merge with stable

mercurial/bundle2.py

@@ -1,1731 +1,1732 b''
 # bundle2.py - generic container format to transmit arbitrary data.
 #
 # Copyright 2013 Facebook, Inc.
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 """Handling of the new bundle2 format
 
 The goal of bundle2 is to act as an atomically packet to transmit a set of
 payloads in an application agnostic way. It consist in a sequence of "parts"
 that will be handed to and processed by the application layer.
 
 
 General format architecture
 ===========================
 
 The format is architectured as follow
 
  - magic string
  - stream level parameters
  - payload parts (any number)
  - end of stream marker.
 
 the Binary format
 ============================
 
 All numbers are unsigned and big-endian.
 
 stream level parameters
 ------------------------
 
 Binary format is as follow
 
 :params size: int32
 
   The total number of Bytes used by the parameters
 
 :params value: arbitrary number of Bytes
 
   A blob of `params size` containing the serialized version of all stream level
   parameters.
 
   The blob contains a space separated list of parameters. Parameters with value
   are stored in the form `<name>=<value>`. Both name and value are urlquoted.
 
   Empty name are obviously forbidden.
 
   Name MUST start with a letter. If this first letter is lower case, the
   parameter is advisory and can be safely ignored. However when the first
   letter is capital, the parameter is mandatory and the bundling process MUST
   stop if he is not able to proceed it.
 
   Stream parameters use a simple textual format for two main reasons:
 
   - Stream level parameters should remain simple and we want to discourage any
     crazy usage.
   - Textual data allow easy human inspection of a bundle2 header in case of
     troubles.
 
   Any Applicative level options MUST go into a bundle2 part instead.
 
 Payload part
 ------------------------
 
 Binary format is as follow
 
 :header size: int32
 
   The total number of Bytes used by the part header. When the header is empty
   (size = 0) this is interpreted as the end of stream marker.
 
 :header:
 
     The header defines how to interpret the part. It contains two piece of
     data: the part type, and the part parameters.
 
     The part type is used to route an application level handler, that can
     interpret payload.
 
     Part parameters are passed to the application level handler.  They are
     meant to convey information that will help the application level object to
     interpret the part payload.
 
     The binary format of the header is has follow
 
     :typesize: (one byte)
 
     :parttype: alphanumerical part name (restricted to [a-zA-Z0-9_:-]*)
 
     :partid: A 32bits integer (unique in the bundle) that can be used to refer
              to this part.
 
     :parameters:
 
         Part's parameter may have arbitrary content, the binary structure is::
 
             <mandatory-count><advisory-count><param-sizes><param-data>
 
         :mandatory-count: 1 byte, number of mandatory parameters
 
         :advisory-count:  1 byte, number of advisory parameters
 
         :param-sizes:
 
             N couple of bytes, where N is the total number of parameters. Each
             couple contains (<size-of-key>, <size-of-value) for one parameter.
 
         :param-data:
 
             A blob of bytes from which each parameter key and value can be
             retrieved using the list of size couples stored in the previous
             field.
 
             Mandatory parameters comes first, then the advisory ones.
 
             Each parameter's key MUST be unique within the part.
 
 :payload:
 
     payload is a series of `<chunksize><chunkdata>`.
 
     `chunksize` is an int32, `chunkdata` are plain bytes (as much as
     `chunksize` says)` The payload part is concluded by a zero size chunk.
 
     The current implementation always produces either zero or one chunk.
     This is an implementation limitation that will ultimately be lifted.
 
     `chunksize` can be negative to trigger special case processing. No such
     processing is in place yet.
 
 Bundle processing
 ============================
 
 Each part is processed in order using a "part handler". Handler are registered
 for a certain part type.
 
 The matching of a part to its handler is case insensitive. The case of the
 part type is used to know if a part is mandatory or advisory. If the Part type
 contains any uppercase char it is considered mandatory. When no handler is
 known for a Mandatory part, the process is aborted and an exception is raised.
 If the part is advisory and no handler is known, the part is ignored. When the
 process is aborted, the full bundle is still read from the stream to keep the
 channel usable. But none of the part read from an abort are processed. In the
 future, dropping the stream may become an option for channel we do not care to
 preserve.
 """
 
 from __future__ import absolute_import
 
 import errno
 import re
 import string
 import struct
 import sys
 
 from .i18n import _
 from . import (
     changegroup,
     error,
     obsolete,
     pushkey,
     pycompat,
     tags,
     url,
     util,
 )
 
 urlerr = util.urlerr
 urlreq = util.urlreq
 
 _pack = struct.pack
 _unpack = struct.unpack
 
 _fstreamparamsize = '>i'
 _fpartheadersize = '>i'
 _fparttypesize = '>B'
 _fpartid = '>I'
 _fpayloadsize = '>i'
 _fpartparamcount = '>BB'
 
 preferedchunksize = 4096
 
 _parttypeforbidden = re.compile('[^a-zA-Z0-9_:-]')
 
 def outdebug(ui, message):
     """debug regarding output stream (bundling)"""
     if ui.configbool('devel', 'bundle2.debug', False):
         ui.debug('bundle2-output: %s\n' % message)
 
 def indebug(ui, message):
     """debug on input stream (unbundling)"""
     if ui.configbool('devel', 'bundle2.debug', False):
         ui.debug('bundle2-input: %s\n' % message)
 
 def validateparttype(parttype):
     """raise ValueError if a parttype contains invalid character"""
     if _parttypeforbidden.search(parttype):
         raise ValueError(parttype)
 
 def _makefpartparamsizes(nbparams):
     """return a struct format to read part parameter sizes
 
     The number parameters is variable so we need to build that format
     dynamically.
     """
     return '>'+('BB'*nbparams)
 
 parthandlermapping = {}
 
 def parthandler(parttype, params=()):
     """decorator that register a function as a bundle2 part handler
 
     eg::
 
         @parthandler('myparttype', ('mandatory', 'param', 'handled'))
         def myparttypehandler(...):
             '''process a part of type "my part".'''
             ...
     """
     validateparttype(parttype)
     def _decorator(func):
         lparttype = parttype.lower() # enforce lower case matching.
         assert lparttype not in parthandlermapping
         parthandlermapping[lparttype] = func
         func.params = frozenset(params)
         return func
     return _decorator
 
 class unbundlerecords(object):
     """keep record of what happens during and unbundle
 
     New records are added using `records.add('cat', obj)`. Where 'cat' is a
     category of record and obj is an arbitrary object.
 
     `records['cat']` will return all entries of this category 'cat'.
 
     Iterating on the object itself will yield `('category', obj)` tuples
     for all entries.
 
     All iterations happens in chronological order.
     """
 
     def __init__(self):
         self._categories = {}
         self._sequences = []
         self._replies = {}
 
     def add(self, category, entry, inreplyto=None):
         """add a new record of a given category.
 
         The entry can then be retrieved in the list returned by
         self['category']."""
         self._categories.setdefault(category, []).append(entry)
         self._sequences.append((category, entry))
         if inreplyto is not None:
             self.getreplies(inreplyto).add(category, entry)
 
     def getreplies(self, partid):
         """get the records that are replies to a specific part"""
         return self._replies.setdefault(partid, unbundlerecords())
 
     def __getitem__(self, cat):
         return tuple(self._categories.get(cat, ()))
 
     def __iter__(self):
         return iter(self._sequences)
 
     def __len__(self):
         return len(self._sequences)
 
     def __nonzero__(self):
         return bool(self._sequences)
 
     __bool__ = __nonzero__
 
 class bundleoperation(object):
     """an object that represents a single bundling process
 
     Its purpose is to carry unbundle-related objects and states.
 
     A new object should be created at the beginning of each bundle processing.
     The object is to be returned by the processing function.
 
     The object has very little content now it will ultimately contain:
     * an access to the repo the bundle is applied to,
     * a ui object,
     * a way to retrieve a transaction to add changes to the repo,
     * a way to record the result of processing each part,
     * a way to construct a bundle response when applicable.
     """
 
     def __init__(self, repo, transactiongetter, captureoutput=True):
         self.repo = repo
         self.ui = repo.ui
         self.records = unbundlerecords()
         self.gettransaction = transactiongetter
         self.reply = None
         self.captureoutput = captureoutput
 
 class TransactionUnavailable(RuntimeError):
     pass
 
 def _notransaction():
     """default method to get a transaction while processing a bundle
 
     Raise an exception to highlight the fact that no transaction was expected
     to be created"""
     raise TransactionUnavailable()
 
 def applybundle(repo, unbundler, tr, source=None, url=None, op=None):
     # transform me into unbundler.apply() as soon as the freeze is lifted
     tr.hookargs['bundle2'] = '1'
     if source is not None and 'source' not in tr.hookargs:
         tr.hookargs['source'] = source
     if url is not None and 'url' not in tr.hookargs:
         tr.hookargs['url'] = url
     return processbundle(repo, unbundler, lambda: tr, op=op)
 
 def processbundle(repo, unbundler, transactiongetter=None, op=None):
     """This function process a bundle, apply effect to/from a repo
 
     It iterates over each part then searches for and uses the proper handling
     code to process the part. Parts are processed in order.
 
     Unknown Mandatory part will abort the process.
 
     It is temporarily possible to provide a prebuilt bundleoperation to the
     function. This is used to ensure output is properly propagated in case of
     an error during the unbundling. This output capturing part will likely be
     reworked and this ability will probably go away in the process.
     """
     if op is None:
         if transactiongetter is None:
             transactiongetter = _notransaction
         op = bundleoperation(repo, transactiongetter)
     # todo:
     # - replace this is a init function soon.
     # - exception catching
     unbundler.params
     if repo.ui.debugflag:
         msg = ['bundle2-input-bundle:']
         if unbundler.params:
             msg.append(' %i params')
         if op.gettransaction is None:
             msg.append(' no-transaction')
         else:
             msg.append(' with-transaction')
         msg.append('\n')
         repo.ui.debug(''.join(msg))
     iterparts = enumerate(unbundler.iterparts())
     part = None
     nbpart = 0
     try:
         for nbpart, part in iterparts:
             _processpart(op, part)
     except Exception as exc:
         # Any exceptions seeking to the end of the bundle at this point are
         # almost certainly related to the underlying stream being bad.
         # And, chances are that the exception we're handling is related to
         # getting in that bad state. So, we swallow the seeking error and
         # re-raise the original error.
         seekerror = False
         try:
             for nbpart, part in iterparts:
                 # consume the bundle content
                 part.seek(0, 2)
         except Exception:
             seekerror = True
 
         # Small hack to let caller code distinguish exceptions from bundle2
         # processing from processing the old format. This is mostly
         # needed to handle different return codes to unbundle according to the
         # type of bundle. We should probably clean up or drop this return code
         # craziness in a future version.
         exc.duringunbundle2 = True
         salvaged = []
         replycaps = None
         if op.reply is not None:
             salvaged = op.reply.salvageoutput()
             replycaps = op.reply.capabilities
         exc._replycaps = replycaps
         exc._bundle2salvagedoutput = salvaged
 
         # Re-raising from a variable loses the original stack. So only use
         # that form if we need to.
         if seekerror:
             raise exc
         else:
             raise
     finally:
         repo.ui.debug('bundle2-input-bundle: %i parts total\n' % nbpart)
 
     return op
 
 def _processpart(op, part):
     """process a single part from a bundle
 
     The part is guaranteed to have been fully consumed when the function exits
     (even if an exception is raised)."""
     status = 'unknown' # used by debug output
     hardabort = False
     try:
         try:
             handler = parthandlermapping.get(part.type)
             if handler is None:
                 status = 'unsupported-type'
                 raise error.BundleUnknownFeatureError(parttype=part.type)
             indebug(op.ui, 'found a handler for part %r' % part.type)
             unknownparams = part.mandatorykeys - handler.params
             if unknownparams:
                 unknownparams = list(unknownparams)
                 unknownparams.sort()
                 status = 'unsupported-params (%s)' % unknownparams
                 raise error.BundleUnknownFeatureError(parttype=part.type,
                                                       params=unknownparams)
             status = 'supported'
         except error.BundleUnknownFeatureError as exc:
             if part.mandatory: # mandatory parts
                 raise
             indebug(op.ui, 'ignoring unsupported advisory part %s' % exc)
             return # skip to part processing
         finally:
             if op.ui.debugflag:
                 msg = ['bundle2-input-part: "%s"' % part.type]
                 if not part.mandatory:
                     msg.append(' (advisory)')
                 nbmp = len(part.mandatorykeys)
                 nbap = len(part.params) - nbmp
                 if nbmp or nbap:
                     msg.append(' (params:')
                     if nbmp:
                         msg.append(' %i mandatory' % nbmp)
                     if nbap:
                         msg.append(' %i advisory' % nbmp)
                     msg.append(')')
                 msg.append(' %s\n' % status)
                 op.ui.debug(''.join(msg))
 
