice using the store as the source of truth is better. This change makes the mutationstore the (sole) source of truth for mutation data. The extras are kept, but they are now only useful as human-readable debug information, and to ensure the commit hash is unique. Collecting the mutation information during commit creation is now done through a new `mutinfo` object. This is a dict with the same keys as the mutation extras, for simplicity, but it is now passed through the `committablectx` and used to generate the mutation store entry directly. The `mutation.enabled` config option is now used to control all aspects of enabling mutation. The `mutation.record` config option is now only used to indicate whether the mutation extras should also added to the commit. Generally this should be set to `true`, however the option is retained so that mutation extras can be stripped by running `hg amend --config mutation.record=false`, which no longer has the side-effect of not recording mutation information to the store. The "remote commit" mutation record origin is now obsolete, and won't be generated anymore. Pushrebase now relies on the obsmarker information coming back from the server in order to correctly generate mutation information. We will need to change this so that the server returns mutation records before we can fully deprecate obsmarkers. Reviewed By: DurhamG Differential Revision: D19410650 fbshipit-source-id: 8d7094e4bfd8d8e97916898d899a8debd339485f
2020-01-16 12:01:10 +03:00
return mutationstore.mutationentry(
node,
nodesfrominfo(mutinfo.get("mutpred")),
nodesfrominfo(mutinfo.get("mutsplit")),
mutinfo.get("mutop", ""),
mutinfo.get("mutuser", ""),
time,
tz,
None,
)
def createsyntheticentry(repo, preds, succ, op, splitting=None, user=None, date=None):
user = user or repo.ui.config("mutation", "user") or repo.ui.username()
date = date or repo.ui.config("mutation", "date")
if date is None:
date = util.makedate()
else:
date = util.parsedate(date)
return mutationstore.mutationentry(
succ, preds, splitting, op, user, int(date[0]), int(date[1]), None
)
def recordentries(repo, entries, skipexisting=True):
count = 0
with repo.transaction("record-mutation") as tr:
ms = repo._mutationstore
tr.addfinalize("mutation", lambda _tr: ms.flush())
for entry in entries:
mutation: make the mutationstore the source of truth for mutation entries Summary: Previously the mutation commit extras were the source of truth for mutation information, and the mutation store served as a kind of cache. This turned out to be less useful than expected, as oftentimes commits are missing, and the store is better indexed, so in practice using the store as the source of truth is better. This change makes the mutationstore the (sole) source of truth for mutation data. The extras are kept, but they are now only useful as human-readable debug information, and to ensure the commit hash is unique. Collecting the mutation information during commit creation is now done through a new `mutinfo` object. This is a dict with the same keys as the mutation extras, for simplicity, but it is now passed through the `committablectx` and used to generate the mutation store entry directly. The `mutation.enabled` config option is now used to control all aspects of enabling mutation. The `mutation.record` config option is now only used to indicate whether the mutation extras should also added to the commit. Generally this should be set to `true`, however the option is retained so that mutation extras can be stripped by running `hg amend --config mutation.record=false`, which no longer has the side-effect of not recording mutation information to the store. The "remote commit" mutation record origin is now obsolete, and won't be generated anymore. Pushrebase now relies on the obsmarker information coming back from the server in order to correctly generate mutation information. We will need to change this so that the server returns mutation records before we can fully deprecate obsmarkers. Reviewed By: DurhamG Differential Revision: D19410650 fbshipit-source-id: 8d7094e4bfd8d8e97916898d899a8debd339485f
2020-01-16 12:01:10 +03:00
if skipexisting and ms.has(entry.succ()):
continue
ms.add(entry)
count += 1
return count
def getdag(repo, *nodes):
"""Get 1:1 mutation subgraph for selected nodes"""
return repo._mutationstore.getdag(nodes)
mutation: make the mutationstore the source of truth for mutation entries Summary: Previously the mutation commit extras were the source of truth for mutation information, and the mutation store served as a kind of cache. This turned out to be less useful than expected, as oftentimes commits are missing, and the store is better indexed, so in practice using the store as the source of truth is better. This change makes the mutationstore the (sole) source of truth for mutation data. The extras are kept, but they are now only useful as human-readable debug information, and to ensure the commit hash is unique. Collecting the mutation information during commit creation is now done through a new `mutinfo` object. This is a dict with the same keys as the mutation extras, for simplicity, but it is now passed through the `committablectx` and used to generate the mutation store entry directly. The `mutation.enabled` config option is now used to control all aspects of enabling mutation. The `mutation.record` config option is now only used to indicate whether the mutation extras should also added to the commit. Generally this should be set to `true`, however the option is retained so that mutation extras can be stripped by running `hg amend --config mutation.record=false`, which no longer has the side-effect of not recording mutation information to the store. The "remote commit" mutation record origin is now obsolete, and won't be generated anymore. Pushrebase now relies on the obsmarker information coming back from the server in order to correctly generate mutation information. We will need to change this so that the server returns mutation records before we can fully deprecate obsmarkers. Reviewed By: DurhamG Differential Revision: D19410650 fbshipit-source-id: 8d7094e4bfd8d8e97916898d899a8debd339485f
2020-01-16 12:01:10 +03:00
def lookup(repo, node):
"""Look up mutation information for the given node"""
return repo._mutationstore.get(node)
def lookupsplit(repo, node):
"""Look up mutation information for the given node, or the main split node
if this node is the result of a split.
