view tests/test-revlog-raw.py @ 39561:d06834e0f48e

wireprotov2peer: stream decoded responses

Previously, wire protocol version 2 would buffer all response data.
Only once all data was received did we CBOR decode it and resolve
the future associated with the command. This was obviously not
desirable. In future commits that introduce large response payloads,
this caused significant memory bloat and slowed down client
operations due to waiting on the server.

This commit refactors the response handling code so that response
data can be streamed.

Command response objects now contain a buffered CBOR decoder. As
new data arrives, it is fed into the decoder. Decoded objects are
made available to the generator as they are decoded.

Because there is a separate thread processing incoming frames and
feeding data into the response object, there is the potential for
race conditions when mutating response objects. So a lock has been
added to guard access to critical state variables.

Because the generator emitting decoded objects needs to wait on
those objects to become available, we've added an Event for the
generator to wait on so it doesn't busy loop. This does mean
there is the potential for deadlocks. And I'm pretty sure they can
occur in some scenarios. We already have a handful of TODOs around
this. But I've added some more. Fixing this will likely require
moving the background thread receiving frames into clienthandler.
We likely would have done this anyway when implementing the client
bits for the SSH transport.

Test output changes because the initial CBOR map holding the overall
response state is now always handled internally by the response
object.

Differential Revision: https://phab.mercurial-scm.org/D4474

# test revlog interaction about raw data (flagprocessor)

from __future__ import absolute_import, print_function

import sys

from mercurial import (
    encoding,
    node,
    revlog,
    transaction,
    vfs,
)

# TESTTMP is optional. This makes it convenient to run without run-tests.py
tvfs = vfs.vfs(encoding.environ.get(b'TESTTMP', b'/tmp'))

# Enable generaldelta otherwise revlog won't use delta as expected by the test
tvfs.options = {b'generaldelta': True, b'revlogv1': True}

# The test wants to control whether to use delta explicitly, based on
# "storedeltachains".
revlog.revlog._isgooddeltainfo = lambda self, d, textlen: self._storedeltachains

def abort(msg):
    print('abort: %s' % msg)
    # Return 0 so run-tests.py could compare the output.
    sys.exit()

# Register a revlog processor for flag EXTSTORED.
#
# It simply prepends a fixed header, and replaces '1' to 'i'. So it has
# insertion and replacement, and may be interesting to test revlog's line-based
# deltas.
_extheader = b'E\n'

def readprocessor(self, rawtext):
    # True: the returned text could be used to verify hash
    text = rawtext[len(_extheader):].replace(b'i', b'1')
    return text, True

def writeprocessor(self, text):
    # False: the returned rawtext shouldn't be used to verify hash
    rawtext = _extheader + text.replace(b'1', b'i')
    return rawtext, False

def rawprocessor(self, rawtext):
    # False: do not verify hash. Only the content returned by "readprocessor"
    # can be used to verify hash.
    return False

revlog.addflagprocessor(revlog.REVIDX_EXTSTORED,
                        (readprocessor, writeprocessor, rawprocessor))

# Utilities about reading and appending revlog

def newtransaction():
    # A transaction is required to write revlogs
    report = lambda msg: None
    return transaction.transaction(report, tvfs, {'plain': tvfs}, b'journal')

def newrevlog(name=b'_testrevlog.i', recreate=False):
    if recreate:
        tvfs.tryunlink(name)
    rlog = revlog.revlog(tvfs, name)
    return rlog

def appendrev(rlog, text, tr, isext=False, isdelta=True):
    '''Append a revision. If isext is True, set the EXTSTORED flag so flag
    processor will be used (and rawtext is different from text). If isdelta is
    True, force the revision to be a delta, otherwise it's full text.
    '''
    nextrev = len(rlog)
    p1 = rlog.node(nextrev - 1)
    p2 = node.nullid
    if isext:
        flags = revlog.REVIDX_EXTSTORED
    else:
        flags = revlog.REVIDX_DEFAULT_FLAGS
    # Change storedeltachains temporarily, to override revlog's delta decision
    rlog._storedeltachains = isdelta
    try:
        rlog.addrevision(text, tr, nextrev, p1, p2, flags=flags)
        return nextrev
    except Exception as ex:
        abort('rev %d: failed to append: %s' % (nextrev, ex))
    finally:
        # Restore storedeltachains. It is always True, see revlog.__init__
        rlog._storedeltachains = True

def addgroupcopy(rlog, tr, destname=b'_destrevlog.i', optimaldelta=True):
    '''Copy revlog to destname using revlog.addgroup. Return the copied revlog.

    This emulates push or pull. They use changegroup. Changegroup requires
    repo to work. We don't have a repo, so a dummy changegroup is used.

    If optimaldelta is True, use optimized delta parent, so the destination
    revlog could probably reuse it. Otherwise it builds sub-optimal delta, and
    the destination revlog needs more work to use it.