         # handler is called outside the above try block so that we don't
         # risk catching KeyErrors from anything other than the
         # parthandlermapping lookup (any KeyError raised by handler()
         # itself represents a defect of a different variety).
         output = None
         if op.captureoutput and op.reply is not None:
             op.ui.pushbuffer(error=True, subproc=True)
             output = ''
         try:
             handler(op, part)
         finally:
             if output is not None:
                 output = op.ui.popbuffer()
             if output:
                 outpart = op.reply.newpart('output', data=output,
                                            mandatory=False)
                 outpart.addparam('in-reply-to', str(part.id), mandatory=False)
     # If exiting or interrupted, do not attempt to seek the stream in the
     # finally block below. This makes abort faster.
     except (SystemExit, KeyboardInterrupt):
         hardabort = True
         raise
     finally:
         # consume the part content to not corrupt the stream.
         if not hardabort:
             part.seek(0, 2)
 
 
 def decodecaps(blob):
     """decode a bundle2 caps bytes blob into a dictionary
 
     The blob is a list of capabilities (one per line)
     Capabilities may have values using a line of the form::
 
         capability=value1,value2,value3
 
     The values are always a list."""
     caps = {}
     for line in blob.splitlines():
         if not line:
             continue
         if '=' not in line:
             key, vals = line, ()
         else:
             key, vals = line.split('=', 1)
             vals = vals.split(',')
         key = urlreq.unquote(key)
         vals = [urlreq.unquote(v) for v in vals]
         caps[key] = vals
     return caps
 
 def encodecaps(caps):
     """encode a bundle2 caps dictionary into a bytes blob"""
     chunks = []
     for ca in sorted(caps):
         vals = caps[ca]
         ca = urlreq.quote(ca)
         vals = [urlreq.quote(v) for v in vals]
         if vals:
             ca = "%s=%s" % (ca, ','.join(vals))
         chunks.append(ca)
     return '\n'.join(chunks)
 
 bundletypes = {
     "": ("", 'UN'),       # only when using unbundle on ssh and old http servers
                           # since the unification ssh accepts a header but there
                           # is no capability signaling it.
     "HG20": (), # special-cased below
     "HG10UN": ("HG10UN", 'UN'),
     "HG10BZ": ("HG10", 'BZ'),
     "HG10GZ": ("HG10GZ", 'GZ'),
 }
 
 # hgweb uses this list to communicate its preferred type
 bundlepriority = ['HG10GZ', 'HG10BZ', 'HG10UN']
 
 class bundle20(object):
     """represent an outgoing bundle2 container
 
     Use the `addparam` method to add stream level parameter. and `newpart` to
     populate it. Then call `getchunks` to retrieve all the binary chunks of
     data that compose the bundle2 container."""
 
     _magicstring = 'HG20'
 
     def __init__(self, ui, capabilities=()):
         self.ui = ui
         self._params = []
         self._parts = []
         self.capabilities = dict(capabilities)
         self._compengine = util.compengines.forbundletype('UN')
         self._compopts = None
 
     def setcompression(self, alg, compopts=None):
         """setup core part compression to <alg>"""
         if alg in (None, 'UN'):
             return
         assert not any(n.lower() == 'compression' for n, v in self._params)
         self.addparam('Compression', alg)
         self._compengine = util.compengines.forbundletype(alg)
         self._compopts = compopts
 
     @property
     def nbparts(self):
         """total number of parts added to the bundler"""
         return len(self._parts)
 
     # methods used to defines the bundle2 content
     def addparam(self, name, value=None):
         """add a stream level parameter"""
         if not name:
             raise ValueError('empty parameter name')
         if name[0] not in string.letters:
             raise ValueError('non letter first character: %r' % name)
         self._params.append((name, value))
 
     def addpart(self, part):
         """add a new part to the bundle2 container
 
         Parts contains the actual applicative payload."""
         assert part.id is None
         part.id = len(self._parts) # very cheap counter
         self._parts.append(part)
 
     def newpart(self, typeid, *args, **kwargs):
         """create a new part and add it to the containers
 
         As the part is directly added to the containers. For now, this means
         that any failure to properly initialize the part after calling
         ``newpart`` should result in a failure of the whole bundling process.
 
         You can still fall back to manually create and add if you need better
         control."""
         part = bundlepart(typeid, *args, **kwargs)
         self.addpart(part)
         return part
 
     # methods used to generate the bundle2 stream
     def getchunks(self):
         if self.ui.debugflag:
             msg = ['bundle2-output-bundle: "%s",' % self._magicstring]
             if self._params:
                 msg.append(' (%i params)' % len(self._params))
             msg.append(' %i parts total\n' % len(self._parts))
             self.ui.debug(''.join(msg))
         outdebug(self.ui, 'start emission of %s stream' % self._magicstring)
         yield self._magicstring
         param = self._paramchunk()
         outdebug(self.ui, 'bundle parameter: %s' % param)
         yield _pack(_fstreamparamsize, len(param))
         if param:
             yield param
         for chunk in self._compengine.compressstream(self._getcorechunk(),
                                                      self._compopts):
             yield chunk
 
     def _paramchunk(self):
         """return a encoded version of all stream parameters"""
         blocks = []
         for par, value in self._params:
             par = urlreq.quote(par)
             if value is not None:
                 value = urlreq.quote(value)
                 par = '%s=%s' % (par, value)
             blocks.append(par)
         return ' '.join(blocks)
 
     def _getcorechunk(self):
         """yield chunk for the core part of the bundle
 
         (all but headers and parameters)"""
         outdebug(self.ui, 'start of parts')
         for part in self._parts:
             outdebug(self.ui, 'bundle part: "%s"' % part.type)
             for chunk in part.getchunks(ui=self.ui):
                 yield chunk
         outdebug(self.ui, 'end of bundle')
         yield _pack(_fpartheadersize, 0)
 
 
     def salvageoutput(self):
         """return a list with a copy of all output parts in the bundle
 
         This is meant to be used during error handling to make sure we preserve
         server output"""
         salvaged = []
         for part in self._parts:
             if part.type.startswith('output'):
                 salvaged.append(part.copy())
         return salvaged
 
 
 class unpackermixin(object):
     """A mixin to extract bytes and struct data from a stream"""
 
     def __init__(self, fp):
         self._fp = fp
 
     def _unpack(self, format):
         """unpack this struct format from the stream
 
         This method is meant for internal usage by the bundle2 protocol only.
         They directly manipulate the low level stream including bundle2 level
         instruction.
 
         Do not use it to implement higher-level logic or methods."""
         data = self._readexact(struct.calcsize(format))
         return _unpack(format, data)
 
     def _readexact(self, size):
         """read exactly <size> bytes from the stream
 
         This method is meant for internal usage by the bundle2 protocol only.
         They directly manipulate the low level stream including bundle2 level
         instruction.
 
         Do not use it to implement higher-level logic or methods."""
         return changegroup.readexactly(self._fp, size)
 
 def getunbundler(ui, fp, magicstring=None):
     """return a valid unbundler object for a given magicstring"""
     if magicstring is None:
         magicstring = changegroup.readexactly(fp, 4)
     magic, version = magicstring[0:2], magicstring[2:4]
     if magic != 'HG':
         raise error.Abort(_('not a Mercurial bundle'))
     unbundlerclass = formatmap.get(version)
     if unbundlerclass is None:
         raise error.Abort(_('unknown bundle version %s') % version)
     unbundler = unbundlerclass(ui, fp)
     indebug(ui, 'start processing of %s stream' % magicstring)
     return unbundler
 
 class unbundle20(unpackermixin):
     """interpret a bundle2 stream
 
     This class is fed with a binary stream and yields parts through its
     `iterparts` methods."""
 
     _magicstring = 'HG20'
 
     def __init__(self, ui, fp):
         """If header is specified, we do not read it out of the stream."""
         self.ui = ui
         self._compengine = util.compengines.forbundletype('UN')
         self._compressed = None
         super(unbundle20, self).__init__(fp)
 
     @util.propertycache
     def params(self):
         """dictionary of stream level parameters"""
         indebug(self.ui, 'reading bundle2 stream parameters')
         params = {}
         paramssize = self._unpack(_fstreamparamsize)[0]
         if paramssize < 0:
             raise error.BundleValueError('negative bundle param size: %i'
                                          % paramssize)
         if paramssize:
             params = self._readexact(paramssize)
             params = self._processallparams(params)
         return params
 
     def _processallparams(self, paramsblock):
         """"""
         params = util.sortdict()
         for p in paramsblock.split(' '):
             p = p.split('=', 1)
             p = [urlreq.unquote(i) for i in p]
             if len(p) < 2:
                 p.append(None)
             self._processparam(*p)
             params[p[0]] = p[1]
         return params
 
 
     def _processparam(self, name, value):
         """process a parameter, applying its effect if needed
 
         Parameter starting with a lower case letter are advisory and will be
         ignored when unknown.  Those starting with an upper case letter are
         mandatory and will this function will raise a KeyError when unknown.
 
         Note: no option are currently supported. Any input will be either
               ignored or failing.
         """
         if not name:
             raise ValueError('empty parameter name')
         if name[0] not in string.letters:
             raise ValueError('non letter first character: %r' % name)
         try:
             handler = b2streamparamsmap[name.lower()]
         except KeyError:
             if name[0].islower():
                 indebug(self.ui, "ignoring unknown parameter %r" % name)
             else:
                 raise error.BundleUnknownFeatureError(params=(name,))
         else:
             handler(self, name, value)
 
     def _forwardchunks(self):
         """utility to transfer a bundle2 as binary
 
         This is made necessary by the fact the 'getbundle' command over 'ssh'
         have no way to know then the reply end, relying on the bundle to be
         interpreted to know its end. This is terrible and we are sorry, but we
         needed to move forward to get general delta enabled.
         """
         yield self._magicstring
         assert 'params' not in vars(self)
         paramssize = self._unpack(_fstreamparamsize)[0]
         if paramssize < 0:
             raise error.BundleValueError('negative bundle param size: %i'
                                          % paramssize)
         yield _pack(_fstreamparamsize, paramssize)
         if paramssize:
             params = self._readexact(paramssize)
             self._processallparams(params)
             yield params
             assert self._compengine.bundletype == 'UN'
         # From there, payload might need to be decompressed
         self._fp = self._compengine.decompressorreader(self._fp)
         emptycount = 0
         while emptycount < 2:
             # so we can brainlessly loop
             assert _fpartheadersize == _fpayloadsize
             size = self._unpack(_fpartheadersize)[0]
             yield _pack(_fpartheadersize, size)
             if size:
                 emptycount = 0
             else:
                 emptycount += 1
                 continue
             if size == flaginterrupt:
                 continue
             elif size < 0:
                 raise error.BundleValueError('negative chunk size: %i')
             yield self._readexact(size)
 
 
     def iterparts(self):
         """yield all parts contained in the stream"""
         # make sure param have been loaded
         self.params
         # From there, payload need to be decompressed
         self._fp = self._compengine.decompressorreader(self._fp)
         indebug(self.ui, 'start extraction of bundle2 parts')
         headerblock = self._readpartheader()
         while headerblock is not None:
             part = unbundlepart(self.ui, headerblock, self._fp)
             yield part
             part.seek(0, 2)
             headerblock = self._readpartheader()
         indebug(self.ui, 'end of bundle2 stream')
 
     def _readpartheader(self):
         """reads a part header size and return the bytes blob
 
         returns None if empty"""
         headersize = self._unpack(_fpartheadersize)[0]
         if headersize < 0:
             raise error.BundleValueError('negative part header size: %i'
                                          % headersize)
         indebug(self.ui, 'part header size: %i' % headersize)
         if headersize:
             return self._readexact(headersize)
         return None
 
     def compressed(self):
         self.params # load params
         return self._compressed
 
     def close(self):
         """close underlying file"""
         if util.safehasattr(self._fp, 'close'):
             return self._fp.close()
 
 formatmap = {'20': unbundle20}
 
 b2streamparamsmap = {}
 
 def b2streamparamhandler(name):
     """register a handler for a stream level parameter"""
     def decorator(func):
         assert name not in formatmap
         b2streamparamsmap[name] = func
         return func
     return decorator
 
 @b2streamparamhandler('compression')
 def processcompression(unbundler, param, value):
     """read compression parameter and install payload decompression"""
     if value not in util.compengines.supportedbundletypes:
         raise error.BundleUnknownFeatureError(params=(param,),
                                               values=(value,))
     unbundler._compengine = util.compengines.forbundletype(value)
     if value is not None:
         unbundler._compressed = True
 
 class bundlepart(object):
     """A bundle2 part contains application level payload
 
     The part `type` is used to route the part to the application level
     handler.
 