"""
ms = repo._mutationstore
mainnode = ms.getsplithead(node) or node
return ms.get(mainnode)
def lookupsuccessors(repo, node):
"""Look up the immediate successors sets for the given node"""
succsets = sorted(repo._mutationstore.getsuccessorssets(node))
return util.removesortedduplicates(succsets)
def allpredecessors(repo, nodes, startdepth=None, stopdepth=None):
"""Yields all the nodes that are predecessors of the given nodes.
Some predecessors may not be known locally."""
depth = 0
thislevel = set(nodes)
nextlevel = set()
seen = {nullid}
ispublic = getispublicfunc(repo)
while thislevel and (stopdepth is None or depth < stopdepth):
for current in thislevel:
if current in seen:
continue
seen.add(current)
if startdepth is None or depth >= startdepth:
yield current
pred = None
entry = lookupsplit(repo, current)
if entry is not None:
pred = entry.preds()
if pred is not None:
for nextnode in pred:
if nextnode not in seen and not ispublic(nextnode):
nextlevel.add(nextnode)
depth += 1
thislevel = nextlevel
nextlevel = set()
def allsuccessors(repo, nodes, startdepth=None, stopdepth=None):
"""Yields all the nodes that are successors of the given nodes.
Successors that are not known locally may be omitted."""
depth = 0
thislevel = set(nodes)
nextlevel = set()
seen = set()
ispublic = getispublicfunc(repo)
while thislevel and (stopdepth is None or depth < stopdepth):
for current in thislevel:
if current in seen:
continue
seen.add(current)
if startdepth is None or depth >= startdepth:
yield current
if ispublic(current):
continue
succsets = lookupsuccessors(repo, current)
if succsets:
nextlevel = nextlevel.union(*succsets)
depth += 1
thislevel = nextlevel
nextlevel = set()
class obsoletecache(object):
def __init__(self):
# Set of commits that are known to be obsolete for each filter level.
self.obsolete = defaultdict(set)
# Set of commits that are known to be not obsolete for each filter level.
self.notobsolete = defaultdict(set)
# If true, then the full set of obsolete commits is known for this
# filter level, and is stored in ``self.obsolete``.
self.complete = defaultdict(bool)
def isobsolete(self, repo, node):
"""Returns true if the node is obsolete in the repository."""
if node is None:
return False
if node not in repo:
return False
ispublic = getispublicfunc(repo)
if ispublic(node):
return False
obsolete = self.obsolete[repo.filtername]
if node in obsolete:
return True
if self.complete[repo.filtername] or node in self.notobsolete[repo.filtername]:
return False
unfi = repo.unfiltered()
clhasnode = getisvisiblefunc(repo)
clrev = unfi.changelog.rev
hiddenrevs = repoview.filterrevs(repo, "visible")
for succ in allsuccessors(repo, [node], startdepth=1):
# If any successor is already known to be obsolete, we can
# assume that the current node is obsolete without checking further.
if succ in obsolete:
obsolete.add(node)
return True
# The node is obsolete if any successor is visible in the normal
# filtered repo.
if clhasnode(succ) and clrev(succ) not in hiddenrevs:
obsolete.add(node)
return True
self.notobsolete[repo.filtername].add(node)
return False
def obsoletenodes(self, repo):
if self.complete[repo.filtername]:
return self.obsolete[repo.filtername]
with perftrace.trace("Compute Obsolete Nodes"):
perftrace.traceflag("mutation")