     The part payload is contained in ``part.data``. It could be raw bytes or a
     generator of byte chunks.
 
     You can add parameters to the part using the ``addparam`` method.
     Parameters can be either mandatory (default) or advisory. Remote side
     should be able to safely ignore the advisory ones.
 
     Both data and parameters cannot be modified after the generation has begun.
     """
 
     def __init__(self, parttype, mandatoryparams=(), advisoryparams=(),
                  data='', mandatory=True):
         validateparttype(parttype)
         self.id = None
         self.type = parttype
         self._data = data
         self._mandatoryparams = list(mandatoryparams)
         self._advisoryparams = list(advisoryparams)
         # checking for duplicated entries
         self._seenparams = set()
         for pname, __ in self._mandatoryparams + self._advisoryparams:
             if pname in self._seenparams:
                 raise error.ProgrammingError('duplicated params: %s' % pname)
             self._seenparams.add(pname)
         # status of the part's generation:
         # - None: not started,
         # - False: currently generated,
         # - True: generation done.
         self._generated = None
         self.mandatory = mandatory
 
     def __repr__(self):
         cls = "%s.%s" % (self.__class__.__module__, self.__class__.__name__)
         return ('<%s object at %x; id: %s; type: %s; mandatory: %s>'
                 % (cls, id(self), self.id, self.type, self.mandatory))
 
     def copy(self):
         """return a copy of the part
 
         The new part have the very same content but no partid assigned yet.
         Parts with generated data cannot be copied."""
         assert not util.safehasattr(self.data, 'next')
         return self.__class__(self.type, self._mandatoryparams,
                               self._advisoryparams, self._data, self.mandatory)
 
     # methods used to defines the part content
     @property
     def data(self):
         return self._data
 
     @data.setter
     def data(self, data):
         if self._generated is not None:
             raise error.ReadOnlyPartError('part is being generated')
         self._data = data
 
     @property
     def mandatoryparams(self):
         # make it an immutable tuple to force people through ``addparam``
         return tuple(self._mandatoryparams)
 
     @property
     def advisoryparams(self):
         # make it an immutable tuple to force people through ``addparam``
         return tuple(self._advisoryparams)
 
     def addparam(self, name, value='', mandatory=True):
         """add a parameter to the part
 
         If 'mandatory' is set to True, the remote handler must claim support
         for this parameter or the unbundling will be aborted.
 
         The 'name' and 'value' cannot exceed 255 bytes each.
         """
         if self._generated is not None:
             raise error.ReadOnlyPartError('part is being generated')
         if name in self._seenparams:
             raise ValueError('duplicated params: %s' % name)
         self._seenparams.add(name)
         params = self._advisoryparams
         if mandatory:
             params = self._mandatoryparams
         params.append((name, value))
 
     # methods used to generates the bundle2 stream
     def getchunks(self, ui):
         if self._generated is not None:
             raise error.ProgrammingError('part can only be consumed once')
         self._generated = False
 
         if ui.debugflag:
             msg = ['bundle2-output-part: "%s"' % self.type]
             if not self.mandatory:
                 msg.append(' (advisory)')
             nbmp = len(self.mandatoryparams)
             nbap = len(self.advisoryparams)
             if nbmp or nbap:
                 msg.append(' (params:')
                 if nbmp:
                     msg.append(' %i mandatory' % nbmp)
                 if nbap:
                     msg.append(' %i advisory' % nbmp)
                 msg.append(')')
             if not self.data:
                 msg.append(' empty payload')
             elif util.safehasattr(self.data, 'next'):
                 msg.append(' streamed payload')
             else:
                 msg.append(' %i bytes payload' % len(self.data))
             msg.append('\n')
             ui.debug(''.join(msg))
 
         #### header
         if self.mandatory:
             parttype = self.type.upper()
         else:
             parttype = self.type.lower()
         outdebug(ui, 'part %s: "%s"' % (self.id, parttype))
         ## parttype
         header = [_pack(_fparttypesize, len(parttype)),
                   parttype, _pack(_fpartid, self.id),
                  ]
         ## parameters
         # count
         manpar = self.mandatoryparams
         advpar = self.advisoryparams
         header.append(_pack(_fpartparamcount, len(manpar), len(advpar)))
         # size
         parsizes = []
         for key, value in manpar:
             parsizes.append(len(key))
             parsizes.append(len(value))
         for key, value in advpar:
             parsizes.append(len(key))
             parsizes.append(len(value))
         paramsizes = _pack(_makefpartparamsizes(len(parsizes) / 2), *parsizes)
         header.append(paramsizes)
         # key, value
         for key, value in manpar:
             header.append(key)
             header.append(value)
         for key, value in advpar:
             header.append(key)
             header.append(value)
         ## finalize header
         headerchunk = ''.join(header)
         outdebug(ui, 'header chunk size: %i' % len(headerchunk))
         yield _pack(_fpartheadersize, len(headerchunk))
         yield headerchunk
         ## payload
         try:
             for chunk in self._payloadchunks():
                 outdebug(ui, 'payload chunk size: %i' % len(chunk))
                 yield _pack(_fpayloadsize, len(chunk))
                 yield chunk
         except GeneratorExit:
             # GeneratorExit means that nobody is listening for our
             # results anyway, so just bail quickly rather than trying
             # to produce an error part.
             ui.debug('bundle2-generatorexit\n')
             raise
         except BaseException as exc:
             # backup exception data for later
             ui.debug('bundle2-input-stream-interrupt: encoding exception %s'
                      % exc)
             tb = sys.exc_info()[2]
             msg = 'unexpected error: %s' % exc
             interpart = bundlepart('error:abort', [('message', msg)],
                                    mandatory=False)
             interpart.id = 0
             yield _pack(_fpayloadsize, -1)
             for chunk in interpart.getchunks(ui=ui):
                 yield chunk
             outdebug(ui, 'closing payload chunk')
             # abort current part payload
             yield _pack(_fpayloadsize, 0)
             pycompat.raisewithtb(exc, tb)
         # end of payload
         outdebug(ui, 'closing payload chunk')
         yield _pack(_fpayloadsize, 0)
         self._generated = True
 
     def _payloadchunks(self):
         """yield chunks of a the part payload
 
         Exists to handle the different methods to provide data to a part."""
         # we only support fixed size data now.
         # This will be improved in the future.
         if util.safehasattr(self.data, 'next'):
             buff = util.chunkbuffer(self.data)
             chunk = buff.read(preferedchunksize)
             while chunk:
                 yield chunk
                 chunk = buff.read(preferedchunksize)
         elif len(self.data):
             yield self.data
 
 
 flaginterrupt = -1
 
 class interrupthandler(unpackermixin):
     """read one part and process it with restricted capability
 
     This allows to transmit exception raised on the producer size during part
     iteration while the consumer is reading a part.
 
     Part processed in this manner only have access to a ui object,"""
 
     def __init__(self, ui, fp):
         super(interrupthandler, self).__init__(fp)
         self.ui = ui
 
     def _readpartheader(self):
         """reads a part header size and return the bytes blob
 
         returns None if empty"""
         headersize = self._unpack(_fpartheadersize)[0]
         if headersize < 0:
             raise error.BundleValueError('negative part header size: %i'
                                          % headersize)
         indebug(self.ui, 'part header size: %i\n' % headersize)
         if headersize:
             return self._readexact(headersize)
         return None
 
     def __call__(self):
 
         self.ui.debug('bundle2-input-stream-interrupt:'
                       ' opening out of band context\n')
         indebug(self.ui, 'bundle2 stream interruption, looking for a part.')
         headerblock = self._readpartheader()
         if headerblock is None:
             indebug(self.ui, 'no part found during interruption.')
             return
         part = unbundlepart(self.ui, headerblock, self._fp)
         op = interruptoperation(self.ui)
         _processpart(op, part)
         self.ui.debug('bundle2-input-stream-interrupt:'
                       ' closing out of band context\n')
 
 class interruptoperation(object):
     """A limited operation to be use by part handler during interruption
 
     It only have access to an ui object.
     """
 
     def __init__(self, ui):
         self.ui = ui
         self.reply = None
         self.captureoutput = False
 
     @property
     def repo(self):
         raise error.ProgrammingError('no repo access from stream interruption')
 
     def gettransaction(self):
         raise TransactionUnavailable('no repo access from stream interruption')
 
 class unbundlepart(unpackermixin):
     """a bundle part read from a bundle"""
 
     def __init__(self, ui, header, fp):
         super(unbundlepart, self).__init__(fp)
         self._seekable = (util.safehasattr(fp, 'seek') and
                           util.safehasattr(fp, 'tell'))
         self.ui = ui
         # unbundle state attr
         self._headerdata = header
         self._headeroffset = 0
         self._initialized = False
         self.consumed = False
         # part data
         self.id = None
         self.type = None
         self.mandatoryparams = None
         self.advisoryparams = None
         self.params = None
         self.mandatorykeys = ()
         self._payloadstream = None
         self._readheader()
         self._mandatory = None
         self._chunkindex = [] #(payload, file) position tuples for chunk starts
         self._pos = 0
 
     def _fromheader(self, size):
         """return the next <size> byte from the header"""
         offset = self._headeroffset
         data = self._headerdata[offset:(offset + size)]
         self._headeroffset = offset + size
         return data
 
     def _unpackheader(self, format):
         """read given format from header
 
         This automatically compute the size of the format to read."""
         data = self._fromheader(struct.calcsize(format))
         return _unpack(format, data)
 
     def _initparams(self, mandatoryparams, advisoryparams):
         """internal function to setup all logic related parameters"""
         # make it read only to prevent people touching it by mistake.
         self.mandatoryparams = tuple(mandatoryparams)
         self.advisoryparams  = tuple(advisoryparams)
         # user friendly UI
         self.params = util.sortdict(self.mandatoryparams)
         self.params.update(self.advisoryparams)
         self.mandatorykeys = frozenset(p[0] for p in mandatoryparams)
 
     def _payloadchunks(self, chunknum=0):
         '''seek to specified chunk and start yielding data'''
         if len(self._chunkindex) == 0:
             assert chunknum == 0, 'Must start with chunk 0'
             self._chunkindex.append((0, self._tellfp()))
         else:
             assert chunknum < len(self._chunkindex), \
                    'Unknown chunk %d' % chunknum
             self._seekfp(self._chunkindex[chunknum][1])
 
         pos = self._chunkindex[chunknum][0]
         payloadsize = self._unpack(_fpayloadsize)[0]
         indebug(self.ui, 'payload chunk size: %i' % payloadsize)
         while payloadsize:
             if payloadsize == flaginterrupt:
                 # interruption detection, the handler will now read a
                 # single part and process it.
                 interrupthandler(self.ui, self._fp)()
             elif payloadsize < 0:
                 msg = 'negative payload chunk size: %i' %  payloadsize
                 raise error.BundleValueError(msg)
             else:
                 result = self._readexact(payloadsize)
                 chunknum += 1
                 pos += payloadsize
                 if chunknum == len(self._chunkindex):
                     self._chunkindex.append((pos, self._tellfp()))
                 yield result
             payloadsize = self._unpack(_fpayloadsize)[0]
             indebug(self.ui, 'payload chunk size: %i' % payloadsize)
 
     def _findchunk(self, pos):
         '''for a given payload position, return a chunk number and offset'''
         for chunk, (ppos, fpos) in enumerate(self._chunkindex):
             if ppos == pos:
                 return chunk, 0
             elif ppos > pos:
                 return chunk - 1, pos - self._chunkindex[chunk - 1][0]
         raise ValueError('Unknown chunk')
 
     def _readheader(self):
         """read the header and setup the object"""
         typesize = self._unpackheader(_fparttypesize)[0]
         self.type = self._fromheader(typesize)
         indebug(self.ui, 'part type: "%s"' % self.type)
         self.id = self._unpackheader(_fpartid)[0]
         indebug(self.ui, 'part id: "%s"' % self.id)
         # extract mandatory bit from type
         self.mandatory = (self.type != self.type.lower())
         self.type = self.type.lower()
         ## reading parameters
         # param count
         mancount, advcount = self._unpackheader(_fpartparamcount)
         indebug(self.ui, 'part parameters: %i' % (mancount + advcount))
         # param size
         fparamsizes = _makefpartparamsizes(mancount + advcount)
         paramsizes = self._unpackheader(fparamsizes)
         # make it a list of couple again
         paramsizes = zip(paramsizes[::2], paramsizes[1::2])
         # split mandatory from advisory
         mansizes = paramsizes[:mancount]
         advsizes = paramsizes[mancount:]
         # retrieve param value
         manparams = []
         for key, value in mansizes:
             manparams.append((self._fromheader(key), self._fromheader(value)))
         advparams = []
         for key, value in advsizes:
             advparams.append((self._fromheader(key), self._fromheader(value)))
         self._initparams(manparams, advparams)
         ## part payload
         self._payloadstream = util.chunkbuffer(self._payloadchunks())
         # we read the data, tell it
         self._initialized = True
 
     def read(self, size=None):
         """read payload data"""
         if not self._initialized:
             self._readheader()
         if size is None:
             data = self._payloadstream.read()
         else:
             data = self._payloadstream.read(size)
         self._pos += len(data)
         if size is None or len(data) < size:
             if not self.consumed and self._pos:
                 self.ui.debug('bundle2-input-part: total payload size %i\n'
                               % self._pos)
             self.consumed = True
         return data
 
     def tell(self):
         return self._pos
 
     def seek(self, offset, whence=0):
         if whence == 0:
             newpos = offset
         elif whence == 1:
             newpos = self._pos + offset
         elif whence == 2:
             if not self.consumed:
                 self.read()
             newpos = self._chunkindex[-1][0] - offset
         else:
             raise ValueError('Unknown whence value: %r' % (whence,))
 
         if newpos > self._chunkindex[-1][0] and not self.consumed:
             self.read()
         if not 0 <= newpos <= self._chunkindex[-1][0]:
             raise ValueError('Offset out of range')
 
         if self._pos != newpos:
             chunk, internaloffset = self._findchunk(newpos)
             self._payloadstream = util.chunkbuffer(self._payloadchunks(chunk))
             adjust = self.read(internaloffset)
             if len(adjust) != internaloffset:
                 raise error.Abort(_('Seek failed\n'))
             self._pos = newpos
 
     def _seekfp(self, offset, whence=0):
         """move the underlying file pointer
 
         This method is meant for internal usage by the bundle2 protocol only.
         They directly manipulate the low level stream including bundle2 level
         instruction.
 
         Do not use it to implement higher-level logic or methods."""
         if self._seekable:
             return self._fp.seek(offset, whence)
         else:
             raise NotImplementedError(_('File pointer is not seekable'))
 
     def _tellfp(self):
         """return the file offset, or None if file is not seekable
 
         This method is meant for internal usage by the bundle2 protocol only.
         They directly manipulate the low level stream including bundle2 level
         instruction.
 
         Do not use it to implement higher-level logic or methods."""
         if self._seekable:
             try:
                 return self._fp.tell()
             except IOError as e:
                 if e.errno == errno.ESPIPE:
                     self._seekable = False
                 else:
                     raise
         return None
 
 # These are only the static capabilities.
 # Check the 'getrepocaps' function for the rest.
 capabilities = {'HG20': (),
                 'error': ('abort', 'unsupportedcontent', 'pushraced',
                           'pushkey'),
                 'listkeys': (),
                 'pushkey': (),
                 'digests': tuple(sorted(util.DIGESTS.keys())),
                 'remote-changegroup': ('http', 'https'),
                 'hgtagsfnodes': (),
                }
 
 def getrepocaps(repo, allowpushback=False):
     """return the bundle2 capabilities for a given repo
 
     Exists to allow extensions (like evolution) to mutate the capabilities.
     """
     caps = capabilities.copy()
     caps['changegroup'] = tuple(sorted(
         changegroup.supportedincomingversions(repo)))
     if obsolete.isenabled(repo, obsolete.exchangeopt):
         supportedformat = tuple('V%i' % v for v in obsolete.formats)
         caps['obsmarkers'] = supportedformat
     if allowpushback:
         caps['pushback'] = ()
     return caps
 
 def bundle2caps(remote):
     """return the bundle capabilities of a peer as dict"""
     raw = remote.capable('bundle2')
     if not raw and raw != '':
         return {}
     capsblob = urlreq.unquote(remote.capable('bundle2'))
     return decodecaps(capsblob)
 
 def obsmarkersversion(caps):
     """extract the list of supported obsmarkers versions from a bundle2caps dict
     """
     obscaps = caps.get('obsmarkers', ())
     return [int(c[1:]) for c in obscaps if c.startswith('V')]
 
 def writenewbundle(ui, repo, source, filename, bundletype, outgoing, opts,
                    vfs=None, compression=None, compopts=None):
     if bundletype.startswith('HG10'):
         cg = changegroup.getchangegroup(repo, source, outgoing, version='01')
         return writebundle(ui, cg, filename, bundletype, vfs=vfs,
                            compression=compression, compopts=compopts)
     elif not bundletype.startswith('HG20'):
         raise error.ProgrammingError('unknown bundle type: %s' % bundletype)
 
     bundle = bundle20(ui)
     bundle.setcompression(compression, compopts)
     _addpartsfromopts(ui, repo, bundle, source, outgoing, opts)
     chunkiter = bundle.getchunks()
 
     return changegroup.writechunks(ui, chunkiter, filename, vfs=vfs)
 
 def _addpartsfromopts(ui, repo, bundler, source, outgoing, opts):
     # We should eventually reconcile this logic with the one behind
     # 'exchange.getbundle2partsgenerator'.
     #
     # The type of input from 'getbundle' and 'writenewbundle' are a bit
     # different right now. So we keep them separated for now for the sake of
     # simplicity.
 
     # we always want a changegroup in such bundle
     cgversion = opts.get('cg.version')
     if cgversion is None:
         cgversion = changegroup.safeversion(repo)
     cg = changegroup.getchangegroup(repo, source, outgoing,
                                     version=cgversion)
     part = bundler.newpart('changegroup', data=cg.getchunks())
     part.addparam('version', cg.version)
     if 'clcount' in cg.extras:
         part.addparam('nbchanges', str(cg.extras['clcount']),
                       mandatory=False)
 
     addparttagsfnodescache(repo, bundler, outgoing)
 
 def addparttagsfnodescache(repo, bundler, outgoing):
     # we include the tags fnode cache for the bundle changeset
     # (as an optional parts)
     cache = tags.hgtagsfnodescache(repo.unfiltered())
     chunks = []
 
     # .hgtags fnodes are only relevant for head changesets. While we could
     # transfer values for all known nodes, there will likely be little to
     # no benefit.
     #
     # We don't bother using a generator to produce output data because
     # a) we only have 40 bytes per head and even esoteric numbers of heads
     # consume little memory (1M heads is 40MB) b) we don't want to send the
     # part if we don't have entries and knowing if we have entries requires
     # cache lookups.
     for node in outgoing.missingheads:
         # Don't compute missing, as this may slow down serving.
         fnode = cache.getfnode(node, computemissing=False)
         if fnode is not None:
             chunks.extend([node, fnode])
 
     if chunks:
         bundler.newpart('hgtagsfnodes', data=''.join(chunks))
 
 def writebundle(ui, cg, filename, bundletype, vfs=None, compression=None,
                 compopts=None):
     """Write a bundle file and return its filename.
 
     Existing files will not be overwritten.
     If no filename is specified, a temporary file is created.
     bz2 compression can be turned off.
     The bundle file will be deleted in case of errors.
     """
 
     if bundletype == "HG20":
         bundle = bundle20(ui)
         bundle.setcompression(compression, compopts)
         part = bundle.newpart('changegroup', data=cg.getchunks())
         part.addparam('version', cg.version)
         if 'clcount' in cg.extras:
             part.addparam('nbchanges', str(cg.extras['clcount']),
                           mandatory=False)
         chunkiter = bundle.getchunks()
     else:
         # compression argument is only for the bundle2 case
         assert compression is None
         if cg.version != '01':
             raise error.Abort(_('old bundle types only supports v1 '
                                 'changegroups'))
         header, comp = bundletypes[bundletype]
         if comp not in util.compengines.supportedbundletypes:
             raise error.Abort(_('unknown stream compression type: %s')
                               % comp)
         compengine = util.compengines.forbundletype(comp)
         def chunkiter():
             yield header
             for chunk in compengine.compressstream(cg.getchunks(), compopts):
                 yield chunk
         chunkiter = chunkiter()
 
     # parse the changegroup data, otherwise we will block
     # in case of sshrepo because we don't know the end of the stream
     return changegroup.writechunks(ui, chunkiter, filename, vfs=vfs)
 
 @parthandler('changegroup', ('version', 'nbchanges', 'treemanifest'))
 def handlechangegroup(op, inpart):
     """apply a changegroup part on the repo
 
     This is a very early implementation that will massive rework before being
     inflicted to any end-user.
     """
     # Make sure we trigger a transaction creation
     #
     # The addchangegroup function will get a transaction object by itself, but
     # we need to make sure we trigger the creation of a transaction object used
     # for the whole processing scope.
     op.gettransaction()
     unpackerversion = inpart.params.get('version', '01')
     # We should raise an appropriate exception here
     cg = changegroup.getunbundler(unpackerversion, inpart, None)
     # the source and url passed here are overwritten by the one contained in
     # the transaction.hookargs argument. So 'bundle2' is a placeholder
     nbchangesets = None
     if 'nbchanges' in inpart.params:
         nbchangesets = int(inpart.params.get('nbchanges'))
     if ('treemanifest' in inpart.params and
         'treemanifest' not in op.repo.requirements):
         if len(op.repo.changelog) != 0:
             raise error.Abort(_(
                 "bundle contains tree manifests, but local repo is "
                 "non-empty and does not use tree manifests"))
         op.repo.requirements.add('treemanifest')
         op.repo._applyopenerreqs()
         op.repo._writerequirements()
     ret = cg.apply(op.repo, 'bundle2', 'bundle2', expectedtotal=nbchangesets)
     op.records.add('changegroup', {'return': ret})
     if op.reply is not None:
         # This is definitely not the final form of this
         # return. But one need to start somewhere.
         part = op.reply.newpart('reply:changegroup', mandatory=False)
         part.addparam('in-reply-to', str(inpart.id), mandatory=False)
         part.addparam('return', '%i' % ret, mandatory=False)
     assert not inpart.read()
 
 _remotechangegroupparams = tuple(['url', 'size', 'digests'] +
     ['digest:%s' % k for k in util.DIGESTS.keys()])
 @parthandler('remote-changegroup', _remotechangegroupparams)
 def handleremotechangegroup(op, inpart):
     """apply a bundle10 on the repo, given an url and validation information
 
     All the information about the remote bundle to import are given as
     parameters. The parameters include:
       - url: the url to the bundle10.
       - size: the bundle10 file size. It is used to validate what was
         retrieved by the client matches the server knowledge about the bundle.
       - digests: a space separated list of the digest types provided as
         parameters.
       - digest:<digest-type>: the hexadecimal representation of the digest with
         that name. Like the size, it is used to validate what was retrieved by
         the client matches what the server knows about the bundle.
 