# Testing each node separately will result in lots of repeated tests.
# Instead, we can do the following:
# - Compute all nodes that are obsolete because one of their closest
# successors is visible.
# - Work back from these commits marking all of their predecessors as
# obsolete.
# Note that "visible" here means "visible in a normal filtered repo",
# even if the filter for this repo includes other commits.
clhasnode = getisvisiblefunc(repo)
clrev = repo.changelog.rev
obsolete = self.obsolete[repo.filtername]
hiddenrevs = repoview.filterrevs(repo, "visible")
for node in repo.nodes("not public()"):
succsets = successorssets(repo, node, closest=True)
if succsets != [[node]]:
if any(
clrev(succ) not in hiddenrevs
for succset in succsets
for succ in succset
):
obsolete.add(node)
candidates = set(obsolete)
seen = set(obsolete)
while candidates:
candidate = candidates.pop()
entry = lookupsplit(repo, candidate)
if entry:
for pred in entry.preds():
if pred not in obsolete and pred not in seen:
candidates.add(pred)
seen.add(pred)
if clhasnode(pred) and pred != nullid:
obsolete.add(pred)
self.obsolete[repo.filtername] = frozenset(obsolete)
self.complete[repo.filtername] = True
# Since we know all obsolete commits, no need to remember which ones
# are not obsolete.
if repo.filtername in self.notobsolete:
del self.notobsolete[repo.filtername]
return self.obsolete[repo.filtername]
def isobsolete(repo, node):
if not util.safehasattr(repo, "_mutationobsolete"):
repo._mutationobsolete = obsoletecache()
return repo._mutationobsolete.isobsolete(repo, node)
def obsoletenodes(repo):
if not util.safehasattr(repo, "_mutationobsolete"):
repo._mutationobsolete = obsoletecache()
return repo._mutationobsolete.obsoletenodes(repo)
def clearobsoletecache(repo):
if util.safehasattr(repo, "_mutationobsolete"):
del repo._mutationobsolete
def fate(repo, node):
"""Returns the fate of a node.
This returns a list of ([nodes], operation) pairs, indicating mutations that
happened to this node that resulted in one or more visible commits.
"""
ispublic = getispublicfunc(repo)
fate = []
if isobsolete(repo, node):
for succset in successorssets(repo, node, closest=True):
if succset == [node]:
pass
elif len(succset) > 1:
fate.append((succset, "split"))
else:
succ = succset[0]
# Base the default operation name on the successor's phase
if ispublic(succ):
op = "land"
else:
op = "rewrite"
# Try to find the real operation name.
entry = lookup(repo, succ)
if entry is not None:
preds = entry.preds()
if preds is not None and node in preds:
op = entry.op() or op
fate.append((succset, op))
return fate
def predecessorsset(repo, startnode, closest=False):
"""Return a list of the commits that were replaced by the startnode.
If there are no such commits, returns a list containing the startnode.
If ``closest`` is True, returns a list of the visible commits that are the
closest previous version of the start node.
If ``closest`` is False, returns a list of the earliest original versions of
the start node.
"""
seen = {startnode}
ispublic = getispublicfunc(repo)
def get(node):
"""Get immediate predecessors
Returns a list of immediate predecessors of the node, omitting
any predecessors which have already been seen, to prevent loops.
If the node has no predecessors, returns a list containing the node
itself.