     When multiple digest types are given, all of them are checked.
     """
     try:
         raw_url = inpart.params['url']
     except KeyError:
         raise error.Abort(_('remote-changegroup: missing "%s" param') % 'url')
     parsed_url = util.url(raw_url)
     if parsed_url.scheme not in capabilities['remote-changegroup']:
         raise error.Abort(_('remote-changegroup does not support %s urls') %
             parsed_url.scheme)
 
     try:
         size = int(inpart.params['size'])
     except ValueError:
         raise error.Abort(_('remote-changegroup: invalid value for param "%s"')
             % 'size')
     except KeyError:
         raise error.Abort(_('remote-changegroup: missing "%s" param') % 'size')
 
     digests = {}
     for typ in inpart.params.get('digests', '').split():
         param = 'digest:%s' % typ
         try:
             value = inpart.params[param]
         except KeyError:
             raise error.Abort(_('remote-changegroup: missing "%s" param') %
                 param)
         digests[typ] = value
 
     real_part = util.digestchecker(url.open(op.ui, raw_url), size, digests)
 
     # Make sure we trigger a transaction creation
     #
     # The addchangegroup function will get a transaction object by itself, but
     # we need to make sure we trigger the creation of a transaction object used
     # for the whole processing scope.
     op.gettransaction()
     from . import exchange
     cg = exchange.readbundle(op.repo.ui, real_part, raw_url)
     if not isinstance(cg, changegroup.cg1unpacker):
         raise error.Abort(_('%s: not a bundle version 1.0') %
             util.hidepassword(raw_url))
     ret = cg.apply(op.repo, 'bundle2', 'bundle2')
     op.records.add('changegroup', {'return': ret})
     if op.reply is not None:
         # This is definitely not the final form of this
         # return. But one need to start somewhere.
         part = op.reply.newpart('reply:changegroup')
         part.addparam('in-reply-to', str(inpart.id), mandatory=False)
         part.addparam('return', '%i' % ret, mandatory=False)
     try:
         real_part.validate()
     except error.Abort as e:
         raise error.Abort(_('bundle at %s is corrupted:\n%s') %
             (util.hidepassword(raw_url), str(e)))
     assert not inpart.read()
 
 @parthandler('reply:changegroup', ('return', 'in-reply-to'))
 def handlereplychangegroup(op, inpart):
     ret = int(inpart.params['return'])
     replyto = int(inpart.params['in-reply-to'])
     op.records.add('changegroup', {'return': ret}, replyto)
 
 @parthandler('check:heads')
 def handlecheckheads(op, inpart):
     """check that head of the repo did not change
 
     This is used to detect a push race when using unbundle.
     This replaces the "heads" argument of unbundle."""
     h = inpart.read(20)
     heads = []
     while len(h) == 20:
         heads.append(h)
         h = inpart.read(20)
     assert not h
     # Trigger a transaction so that we are guaranteed to have the lock now.
     if op.ui.configbool('experimental', 'bundle2lazylocking'):
         op.gettransaction()
     if sorted(heads) != sorted(op.repo.heads()):
         raise error.PushRaced('repository changed while pushing - '
                               'please try again')
 
 @parthandler('output')
 def handleoutput(op, inpart):
     """forward output captured on the server to the client"""
     for line in inpart.read().splitlines():
         op.ui.status(_('remote: %s\n') % line)
 
 @parthandler('replycaps')
 def handlereplycaps(op, inpart):
     """Notify that a reply bundle should be created
 
     The payload contains the capabilities information for the reply"""
     caps = decodecaps(inpart.read())
     if op.reply is None:
         op.reply = bundle20(op.ui, caps)
 
 class AbortFromPart(error.Abort):
     """Sub-class of Abort that denotes an error from a bundle2 part."""
 
 @parthandler('error:abort', ('message', 'hint'))
 def handleerrorabort(op, inpart):
     """Used to transmit abort error over the wire"""
     raise AbortFromPart(inpart.params['message'],
                         hint=inpart.params.get('hint'))
 
 @parthandler('error:pushkey', ('namespace', 'key', 'new', 'old', 'ret',
                                'in-reply-to'))
 def handleerrorpushkey(op, inpart):
     """Used to transmit failure of a mandatory pushkey over the wire"""
     kwargs = {}
     for name in ('namespace', 'key', 'new', 'old', 'ret'):
         value = inpart.params.get(name)
         if value is not None:
             kwargs[name] = value
     raise error.PushkeyFailed(inpart.params['in-reply-to'], **kwargs)
 
 @parthandler('error:unsupportedcontent', ('parttype', 'params'))
 def handleerrorunsupportedcontent(op, inpart):
     """Used to transmit unknown content error over the wire"""
     kwargs = {}
     parttype = inpart.params.get('parttype')
     if parttype is not None:
         kwargs['parttype'] = parttype
     params = inpart.params.get('params')
     if params is not None:
         kwargs['params'] = params.split('\0')
 
     raise error.BundleUnknownFeatureError(**kwargs)
 
 @parthandler('error:pushraced', ('message',))
 def handleerrorpushraced(op, inpart):
     """Used to transmit push race error over the wire"""
     raise error.ResponseError(_('push failed:'), inpart.params['message'])
 
 @parthandler('listkeys', ('namespace',))
 def handlelistkeys(op, inpart):
     """retrieve pushkey namespace content stored in a bundle2"""
     namespace = inpart.params['namespace']
     r = pushkey.decodekeys(inpart.read())
     op.records.add('listkeys', (namespace, r))
 
 @parthandler('pushkey', ('namespace', 'key', 'old', 'new'))
 def handlepushkey(op, inpart):
     """process a pushkey request"""
     dec = pushkey.decode
     namespace = dec(inpart.params['namespace'])
     key = dec(inpart.params['key'])
     old = dec(inpart.params['old'])
     new = dec(inpart.params['new'])
     # Grab the transaction to ensure that we have the lock before performing the
     # pushkey.
     if op.ui.configbool('experimental', 'bundle2lazylocking'):
         op.gettransaction()
     ret = op.repo.pushkey(namespace, key, old, new)
     record = {'namespace': namespace,
               'key': key,
               'old': old,
               'new': new}
     op.records.add('pushkey', record)
     if op.reply is not None:
         rpart = op.reply.newpart('reply:pushkey')
         rpart.addparam('in-reply-to', str(inpart.id), mandatory=False)
         rpart.addparam('return', '%i' % ret, mandatory=False)
     if inpart.mandatory and not ret:
         kwargs = {}
         for key in ('namespace', 'key', 'new', 'old', 'ret'):
             if key in inpart.params:
                 kwargs[key] = inpart.params[key]
         raise error.PushkeyFailed(partid=str(inpart.id), **kwargs)
 
 @parthandler('reply:pushkey', ('return', 'in-reply-to'))
 def handlepushkeyreply(op, inpart):
     """retrieve the result of a pushkey request"""
     ret = int(inpart.params['return'])
     partid = int(inpart.params['in-reply-to'])
     op.records.add('pushkey', {'return': ret}, partid)
 
 @parthandler('obsmarkers')
 def handleobsmarker(op, inpart):
     """add a stream of obsmarkers to the repo"""
     tr = op.gettransaction()
     markerdata = inpart.read()
     if op.ui.config('experimental', 'obsmarkers-exchange-debug', False):
         op.ui.write(('obsmarker-exchange: %i bytes received\n')
                     % len(markerdata))
     # The mergemarkers call will crash if marker creation is not enabled.
     # we want to avoid this if the part is advisory.
     if not inpart.mandatory and op.repo.obsstore.readonly:
         op.repo.ui.debug('ignoring obsolescence markers, feature not enabled')
         return
     new = op.repo.obsstore.mergemarkers(tr, markerdata)
+    op.repo.invalidatevolatilesets()
     if new:
         op.repo.ui.status(_('%i new obsolescence markers\n') % new)
     op.records.add('obsmarkers', {'new': new})
     if op.reply is not None:
         rpart = op.reply.newpart('reply:obsmarkers')
         rpart.addparam('in-reply-to', str(inpart.id), mandatory=False)
         rpart.addparam('new', '%i' % new, mandatory=False)
 
 
 @parthandler('reply:obsmarkers', ('new', 'in-reply-to'))
 def handleobsmarkerreply(op, inpart):
     """retrieve the result of a pushkey request"""
     ret = int(inpart.params['new'])
     partid = int(inpart.params['in-reply-to'])
     op.records.add('obsmarkers', {'new': ret}, partid)
 
 @parthandler('hgtagsfnodes')
 def handlehgtagsfnodes(op, inpart):
     """Applies .hgtags fnodes cache entries to the local repo.
 
     Payload is pairs of 20 byte changeset nodes and filenodes.
     """
     # Grab the transaction so we ensure that we have the lock at this point.
     if op.ui.configbool('experimental', 'bundle2lazylocking'):
         op.gettransaction()
     cache = tags.hgtagsfnodescache(op.repo.unfiltered())
 
     count = 0
     while True:
         node = inpart.read(20)
         fnode = inpart.read(20)
         if len(node) < 20 or len(fnode) < 20:
             op.ui.debug('ignoring incomplete received .hgtags fnodes data\n')
             break
         cache.setfnode(node, fnode)
         count += 1
 
     cache.write()
     op.ui.debug('applied %i hgtags fnodes cache entries\n' % count)

mercurial/obsolete.py

@@ -1,1292 +1,1293 b''
 # obsolete.py - obsolete markers handling
 #
 # Copyright 2012 Pierre-Yves David <[e-mail]>
 #                Logilab SA        <[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.
 
 """Obsolete marker handling
 
 An obsolete marker maps an old changeset to a list of new
 changesets. If the list of new changesets is empty, the old changeset
 is said to be "killed". Otherwise, the old changeset is being
 "replaced" by the new changesets.
 
 Obsolete markers can be used to record and distribute changeset graph
 transformations performed by history rewrite operations, and help
 building new tools to reconcile conflicting rewrite actions. To
 facilitate conflict resolution, markers include various annotations
 besides old and news changeset identifiers, such as creation date or
 author name.
 
 The old obsoleted changeset is called a "precursor" and possible
 replacements are called "successors". Markers that used changeset X as
 a precursor are called "successor markers of X" because they hold
 information about the successors of X. Markers that use changeset Y as
 a successors are call "precursor markers of Y" because they hold
 information about the precursors of Y.
 
 Examples:
 
 - When changeset A is replaced by changeset A', one marker is stored:
 
     (A, (A',))
 
 - When changesets A and B are folded into a new changeset C, two markers are
   stored:
 
     (A, (C,)) and (B, (C,))
 
 - When changeset A is simply "pruned" from the graph, a marker is created:
 
     (A, ())
 
 - When changeset A is split into B and C, a single marker is used:
 
     (A, (B, C))
 
   We use a single marker to distinguish the "split" case from the "divergence"
   case. If two independent operations rewrite the same changeset A in to A' and
   A'', we have an error case: divergent rewriting. We can detect it because
   two markers will be created independently:
 
   (A, (B,)) and (A, (C,))
 
 Format
 ------
 
 Markers are stored in an append-only file stored in
 '.hg/store/obsstore'.
 
 The file starts with a version header:
 
 - 1 unsigned byte: version number, starting at zero.
 
 The header is followed by the markers. Marker format depend of the version. See
 comment associated with each format for details.
 
 """
 from __future__ import absolute_import
 
 import errno
 import struct
 
 from .i18n import _
 from . import (
     error,
     node,
     phases,
     policy,
     util,
 )
 
 parsers = policy.importmod(r'parsers')
 
 _pack = struct.pack
 _unpack = struct.unpack
 _calcsize = struct.calcsize
 propertycache = util.propertycache
 
 # the obsolete feature is not mature enough to be enabled by default.
 # you have to rely on third party extension extension to enable this.
 _enabled = False
 
 # Options for obsolescence
 createmarkersopt = 'createmarkers'
 allowunstableopt = 'allowunstable'
 exchangeopt = 'exchange'
 
 def isenabled(repo, option):
     """Returns True if the given repository has the given obsolete option
     enabled.
     """
     result = set(repo.ui.configlist('experimental', 'evolution'))
     if 'all' in result:
         return True
 
     # For migration purposes, temporarily return true if the config hasn't been
     # set but _enabled is true.
     if len(result) == 0 and _enabled:
         return True
 
     # createmarkers must be enabled if other options are enabled
     if ((allowunstableopt in result or exchangeopt in result) and
         not createmarkersopt in result):
         raise error.Abort(_("'createmarkers' obsolete option must be enabled "
                            "if other obsolete options are enabled"))
 
     return option in result
 
 ### obsolescence marker flag
 
 ## bumpedfix flag
 #
 # When a changeset A' succeed to a changeset A which became public, we call A'
 # "bumped" because it's a successors of a public changesets
 #
 # o    A' (bumped)
 # |`:
 # | o  A
 # |/
 # o    Z
 #
 # The way to solve this situation is to create a new changeset Ad as children
 # of A. This changeset have the same content than A'. So the diff from A to A'
 # is the same than the diff from A to Ad. Ad is marked as a successors of A'
 #
 # o   Ad
 # |`:
 # | x A'
 # |'|
 # o | A
 # |/
 # o Z
 #
 # But by transitivity Ad is also a successors of A. To avoid having Ad marked
 # as bumped too, we add the `bumpedfix` flag to the marker. <A', (Ad,)>.
 # This flag mean that the successors express the changes between the public and
 # bumped version and fix the situation, breaking the transitivity of
 # "bumped" here.
 bumpedfix = 1
 usingsha256 = 2
 
 ## Parsing and writing of version "0"
 #
 # The header is followed by the markers. Each marker is made of:
 #
 # - 1 uint8 : number of new changesets "N", can be zero.
 #
 # - 1 uint32: metadata size "M" in bytes.
 #
 # - 1 byte: a bit field. It is reserved for flags used in common
 #   obsolete marker operations, to avoid repeated decoding of metadata
 #   entries.
 #
 # - 20 bytes: obsoleted changeset identifier.
 #
 # - N*20 bytes: new changesets identifiers.
 #
 # - M bytes: metadata as a sequence of nul-terminated strings. Each
 #   string contains a key and a value, separated by a colon ':', without
 #   additional encoding. Keys cannot contain '\0' or ':' and values
 #   cannot contain '\0'.
 _fm0version = 0
 _fm0fixed   = '>BIB20s'
 _fm0node = '20s'
 _fm0fsize = _calcsize(_fm0fixed)
 _fm0fnodesize = _calcsize(_fm0node)
 
 def _fm0readmarkers(data, off):
     # Loop on markers
     l = len(data)
     while off + _fm0fsize <= l:
         # read fixed part
         cur = data[off:off + _fm0fsize]
         off += _fm0fsize
         numsuc, mdsize, flags, pre = _unpack(_fm0fixed, cur)
         # read replacement
         sucs = ()
         if numsuc:
             s = (_fm0fnodesize * numsuc)
             cur = data[off:off + s]
             sucs = _unpack(_fm0node * numsuc, cur)
             off += s
         # read metadata
         # (metadata will be decoded on demand)
         metadata = data[off:off + mdsize]
         if len(metadata) != mdsize:
             raise error.Abort(_('parsing obsolete marker: metadata is too '
                                'short, %d bytes expected, got %d')
                              % (mdsize, len(metadata)))
         off += mdsize
         metadata = _fm0decodemeta(metadata)
         try:
             when, offset = metadata.pop('date', '0 0').split(' ')
             date = float(when), int(offset)
         except ValueError:
             date = (0., 0)
         parents = None
         if 'p2' in metadata:
             parents = (metadata.pop('p1', None), metadata.pop('p2', None))
         elif 'p1' in metadata:
             parents = (metadata.pop('p1', None),)
         elif 'p0' in metadata:
             parents = ()
         if parents is not None:
             try:
                 parents = tuple(node.bin(p) for p in parents)
                 # if parent content is not a nodeid, drop the data
                 for p in parents:
                     if len(p) != 20:
                         parents = None
                         break
             except TypeError:
                 # if content cannot be translated to nodeid drop the data.
                 parents = None
 
         metadata = tuple(sorted(metadata.iteritems()))
 
         yield (pre, sucs, flags, metadata, date, parents)
 
 def _fm0encodeonemarker(marker):
     pre, sucs, flags, metadata, date, parents = marker
     if flags & usingsha256:
         raise error.Abort(_('cannot handle sha256 with old obsstore format'))
     metadata = dict(metadata)
     time, tz = date
     metadata['date'] = '%r %i' % (time, tz)
     if parents is not None:
         if not parents:
             # mark that we explicitly recorded no parents
             metadata['p0'] = ''
         for i, p in enumerate(parents, 1):
             metadata['p%i' % i] = node.hex(p)
     metadata = _fm0encodemeta(metadata)
     numsuc = len(sucs)
     format = _fm0fixed + (_fm0node * numsuc)
     data = [numsuc, len(metadata), flags, pre]
     data.extend(sucs)
     return _pack(format, *data) + metadata
 
 def _fm0encodemeta(meta):
     """Return encoded metadata string to string mapping.
 
     Assume no ':' in key and no '\0' in both key and value."""
     for key, value in meta.iteritems():
         if ':' in key or '\0' in key:
             raise ValueError("':' and '\0' are forbidden in metadata key'")
         if '\0' in value:
             raise ValueError("':' is forbidden in metadata value'")
     return '\0'.join(['%s:%s' % (k, meta[k]) for k in sorted(meta)])
 
 def _fm0decodemeta(data):
     """Return string to string dictionary from encoded version."""
     d = {}
     for l in data.split('\0'):
         if l:
             key, value = l.split(':')
             d[key] = value
     return d
 
 ## Parsing and writing of version "1"
 #
 # The header is followed by the markers. Each marker is made of:
 #
 # - uint32: total size of the marker (including this field)
 #
 # - float64: date in seconds since epoch
 #
 # - int16: timezone offset in minutes
 #
 # - uint16: a bit field. It is reserved for flags used in common
 #   obsolete marker operations, to avoid repeated decoding of metadata
 #   entries.
 #
 # - uint8: number of successors "N", can be zero.
 #
 # - uint8: number of parents "P", can be zero.
 #
 #     0: parents data stored but no parent,
 #     1: one parent stored,
 #     2: two parents stored,
 #     3: no parent data stored
 #
 # - uint8: number of metadata entries M
 #
 # - 20 or 32 bytes: precursor changeset identifier.
 #
 # - N*(20 or 32) bytes: successors changesets identifiers.
 #
 # - P*(20 or 32) bytes: parents of the precursors changesets.
 #
 # - M*(uint8, uint8): size of all metadata entries (key and value)
 #
 # - remaining bytes: the metadata, each (key, value) pair after the other.
 _fm1version = 1
 _fm1fixed = '>IdhHBBB20s'
 _fm1nodesha1 = '20s'
 _fm1nodesha256 = '32s'
 _fm1nodesha1size = _calcsize(_fm1nodesha1)
 _fm1nodesha256size = _calcsize(_fm1nodesha256)
 _fm1fsize = _calcsize(_fm1fixed)
 _fm1parentnone = 3
 _fm1parentshift = 14
 _fm1parentmask = (_fm1parentnone << _fm1parentshift)
 _fm1metapair = 'BB'
 _fm1metapairsize = _calcsize('BB')
 
 def _fm1purereadmarkers(data, off):
     # make some global constants local for performance
     noneflag = _fm1parentnone
     sha2flag = usingsha256
     sha1size = _fm1nodesha1size
     sha2size = _fm1nodesha256size
     sha1fmt = _fm1nodesha1
     sha2fmt = _fm1nodesha256
     metasize = _fm1metapairsize
     metafmt = _fm1metapair
     fsize = _fm1fsize
     unpack = _unpack
 
     # Loop on markers
     stop = len(data) - _fm1fsize
     ufixed = struct.Struct(_fm1fixed).unpack
 
     while off <= stop:
         # read fixed part
         o1 = off + fsize
         t, secs, tz, flags, numsuc, numpar, nummeta, prec = ufixed(data[off:o1])
 
         if flags & sha2flag:
             # FIXME: prec was read as a SHA1, needs to be amended
 
             # read 0 or more successors
             if numsuc == 1:
                 o2 = o1 + sha2size
                 sucs = (data[o1:o2],)
             else:
                 o2 = o1 + sha2size * numsuc
                 sucs = unpack(sha2fmt * numsuc, data[o1:o2])
 
             # read parents
             if numpar == noneflag:
                 o3 = o2
                 parents = None
             elif numpar == 1:
                 o3 = o2 + sha2size
                 parents = (data[o2:o3],)
             else:
                 o3 = o2 + sha2size * numpar
                 parents = unpack(sha2fmt * numpar, data[o2:o3])
         else:
             # read 0 or more successors
             if numsuc == 1:
                 o2 = o1 + sha1size
                 sucs = (data[o1:o2],)
             else:
                 o2 = o1 + sha1size * numsuc
                 sucs = unpack(sha1fmt * numsuc, data[o1:o2])
 
             # read parents
             if numpar == noneflag:
                 o3 = o2
                 parents = None
             elif numpar == 1:
                 o3 = o2 + sha1size
                 parents = (data[o2:o3],)
             else:
                 o3 = o2 + sha1size * numpar
                 parents = unpack(sha1fmt * numpar, data[o2:o3])
 
         # read metadata
         off = o3 + metasize * nummeta
         metapairsize = unpack('>' + (metafmt * nummeta), data[o3:off])
         metadata = []
         for idx in xrange(0, len(metapairsize), 2):
             o1 = off + metapairsize[idx]
             o2 = o1 + metapairsize[idx + 1]
             metadata.append((data[off:o1], data[o1:o2]))
             off = o2
 
         yield (prec, sucs, flags, tuple(metadata), (secs, tz * 60), parents)
 
 def _fm1encodeonemarker(marker):
     pre, sucs, flags, metadata, date, parents = marker
     # determine node size
     _fm1node = _fm1nodesha1
     if flags & usingsha256:
         _fm1node = _fm1nodesha256
     numsuc = len(sucs)
     numextranodes = numsuc
     if parents is None:
         numpar = _fm1parentnone
     else:
         numpar = len(parents)
         numextranodes += numpar
     formatnodes = _fm1node * numextranodes
     formatmeta = _fm1metapair * len(metadata)
     format = _fm1fixed + formatnodes + formatmeta
     # tz is stored in minutes so we divide by 60
     tz = date[1]//60
     data = [None, date[0], tz, flags, numsuc, numpar, len(metadata), pre]
     data.extend(sucs)
     if parents is not None:
         data.extend(parents)
     totalsize = _calcsize(format)
     for key, value in metadata:
         lk = len(key)
         lv = len(value)
         data.append(lk)
         data.append(lv)
         totalsize += lk + lv
     data[0] = totalsize
     data = [_pack(format, *data)]
     for key, value in metadata:
         data.append(key)
         data.append(value)
     return ''.join(data)
 
 def _fm1readmarkers(data, off):
     native = getattr(parsers, 'fm1readmarkers', None)
     if not native:
         return _fm1purereadmarkers(data, off)
     stop = len(data) - _fm1fsize
     return native(data, off, stop)
 
 # mapping to read/write various marker formats
 # <version> -> (decoder, encoder)
 formats = {_fm0version: (_fm0readmarkers, _fm0encodeonemarker),
            _fm1version: (_fm1readmarkers, _fm1encodeonemarker)}
 
 @util.nogc
 def _readmarkers(data):
     """Read and enumerate markers from raw data"""
     off = 0
     diskversion = _unpack('>B', data[off:off + 1])[0]
     off += 1
     if diskversion not in formats:
         raise error.Abort(_('parsing obsolete marker: unknown version %r')
                          % diskversion)
     return diskversion, formats[diskversion][0](data, off)
 
 def encodemarkers(markers, addheader=False, version=_fm0version):
     # Kept separate from flushmarkers(), it will be reused for
     # markers exchange.
     encodeone = formats[version][1]
     if addheader:
         yield _pack('>B', version)
     for marker in markers:
         yield encodeone(marker)
 
 
 class marker(object):
     """Wrap obsolete marker raw data"""
 
     def __init__(self, repo, data):
         # the repo argument will be used to create changectx in later version
         self._repo = repo
         self._data = data
         self._decodedmeta = None
 
     def __hash__(self):
         return hash(self._data)
 
     def __eq__(self, other):
         if type(other) != type(self):
             return False
         return self._data == other._data
 
     def precnode(self):
         """Precursor changeset node identifier"""
         return self._data[0]
 
     def succnodes(self):
         """List of successor changesets node identifiers"""
         return self._data[1]
 
     def parentnodes(self):
         """Parents of the precursors (None if not recorded)"""
         return self._data[5]
 
     def metadata(self):
         """Decoded metadata dictionary"""
         return dict(self._data[3])
 
     def date(self):
         """Creation date as (unixtime, offset)"""
         return self._data[4]
 
     def flags(self):
         """The flags field of the marker"""
         return self._data[2]
 
 @util.nogc
 def _addsuccessors(successors, markers):
     for mark in markers:
         successors.setdefault(mark[0], set()).add(mark)
 
 @util.nogc
 def _addprecursors(precursors, markers):
     for mark in markers:
         for suc in mark[1]:
             precursors.setdefault(suc, set()).add(mark)
 
 @util.nogc
 def _addchildren(children, markers):
     for mark in markers:
         parents = mark[5]
         if parents is not None:
             for p in parents:
                 children.setdefault(p, set()).add(mark)
 
 def _checkinvalidmarkers(markers):
     """search for marker with invalid data and raise error if needed
 