"""
entry = lookupsplit(repo, node)
if entry is not None:
preds = entry.preds()
if preds is not None:
return [
pred for pred in preds if pred not in seen and not ispublic(pred)
] or [node]
return [node]
preds = [startnode]
nextpreds = sum((get(p) for p in preds), [])
expanded = nextpreds != preds
while expanded:
seen.update(nextpreds)
if all(p in repo for p in nextpreds):
# We have found a set of predecessors that are all visible - this is
# a valid set to return.
preds = nextpreds
if closest:
break
# Now look at the next predecessors of each commit.
newnextpreds = sum((get(p) for p in nextpreds), [])
else:
# Expand out to the predecessors of the commits until we find visible
# ones.
newnextpreds = sum(([p] if p in repo else get(p) for p in nextpreds), [])
expanded = newnextpreds != nextpreds
nextpreds = newnextpreds
if not expanded:
# We've reached a stable state and some of the commits might not be
# visible. Remove the invisible commits, and continue with what's
# left.
newnextpreds = [p for p in nextpreds if p in repo]
if newnextpreds:
expanded = newnextpreds != nextpreds
nextpreds = newnextpreds
# Remove duplicates from the list of predecessors. Splits and folds
# can lead to the same commits via multiple routes, but we don't need
# to process them multiple times.
nextpreds = util.removeduplicates(nextpreds)
return preds
def _succproduct(succsetlist):
"""Takes a list of successorsset lists and returns a single successorsset
list representing the cartesian product of those successorsset lists.
The ordering of elements within the lists must be preserved.
>>> _succproduct([[[1]], [[2]]])
[[1, 2]]
>>> _succproduct([[[1, 2]], [[3, 4]]])
[[1, 2, 3, 4]]
>>> _succproduct([[[1, 2], [3, 4]], [[5, 6]]])
[[1, 2, 5, 6], [3, 4, 5, 6]]
>>> _succproduct([[[1, 2], [3, 4]], [[5, 6], [7, 8]]])
[[1, 2, 5, 6], [3, 4, 5, 6], [1, 2, 7, 8], [3, 4, 7, 8]]
>>> _succproduct([[[1, 2], [3, 4]], [[2, 3], [7, 8]]])
[[1, 2, 3], [3, 4, 2], [1, 2, 7, 8], [3, 4, 7, 8]]
>>> _succproduct([[[1, 2], [3, 4]], [[1, 2], [7, 8]]])
[[1, 2], [3, 4, 1, 2], [1, 2, 7, 8], [3, 4, 7, 8]]
>>> _succproduct([[[1], [2]], [[3], [4]]])
[[1, 3], [2, 3], [1, 4], [2, 4]]
>>> _succproduct([[[5]], [[4]], [[3]], [[2]], [[1]]])
[[5, 4, 3, 2, 1]]
"""
# Start with the first successorsset.
product = succsetlist[0]
for succsets in succsetlist[1:]:
# For each of the remaining successorssets, compute the product with
# the successorsset so far.
newproduct = []
for succset in succsets:
for p in product:
newproduct.append(p + [s for s in succset if s not in p])
product = newproduct
return product
def successorssets(repo, startnode, closest=False, cache=None):
"""Return a list of lists of commits that replace the startnode.
If there are no such commits, returns a list containing a single list
containing the startnode.
If ``closest`` is True, the lists contain the visible commits that are the
closest next version of the start node.
If ``closest`` is False, the lists contain the latest versions of the start
node.
The ``cache`` parameter is unused. It is provided to make this function
signature-compatible with ``obsutil.successorssets``.
"""
seen = {startnode}
ispublic = getispublicfunc(repo)
def getsets(node):
"""Get immediate successors sets
Returns a list of immediate successors sets of the node, omitting
any sets which contain nodes already seen, to prevent loops.
If the node has no successors, returns a list containing a single
successors set which contains the node itself.
"""
if ispublic(node):
return [[node]]
succsets = [
succset
for succset in lookupsuccessors(repo, node)
if not any(succ in seen for succ in succset)
]
return succsets or [[node]]
clhasnode = getisvisiblefunc(repo)
succsets = [[startnode]]
nextsuccsets = getsets(startnode)
expanded = nextsuccsets != succsets
while expanded:
seen.update(node for succset in nextsuccsets for node in succset)
if all(clhasnode(s) for succset in nextsuccsets for s in succset):
# We have found a set of successor sets that all contain visible
# commits - this is a valid set to return.
succsets = nextsuccsets
if closest:
break
# Now look at the next successors of each successors set. When
# commits are modified in different ways (i.e. they have been
# diverged), we need to find all possible permutations that replace
# the original nodes. To do this, we compute the cartesian product
# of the successors sets of each successor in the original
# successors set.
#
# For example, if A is split into B and C, B is diverged to B1 and
# B2, and C is also diverged to C1 and C2, then the successors sets
# of A are: [B1, C1], [B1, C2], [B2, C1], [B2, C2], which is the
# cartesian product: [B1, B2] x [C1, C2].
newnextsuccsets = sum(
[
_succproduct([getsets(succ) for succ in succset])
for succset in nextsuccsets
],
[],
)
else:
# Expand each successors set out to its successors until we find
# visible commit. Again, use the cartesian product to find all
# permutations.
newnextsuccsets = sum(
[
_succproduct(
[
[[succ]] if clhasnode(succ) else getsets(succ)
for succ in succset
]
)
for succset in nextsuccsets
],
[],
)
expanded = newnextsuccsets != nextsuccsets
nextsuccsets = newnextsuccsets
if not expanded:
# We've reached a stable state and some of the commits might not be
# visible. Remove the invisible commits, and continue with what's
# left.
newnextsuccsets = [
[s for s in succset if clhasnode(s)] for succset in nextsuccsets
]
# Remove sets that are now empty.
newnextsuccsets = [succset for succset in newnextsuccsets if succset]
if newnextsuccsets:
expanded = newnextsuccsets != nextsuccsets
nextsuccsets = newnextsuccsets
# Remove duplicates from the list of successors sets. Splits and folds
# can lead to the same commits via multiple routes, but we don't need
# to process them multiple times.
nextsuccsets = util.removeduplicates(nextsuccsets, key=frozenset)
return succsets
def foreground(repo, nodes):
"""Returns all nodes in the "foreground" of the given nodes.
The foreground of a commit is the transitive closure of all descendants
and successors of the commit.
"""
unfi = repo.unfiltered()
nm = unfi.changelog.nodemap
foreground = set(nodes)
newnodes = set(nodes)
while newnodes:
newnodes.update(n for n in allsuccessors(repo, newnodes) if n in nm)
newnodes.update(unfi.nodes("%ln::", newnodes))
newnodes.difference_update(foreground)
foreground.update(newnodes)
return foreground
def toposortrevs(repo, revs, predmap):
"""topologically sort revs according to the given predecessor map"""
dag = {}
valid = set(revs)
heads = set(revs)
clparentrevs = repo.changelog.parentrevs
for rev in revs:
prev = [p for p in clparentrevs(rev) if p in valid]
prev.extend(predmap[rev])
heads.difference_update(prev)
dag[rev] = prev
if not heads:
raise error.Abort("commit predecessors and ancestors contain a cycle")
seen = set()
sortedrevs = []
revstack = list(reversed(sorted(heads)))
while revstack:
rev = revstack[-1]
if rev not in seen:
seen.add(rev)
for next in reversed(dag[rev]):
if next not in seen:
revstack.append(next)
else:
sortedrevs.append(rev)
revstack.pop()
return sortedrevs
def toposort(repo, items, nodefn=None):
"""topologically sort nodes according to the given predecessor map
items can either be nodes, or something convertible to nodes by a provided
node function.