     Exist as a separated function to allow the evolve extension for a more
     subtle handling.
     """
     for mark in markers:
         if node.nullid in mark[1]:
             raise error.Abort(_('bad obsolescence marker detected: '
                                'invalid successors nullid'))
 
 class obsstore(object):
     """Store obsolete markers
 
     Markers can be accessed with two mappings:
     - precursors[x] -> set(markers on precursors edges of x)
     - successors[x] -> set(markers on successors edges of x)
     - children[x]   -> set(markers on precursors edges of children(x)
     """
 
     fields = ('prec', 'succs', 'flag', 'meta', 'date', 'parents')
     # prec:    nodeid, precursor changesets
     # succs:   tuple of nodeid, successor changesets (0-N length)
     # flag:    integer, flag field carrying modifier for the markers (see doc)
     # meta:    binary blob, encoded metadata dictionary
     # date:    (float, int) tuple, date of marker creation
     # parents: (tuple of nodeid) or None, parents of precursors
     #          None is used when no data has been recorded
 
     def __init__(self, svfs, defaultformat=_fm1version, readonly=False):
         # caches for various obsolescence related cache
         self.caches = {}
         self.svfs = svfs
         self._version = defaultformat
         self._readonly = readonly
 
     def __iter__(self):
         return iter(self._all)
 
     def __len__(self):
         return len(self._all)
 
     def __nonzero__(self):
         if not self._cached('_all'):
             try:
                 return self.svfs.stat('obsstore').st_size > 1
             except OSError as inst:
                 if inst.errno != errno.ENOENT:
                     raise
                 # just build an empty _all list if no obsstore exists, which
                 # avoids further stat() syscalls
                 pass
         return bool(self._all)
 
     __bool__ = __nonzero__
 
     @property
     def readonly(self):
         """True if marker creation is disabled
 
         Remove me in the future when obsolete marker is always on."""
         return self._readonly
 
     def create(self, transaction, prec, succs=(), flag=0, parents=None,
                date=None, metadata=None):
         """obsolete: add a new obsolete marker
 
         * ensuring it is hashable
         * check mandatory metadata
         * encode metadata
 
         If you are a human writing code creating marker you want to use the
         `createmarkers` function in this module instead.
 
         return True if a new marker have been added, False if the markers
         already existed (no op).
         """
         if metadata is None:
             metadata = {}
         if date is None:
             if 'date' in metadata:
                 # as a courtesy for out-of-tree extensions
                 date = util.parsedate(metadata.pop('date'))
             else:
                 date = util.makedate()
         if len(prec) != 20:
             raise ValueError(prec)
         for succ in succs:
             if len(succ) != 20:
                 raise ValueError(succ)
         if prec in succs:
             raise ValueError(_('in-marker cycle with %s') % node.hex(prec))
 
         metadata = tuple(sorted(metadata.iteritems()))
 
         marker = (str(prec), tuple(succs), int(flag), metadata, date, parents)
         return bool(self.add(transaction, [marker]))
 
     def add(self, transaction, markers):
         """Add new markers to the store
 
         Take care of filtering duplicate.
         Return the number of new marker."""
         if self._readonly:
             raise error.Abort(_('creating obsolete markers is not enabled on '
                               'this repo'))
         known = set(self._all)
         new = []
         for m in markers:
             if m not in known:
                 known.add(m)
                 new.append(m)
         if new:
             f = self.svfs('obsstore', 'ab')
             try:
                 offset = f.tell()
                 transaction.add('obsstore', offset)
                 # offset == 0: new file - add the version header
                 for bytes in encodemarkers(new, offset == 0, self._version):
                     f.write(bytes)
             finally:
                 # XXX: f.close() == filecache invalidation == obsstore rebuilt.
                 # call 'filecacheentry.refresh()'  here
                 f.close()
             self._addmarkers(new)
             # new marker *may* have changed several set. invalidate the cache.
             self.caches.clear()
         # records the number of new markers for the transaction hooks
         previous = int(transaction.hookargs.get('new_obsmarkers', '0'))
         transaction.hookargs['new_obsmarkers'] = str(previous + len(new))
         return len(new)
 
     def mergemarkers(self, transaction, data):
         """merge a binary stream of markers inside the obsstore
 
         Returns the number of new markers added."""
         version, markers = _readmarkers(data)
         return self.add(transaction, markers)
 
     @propertycache
     def _all(self):
         data = self.svfs.tryread('obsstore')
         if not data:
             return []
         self._version, markers = _readmarkers(data)
         markers = list(markers)
         _checkinvalidmarkers(markers)
         return markers
 
     @propertycache
     def successors(self):
         successors = {}
         _addsuccessors(successors, self._all)
         return successors
 
     @propertycache
     def precursors(self):
         precursors = {}
         _addprecursors(precursors, self._all)
         return precursors
 
     @propertycache
     def children(self):
         children = {}
         _addchildren(children, self._all)
         return children
 
     def _cached(self, attr):
         return attr in self.__dict__
 
     def _addmarkers(self, markers):
         markers = list(markers) # to allow repeated iteration
         self._all.extend(markers)
         if self._cached('successors'):
             _addsuccessors(self.successors, markers)
         if self._cached('precursors'):
             _addprecursors(self.precursors, markers)
         if self._cached('children'):
             _addchildren(self.children, markers)
         _checkinvalidmarkers(markers)
 
     def relevantmarkers(self, nodes):
         """return a set of all obsolescence markers relevant to a set of nodes.
 
         "relevant" to a set of nodes mean:
 
         - marker that use this changeset as successor
         - prune marker of direct children on this changeset
         - recursive application of the two rules on precursors of these markers
 
         It is a set so you cannot rely on order."""
 
         pendingnodes = set(nodes)
         seenmarkers = set()
         seennodes = set(pendingnodes)
         precursorsmarkers = self.precursors
         children = self.children
         while pendingnodes:
             direct = set()
             for current in pendingnodes:
                 direct.update(precursorsmarkers.get(current, ()))
                 pruned = [m for m in children.get(current, ()) if not m[1]]
                 direct.update(pruned)
             direct -= seenmarkers
             pendingnodes = set([m[0] for m in direct])
             seenmarkers |= direct
             pendingnodes -= seennodes
             seennodes |= pendingnodes
         return seenmarkers
 
 def commonversion(versions):
     """Return the newest version listed in both versions and our local formats.
 
     Returns None if no common version exists.
     """
     versions.sort(reverse=True)
     # search for highest version known on both side
     for v in versions:
         if v in formats:
             return v
     return None
 
 # arbitrary picked to fit into 8K limit from HTTP server
 # you have to take in account:
 # - the version header
 # - the base85 encoding
 _maxpayload = 5300
 
 def _pushkeyescape(markers):
     """encode markers into a dict suitable for pushkey exchange
 
     - binary data is base85 encoded
     - split in chunks smaller than 5300 bytes"""
     keys = {}
     parts = []
     currentlen = _maxpayload * 2  # ensure we create a new part
     for marker in markers:
         nextdata = _fm0encodeonemarker(marker)
         if (len(nextdata) + currentlen > _maxpayload):
             currentpart = []
             currentlen = 0
             parts.append(currentpart)
         currentpart.append(nextdata)
         currentlen += len(nextdata)
     for idx, part in enumerate(reversed(parts)):
         data = ''.join([_pack('>B', _fm0version)] + part)
         keys['dump%i' % idx] = util.b85encode(data)
     return keys
 
 def listmarkers(repo):
     """List markers over pushkey"""
     if not repo.obsstore:
         return {}
     return _pushkeyescape(sorted(repo.obsstore))
 
 def pushmarker(repo, key, old, new):
     """Push markers over pushkey"""
     if not key.startswith('dump'):
         repo.ui.warn(_('unknown key: %r') % key)
         return 0
     if old:
         repo.ui.warn(_('unexpected old value for %r') % key)
         return 0
     data = util.b85decode(new)
     lock = repo.lock()
     try:
         tr = repo.transaction('pushkey: obsolete markers')
         try:
             repo.obsstore.mergemarkers(tr, data)
+            repo.invalidatevolatilesets()
             tr.close()
             return 1
         finally:
             tr.release()
     finally:
         lock.release()
 
 def getmarkers(repo, nodes=None):
     """returns markers known in a repository
 
     If <nodes> is specified, only markers "relevant" to those nodes are are
     returned"""
     if nodes is None:
         rawmarkers = repo.obsstore
     else:
         rawmarkers = repo.obsstore.relevantmarkers(nodes)
 
     for markerdata in rawmarkers:
         yield marker(repo, markerdata)
 
 def relevantmarkers(repo, node):
     """all obsolete markers relevant to some revision"""
     for markerdata in repo.obsstore.relevantmarkers(node):
         yield marker(repo, markerdata)
 
 
 def precursormarkers(ctx):
     """obsolete marker marking this changeset as a successors"""
     for data in ctx.repo().obsstore.precursors.get(ctx.node(), ()):
         yield marker(ctx.repo(), data)
 
 def successormarkers(ctx):
     """obsolete marker making this changeset obsolete"""
     for data in ctx.repo().obsstore.successors.get(ctx.node(), ()):
         yield marker(ctx.repo(), data)
 
 def allsuccessors(obsstore, nodes, ignoreflags=0):
     """Yield node for every successor of <nodes>.
 
     Some successors may be unknown locally.
 
     This is a linear yield unsuited to detecting split changesets. It includes
     initial nodes too."""
     remaining = set(nodes)
     seen = set(remaining)
     while remaining:
         current = remaining.pop()
         yield current
         for mark in obsstore.successors.get(current, ()):
             # ignore marker flagged with specified flag
             if mark[2] & ignoreflags:
                 continue
             for suc in mark[1]:
                 if suc not in seen:
                     seen.add(suc)
                     remaining.add(suc)
 
 def allprecursors(obsstore, nodes, ignoreflags=0):
     """Yield node for every precursors of <nodes>.
 
     Some precursors may be unknown locally.
 
     This is a linear yield unsuited to detecting folded changesets. It includes
     initial nodes too."""
 
     remaining = set(nodes)
     seen = set(remaining)
     while remaining:
         current = remaining.pop()
         yield current
         for mark in obsstore.precursors.get(current, ()):
             # ignore marker flagged with specified flag
             if mark[2] & ignoreflags:
                 continue
             suc = mark[0]
             if suc not in seen:
                 seen.add(suc)
                 remaining.add(suc)
 
 def foreground(repo, nodes):
     """return all nodes in the "foreground" of other node
 
     The foreground of a revision is anything reachable using parent -> children
     or precursor -> successor relation. It is very similar to "descendant" but
     augmented with obsolescence information.
 
     Beware that possible obsolescence cycle may result if complex situation.
     """
     repo = repo.unfiltered()
     foreground = set(repo.set('%ln::', nodes))
     if repo.obsstore:
         # We only need this complicated logic if there is obsolescence
         # XXX will probably deserve an optimised revset.
         nm = repo.changelog.nodemap
         plen = -1
         # compute the whole set of successors or descendants
         while len(foreground) != plen:
             plen = len(foreground)
             succs = set(c.node() for c in foreground)
             mutable = [c.node() for c in foreground if c.mutable()]
             succs.update(allsuccessors(repo.obsstore, mutable))
             known = (n for n in succs if n in nm)
             foreground = set(repo.set('%ln::', known))
     return set(c.node() for c in foreground)
 
 
 def successorssets(repo, initialnode, cache=None):
     """Return set of all latest successors of initial nodes
 
     The successors set of a changeset A are the group of revisions that succeed
     A. It succeeds A as a consistent whole, each revision being only a partial
     replacement. The successors set contains non-obsolete changesets only.
 
     This function returns the full list of successor sets which is why it
     returns a list of tuples and not just a single tuple. Each tuple is a valid
     successors set. Note that (A,) may be a valid successors set for changeset A
     (see below).
 
     In most cases, a changeset A will have a single element (e.g. the changeset
     A is replaced by A') in its successors set. Though, it is also common for a
     changeset A to have no elements in its successor set (e.g. the changeset
     has been pruned). Therefore, the returned list of successors sets will be
     [(A',)] or [], respectively.
 
     When a changeset A is split into A' and B', however, it will result in a
     successors set containing more than a single element, i.e. [(A',B')].
     Divergent changesets will result in multiple successors sets, i.e. [(A',),
     (A'')].
 
     If a changeset A is not obsolete, then it will conceptually have no
     successors set. To distinguish this from a pruned changeset, the successor
     set will contain itself only, i.e. [(A,)].
 
     Finally, successors unknown locally are considered to be pruned (obsoleted
     without any successors).
 
     The optional `cache` parameter is a dictionary that may contain precomputed
     successors sets. It is meant to reuse the computation of a previous call to
     `successorssets` when multiple calls are made at the same time. The cache
     dictionary is updated in place. The caller is responsible for its life
     span. Code that makes multiple calls to `successorssets` *must* use this
     cache mechanism or suffer terrible performance.
     """
 
     succmarkers = repo.obsstore.successors
 
     # Stack of nodes we search successors sets for
     toproceed = [initialnode]
     # set version of above list for fast loop detection
     # element added to "toproceed" must be added here
     stackedset = set(toproceed)
     if cache is None:
         cache = {}
 
     # This while loop is the flattened version of a recursive search for
     # successors sets
     #
     # def successorssets(x):
     #    successors = directsuccessors(x)
     #    ss = [[]]
     #    for succ in directsuccessors(x):
     #        # product as in itertools cartesian product
     #        ss = product(ss, successorssets(succ))
     #    return ss
     #
     # But we can not use plain recursive calls here:
     # - that would blow the python call stack
     # - obsolescence markers may have cycles, we need to handle them.
     #
     # The `toproceed` list act as our call stack. Every node we search
     # successors set for are stacked there.
     #
     # The `stackedset` is set version of this stack used to check if a node is
     # already stacked. This check is used to detect cycles and prevent infinite
     # loop.
     #
     # successors set of all nodes are stored in the `cache` dictionary.
     #
     # After this while loop ends we use the cache to return the successors sets
     # for the node requested by the caller.
     while toproceed:
         # Every iteration tries to compute the successors sets of the topmost
         # node of the stack: CURRENT.
         #
         # There are four possible outcomes:
         #
         # 1) We already know the successors sets of CURRENT:
         #    -> mission accomplished, pop it from the stack.
         # 2) Node is not obsolete:
         #    -> the node is its own successors sets. Add it to the cache.
         # 3) We do not know successors set of direct successors of CURRENT:
         #    -> We add those successors to the stack.
         # 4) We know successors sets of all direct successors of CURRENT:
         #    -> We can compute CURRENT successors set and add it to the
         #       cache.
         #
         current = toproceed[-1]
         if current in cache:
             # case (1): We already know the successors sets
             stackedset.remove(toproceed.pop())
         elif current not in succmarkers:
             # case (2): The node is not obsolete.
             if current in repo:
                 # We have a valid last successors.
                 cache[current] = [(current,)]
             else:
                 # Final obsolete version is unknown locally.
                 # Do not count that as a valid successors
                 cache[current] = []
         else:
             # cases (3) and (4)
             #
             # We proceed in two phases. Phase 1 aims to distinguish case (3)
             # from case (4):
             #
             #     For each direct successors of CURRENT, we check whether its
             #     successors sets are known. If they are not, we stack the
             #     unknown node and proceed to the next iteration of the while
             #     loop. (case 3)
             #
             #     During this step, we may detect obsolescence cycles: a node
             #     with unknown successors sets but already in the call stack.
             #     In such a situation, we arbitrary set the successors sets of
             #     the node to nothing (node pruned) to break the cycle.
             #
             #     If no break was encountered we proceed to phase 2.
             #
             # Phase 2 computes successors sets of CURRENT (case 4); see details
             # in phase 2 itself.
             #
             # Note the two levels of iteration in each phase.
             # - The first one handles obsolescence markers using CURRENT as
             #   precursor (successors markers of CURRENT).
             #
             #   Having multiple entry here means divergence.
             #
             # - The second one handles successors defined in each marker.
             #
             #   Having none means pruned node, multiple successors means split,
             #   single successors are standard replacement.
             #
             for mark in sorted(succmarkers[current]):
                 for suc in mark[1]:
                     if suc not in cache:
                         if suc in stackedset:
                             # cycle breaking
                             cache[suc] = []
                         else:
                             # case (3) If we have not computed successors sets
                             # of one of those successors we add it to the
                             # `toproceed` stack and stop all work for this
                             # iteration.
                             toproceed.append(suc)
                             stackedset.add(suc)
                             break
                 else:
                     continue
                 break
             else:
                 # case (4): we know all successors sets of all direct
                 # successors
                 #
                 # Successors set contributed by each marker depends on the
                 # successors sets of all its "successors" node.
                 #
                 # Each different marker is a divergence in the obsolescence
                 # history. It contributes successors sets distinct from other
                 # markers.
                 #
                 # Within a marker, a successor may have divergent successors
                 # sets. In such a case, the marker will contribute multiple
                 # divergent successors sets. If multiple successors have
                 # divergent successors sets, a Cartesian product is used.
                 #
                 # At the end we post-process successors sets to remove
                 # duplicated entry and successors set that are strict subset of
                 # another one.
                 succssets = []
                 for mark in sorted(succmarkers[current]):
                     # successors sets contributed by this marker
                     markss = [[]]
                     for suc in mark[1]:
                         # cardinal product with previous successors
                         productresult = []
                         for prefix in markss:
                             for suffix in cache[suc]:
                                 newss = list(prefix)
                                 for part in suffix:
                                     # do not duplicated entry in successors set
                                     # first entry wins.
                                     if part not in newss:
                                         newss.append(part)
                                 productresult.append(newss)
                         markss = productresult
                     succssets.extend(markss)
                 # remove duplicated and subset
                 seen = []
                 final = []
                 candidate = sorted(((set(s), s) for s in succssets if s),
                                    key=lambda x: len(x[1]), reverse=True)
                 for setversion, listversion in candidate:
                     for seenset in seen:
                         if setversion.issubset(seenset):
                             break
                     else:
                         final.append(listversion)
                         seen.append(setversion)
                 final.reverse() # put small successors set first
                 cache[current] = final
     return cache[initialnode]
 
 # mapping of 'set-name' -> <function to compute this set>
 cachefuncs = {}
 def cachefor(name):
     """Decorator to register a function as computing the cache for a set"""
     def decorator(func):
         assert name not in cachefuncs
         cachefuncs[name] = func
         return func
     return decorator
 
 def getrevs(repo, name):
     """Return the set of revision that belong to the <name> set
 
     Such access may compute the set and cache it for future use"""
     repo = repo.unfiltered()
     if not repo.obsstore:
         return frozenset()
     if name not in repo.obsstore.caches:
         repo.obsstore.caches[name] = cachefuncs[name](repo)
     return repo.obsstore.caches[name]
 
 # To be simple we need to invalidate obsolescence cache when:
 #
 # - new changeset is added:
 # - public phase is changed
 # - obsolescence marker are added
 # - strip is used a repo
 def clearobscaches(repo):
     """Remove all obsolescence related cache from a repo
 
     This remove all cache in obsstore is the obsstore already exist on the
     repo.
 
     (We could be smarter here given the exact event that trigger the cache
     clearing)"""
     # only clear cache is there is obsstore data in this repo
     if 'obsstore' in repo._filecache:
         repo.obsstore.caches.clear()
 
 @cachefor('obsolete')
 def _computeobsoleteset(repo):
     """the set of obsolete revisions"""
     obs = set()
     getnode = repo.changelog.node
     notpublic = repo._phasecache.getrevset(repo, (phases.draft, phases.secret))
     for r in notpublic:
         if getnode(r) in repo.obsstore.successors:
             obs.add(r)
     return obs
 
 @cachefor('unstable')
 def _computeunstableset(repo):
     """the set of non obsolete revisions with obsolete parents"""
     revs = [(ctx.rev(), ctx) for ctx in
             repo.set('(not public()) and (not obsolete())')]
     revs.sort(key=lambda x:x[0])
     unstable = set()
     for rev, ctx in revs:
         # A rev is unstable if one of its parent is obsolete or unstable
         # this works since we traverse following growing rev order
         if any((x.obsolete() or (x.rev() in unstable))
                 for x in ctx.parents()):
             unstable.add(rev)
     return unstable
 
 @cachefor('suspended')
 def _computesuspendedset(repo):
     """the set of obsolete parents with non obsolete descendants"""
     suspended = repo.changelog.ancestors(getrevs(repo, 'unstable'))
     return set(r for r in getrevs(repo, 'obsolete') if r in suspended)
 
 @cachefor('extinct')
 def _computeextinctset(repo):
     """the set of obsolete parents without non obsolete descendants"""
     return getrevs(repo, 'obsolete') - getrevs(repo, 'suspended')
 
 
 @cachefor('bumped')
 def _computebumpedset(repo):
     """the set of revs trying to obsolete public revisions"""
     bumped = set()
     # util function (avoid attribute lookup in the loop)
     phase = repo._phasecache.phase # would be faster to grab the full list
     public = phases.public
     cl = repo.changelog
     torev = cl.nodemap.get
     for ctx in repo.set('(not public()) and (not obsolete())'):
         rev = ctx.rev()
         # We only evaluate mutable, non-obsolete revision
         node = ctx.node()
         # (future) A cache of precursors may worth if split is very common
         for pnode in allprecursors(repo.obsstore, [node],
                                    ignoreflags=bumpedfix):
             prev = torev(pnode) # unfiltered! but so is phasecache
             if (prev is not None) and (phase(repo, prev) <= public):
                 # we have a public precursor
                 bumped.add(rev)
                 break # Next draft!
     return bumped
 
 @cachefor('divergent')
 def _computedivergentset(repo):
     """the set of rev that compete to be the final successors of some revision.
     """
     divergent = set()
     obsstore = repo.obsstore
     newermap = {}
     for ctx in repo.set('(not public()) - obsolete()'):
         mark = obsstore.precursors.get(ctx.node(), ())
         toprocess = set(mark)
         seen = set()
         while toprocess:
             prec = toprocess.pop()[0]
             if prec in seen:
                 continue # emergency cycle hanging prevention
             seen.add(prec)
             if prec not in newermap:
                 successorssets(repo, prec, newermap)
             newer = [n for n in newermap[prec] if n]
             if len(newer) > 1:
                 divergent.add(ctx.rev())
                 break
             toprocess.update(obsstore.precursors.get(prec, ()))
     return divergent
 
 
 def createmarkers(repo, relations, flag=0, date=None, metadata=None,
                   operation=None):
     """Add obsolete markers between changesets in a repo
 
     <relations> must be an iterable of (<old>, (<new>, ...)[,{metadata}])
     tuple. `old` and `news` are changectx. metadata is an optional dictionary
     containing metadata for this marker only. It is merged with the global
     metadata specified through the `metadata` argument of this function,
 
     Trying to obsolete a public changeset will raise an exception.
 
     Current user and date are used except if specified otherwise in the
     metadata attribute.
 
     This function operates within a transaction of its own, but does
     not take any lock on the repo.
     """
     # prepare metadata
     if metadata is None:
         metadata = {}
     if 'user' not in metadata:
         metadata['user'] = repo.ui.username()
     useoperation = repo.ui.configbool('experimental',
                                       'evolution.track-operation',
                                       False)
     if useoperation and operation:
         metadata['operation'] = operation
     tr = repo.transaction('add-obsolescence-marker')
     try:
         markerargs = []
         for rel in relations:
             prec = rel[0]
             sucs = rel[1]
             localmetadata = metadata.copy()
             if 2 < len(rel):
                 localmetadata.update(rel[2])
 
             if not prec.mutable():
                 raise error.Abort(_("cannot obsolete public changeset: %s")
                                  % prec,
                                  hint="see 'hg help phases' for details")
             nprec = prec.node()
             nsucs = tuple(s.node() for s in sucs)
             npare = None
             if not nsucs:
                 npare = tuple(p.node() for p in prec.parents())
             if nprec in nsucs:
                 raise error.Abort(_("changeset %s cannot obsolete itself")
                                   % prec)
 
             # Creating the marker causes the hidden cache to become invalid,
             # which causes recomputation when we ask for prec.parents() above.
             # Resulting in n^2 behavior.  So let's prepare all of the args
             # first, then create the markers.
             markerargs.append((nprec, nsucs, npare, localmetadata))
 
         for args in markerargs:
             nprec, nsucs, npare, localmetadata = args
             repo.obsstore.create(tr, nprec, nsucs, flag, parents=npare,
                                  date=date, metadata=localmetadata)
             repo.filteredrevcache.clear()
         tr.close()
     finally:
         tr.release()