"""
if nodefn is None:
nodefn = lambda item: item
clrev = repo.changelog.rev
revmap = {clrev(nodefn(x)): i for i, x in enumerate(items)}
predmap = {}
for item in items:
node = nodefn(item)
rev = clrev(node)
predmap[rev] = [
r
for r in map(clrev, predecessorsset(repo, node, closest=True))
if r != rev and r in revmap
]
sortedrevs = toposortrevs(repo, revmap.keys(), predmap)
return [items[revmap[r]] for r in sortedrevs]
def unbundle(repo, bundledata):
mutation: make the mutationstore the source of truth for mutation entries Summary: Previously the mutation commit extras were the source of truth for mutation information, and the mutation store served as a kind of cache. This turned out to be less useful than expected, as oftentimes commits are missing, and the store is better indexed, so in practice using the store as the source of truth is better. This change makes the mutationstore the (sole) source of truth for mutation data. The extras are kept, but they are now only useful as human-readable debug information, and to ensure the commit hash is unique. Collecting the mutation information during commit creation is now done through a new `mutinfo` object. This is a dict with the same keys as the mutation extras, for simplicity, but it is now passed through the `committablectx` and used to generate the mutation store entry directly. The `mutation.enabled` config option is now used to control all aspects of enabling mutation. The `mutation.record` config option is now only used to indicate whether the mutation extras should also added to the commit. Generally this should be set to `true`, however the option is retained so that mutation extras can be stripped by running `hg amend --config mutation.record=false`, which no longer has the side-effect of not recording mutation information to the store. The "remote commit" mutation record origin is now obsolete, and won't be generated anymore. Pushrebase now relies on the obsmarker information coming back from the server in order to correctly generate mutation information. We will need to change this so that the server returns mutation records before we can fully deprecate obsmarkers. Reviewed By: DurhamG Differential Revision: D19410650 fbshipit-source-id: 8d7094e4bfd8d8e97916898d899a8debd339485f
2020-01-16 12:01:10 +03:00
if enabled(repo):
entries = mutationstore.unbundle(bundledata)
recordentries(repo, entries, skipexisting=True)
def bundle(repo, nodes):
"""Generate bundled mutation data for bundling alongside the given nodes.
This consists of mutation entries for all predecessors of the given nodes,
including the nodes themselves.
"""
nodes = set(nodes)
remaining = set(nodes)
seen = set()
entries = []
while remaining:
current = remaining.pop()
if current in seen:
continue
seen.add(current)
entry = lookupsplit(repo, current)
if entry is not None:
mutation: make the mutationstore the source of truth for mutation entries Summary: Previously the mutation commit extras were the source of truth for mutation information, and the mutation store served as a kind of cache. This turned out to be less useful than expected, as oftentimes commits are missing, and the store is better indexed, so in practice using the store as the source of truth is better. This change makes the mutationstore the (sole) source of truth for mutation data. The extras are kept, but they are now only useful as human-readable debug information, and to ensure the commit hash is unique. Collecting the mutation information during commit creation is now done through a new `mutinfo` object. This is a dict with the same keys as the mutation extras, for simplicity, but it is now passed through the `committablectx` and used to generate the mutation store entry directly. The `mutation.enabled` config option is now used to control all aspects of enabling mutation. The `mutation.record` config option is now only used to indicate whether the mutation extras should also added to the commit. Generally this should be set to `true`, however the option is retained so that mutation extras can be stripped by running `hg amend --config mutation.record=false`, which no longer has the side-effect of not recording mutation information to the store. The "remote commit" mutation record origin is now obsolete, and won't be generated anymore. Pushrebase now relies on the obsmarker information coming back from the server in order to correctly generate mutation information. We will need to change this so that the server returns mutation records before we can fully deprecate obsmarkers. Reviewed By: DurhamG Differential Revision: D19410650 fbshipit-source-id: 8d7094e4bfd8d8e97916898d899a8debd339485f
2020-01-16 12:01:10 +03:00
entries.append(entry)
for nextnode in entry.preds():
if nextnode not in seen:
remaining.add(nextnode)
return mutationstore.bundle(entries)
def convertfromobsmarkers(repo):
"""Convert obsmarkers into mutation records."""
obsmarkers = repo.obsstore._all
# Sort obsmarkers by date. Applying them in probable date order gives us
# a better chance of resolving cycles in the right way.
obsmarkers.sort(key=lambda x: x[4])
newmut = {}
def checkloopfree(pred, succs):
candidates = {pred}
while candidates:
candidate = candidates.pop()
if candidate in newmut:
mutpreds, mutsplit, _markers = newmut[candidate]
for succ in succs:
if succ in mutpreds:
repo.ui.debug(
"ignoring loop: %s -> %s\n history loops at %s -> %s\n"
% (
nodemod.hex(pred),
", ".join([nodemod.hex(s) for s in succs]),
nodemod.hex(succ),
nodemod.hex(candidate),
)
)
return False
candidates.update(mutpreds)
return True
def checkduplicate(marker, markers):
return any(marker[:2] == m[:2] for m in markers)
dropprune = 0
droprevive = 0
dropundo = 0
droploop = 0
dropinvalid = 0
for obsmarker in obsmarkers:
obspred, obssuccs, obsflag, obsmeta, obsdate, obsparents = obsmarker
if not obssuccs:
# Skip prune markers
dropprune += 1
continue
if obssuccs == (obspred,):
# Skip revive markers
droprevive += 1
continue
obsmeta = dict(obsmeta)
if obsmeta.get("operation") in ("undo", "uncommit", "unamend"):
# Skip undo-style markers
dropundo += 1
continue
if not checkloopfree(obspred, obssuccs):
# Skip markers that introduce loops
droploop += 1
continue
if len(obssuccs) > 1:
# Split marker
succ = obssuccs[-1]
if succ in newmut:
preds, split, markers = newmut[succ]
if checkduplicate(obsmarker, markers):
# duplicate
continue
repo.ui.debug(
"invalid obsmarker found: %s -> %s is both split and folded\n"
% (nodemod.hex(obspred), nodemod.hex(succ))
)
dropinvalid += 1
continue
newmut[succ] = ([obspred], obssuccs[:-1], [obsmarker])
elif obssuccs[0] in newmut:
preds, split, markers = newmut[obssuccs[0]]
if checkduplicate(obsmarker, markers):
# duplicate
continue
# Fold marker
preds.append(obspred)
markers.append(obsmarker)
else:
# Normal marker
newmut[obssuccs[0]] = ([obspred], None, [obsmarker])
repo.ui.debug(
"processed %s markers for %s commits\n" % (len(obsmarkers), len(newmut))
)
repo.ui.debug(
"dropped markers: prune: %s, revive: %s, undo: %s, loop: %s, invalid: %s\n"
% (dropprune, droprevive, dropundo, droploop, dropinvalid)
)
entries = []
for succ, (preds, split, obsmarkers) in newmut.items():
if lookup(repo, succ) is not None:
# Have already converted this successor, or already know about it
continue
mutop = ""
mutuser = ""
mutdate = None
for obsmarker in obsmarkers:
obspred, obssuccs, obsflag, obsmeta, obsdate, obsparents = obsmarker
obsmeta = dict(obsmeta)
obsop = obsmeta.get("operation", "")
if not mutop or obsop not in ("", "copy"):
mutop = obsop
obsuser = obsmeta.get("user", "")
if not mutuser and obsuser:
mutuser = obsuser
if not mutdate:
mutdate = obsdate
entries.append(
createsyntheticentry(repo, preds, succ, mutop, split, mutuser, mutdate)
)
with repo.lock():
count = recordentries(repo, entries, skipexisting=False)
return (len(entries), len(newmut), count)
def automigrate(repo):
if (
enabled(repo)
and repo.ui.configbool("mutation", "automigrate")
and not repo.ui.configbool("mutation", "proxy-obsstore")
):
msg = _(
"checking for new obsmarkers to migrate to the mutation store (https://fburl.com/hgmutation)\n"
)
repo.ui.status(msg)
convertfromobsmarkers(repo)
def getisvisiblefunc(repo):
"""get a (node) -> bool function to test visibility
If narrow-heads is set, visible commits are defined as "::head()".
visibility here is used to test where a "successor" exists or not.
"""
if repo.ui.configbool("experimental", "narrow-heads"):
# Use phase to test.
# secret phase means "not reachable from public or draft heads", aka. "hidden"
nodemap = repo.changelog.nodemap
getphase = repo._phasecache.phase
def isvisible(
node, nodemap=nodemap, getphase=getphase, repo=repo, secret=phases.secret
):
try:
rev = nodemap[node]
except error.RevlogError:
return False
return getphase(repo, rev) != secret
return isvisible
else:
# Use cl.hasnode to test
return repo.changelog.hasnode
def getispublicfunc(repo):
"""get a (node) -> bool function to test whether a commit is public"""
nodemap = repo.changelog.nodemap
getphase = repo._phasecache.phase
def ispublic(
node, nodemap=nodemap, getphase=getphase, repo=repo, public=phases.public
):
try:
rev = nodemap[node]
except error.RevlogError:
return False
return getphase(repo, rev) == public
return ispublic