    This exercises some revlog.addgroup (and revlog._addrevision(text=None))
    code path, which is not covered by "appendrev" alone.
    '''
    class dummychangegroup(object):
        @staticmethod
        def deltachunk(pnode):
            pnode = pnode or node.nullid
            parentrev = rlog.rev(pnode)
            r = parentrev + 1
            if r >= len(rlog):
                return {}
            if optimaldelta:
                deltaparent = parentrev
            else:
                # suboptimal deltaparent
                deltaparent = min(0, parentrev)
            if not rlog.candelta(deltaparent, r):
                deltaparent = -1
            return {b'node': rlog.node(r), b'p1': pnode, b'p2': node.nullid,
                    b'cs': rlog.node(rlog.linkrev(r)), b'flags': rlog.flags(r),
                    b'deltabase': rlog.node(deltaparent),
                    b'delta': rlog.revdiff(deltaparent, r)}

        def deltaiter(self):
            chain = None
            for chunkdata in iter(lambda: self.deltachunk(chain), {}):
                node = chunkdata[b'node']
                p1 = chunkdata[b'p1']
                p2 = chunkdata[b'p2']
                cs = chunkdata[b'cs']
                deltabase = chunkdata[b'deltabase']
                delta = chunkdata[b'delta']
                flags = chunkdata[b'flags']

                chain = node

                yield (node, p1, p2, cs, deltabase, delta, flags)

    def linkmap(lnode):
        return rlog.rev(lnode)

    dlog = newrevlog(destname, recreate=True)
    dummydeltas = dummychangegroup().deltaiter()
    dlog.addgroup(dummydeltas, linkmap, tr)
    return dlog

def lowlevelcopy(rlog, tr, destname=b'_destrevlog.i'):
    '''Like addgroupcopy, but use the low level revlog._addrevision directly.

    It exercises some code paths that are hard to reach easily otherwise.
    '''
    dlog = newrevlog(destname, recreate=True)
    for r in rlog:
        p1 = rlog.node(r - 1)
        p2 = node.nullid
        if r == 0 or (rlog.flags(r) & revlog.REVIDX_EXTSTORED):
            text = rlog.revision(r, raw=True)
            cachedelta = None
        else:
            # deltaparent cannot have EXTSTORED flag.
            deltaparent = max([-1] +
                              [p for p in range(r)
                               if rlog.flags(p) & revlog.REVIDX_EXTSTORED == 0])
            text = None
            cachedelta = (deltaparent, rlog.revdiff(deltaparent, r))
        flags = rlog.flags(r)
        ifh = dfh = None
        try:
            ifh = dlog.opener(dlog.indexfile, b'a+')
            if not dlog._inline:
                dfh = dlog.opener(dlog.datafile, b'a+')
            dlog._addrevision(rlog.node(r), text, tr, r, p1, p2, flags,
                              cachedelta, ifh, dfh)
        finally:
            if dfh is not None:
                dfh.close()
            if ifh is not None:
                ifh.close()
    return dlog

# Utilities to generate revisions for testing

def genbits(n):
    '''Given a number n, generate (2 ** (n * 2) + 1) numbers in range(2 ** n).
    i.e. the generated numbers have a width of n bits.

    The combination of two adjacent numbers will cover all possible cases.
    That is to say, given any x, y where both x, and y are in range(2 ** n),
    there is an x followed immediately by y in the generated sequence.
    '''
    m = 2 ** n

    # Gray Code. See https://en.wikipedia.org/wiki/Gray_code
    gray = lambda x: x ^ (x >> 1)
    reversegray = dict((gray(i), i) for i in range(m))

    # Generate (n * 2) bit gray code, yield lower n bits as X, and look for
    # the next unused gray code where higher n bits equal to X.

    # For gray codes whose higher bits are X, a[X] of them have been used.
    a = [0] * m

    # Iterate from 0.
    x = 0
    yield x
    for i in range(m * m):
        x = reversegray[x]
        y = gray(a[x] + x * m) & (m - 1)
        assert a[x] < m
        a[x] += 1
        x = y
        yield x

def gentext(rev):
    '''Given a revision number, generate dummy text'''
    return b''.join(b'%d\n' % j for j in range(-1, rev % 5))

def writecases(rlog, tr):
    '''Write some revisions interested to the test.

    The test is interested in 3 properties of a revision:

        - Is it a delta or a full text? (isdelta)
          This is to catch some delta application issues.
        - Does it have a flag of EXTSTORED? (isext)
          This is to catch some flag processor issues. Especially when
          interacted with revlog deltas.
        - Is its text empty? (isempty)
          This is less important. It is intended to try to catch some careless
          checks like "if text" instead of "if text is None". Note: if flag
          processor is involved, raw text may be not empty.

    Write 65 revisions. So that all combinations of the above flags for
    adjacent revisions are covered. That is to say,

        len(set(
            (r.delta, r.ext, r.empty, (r+1).delta, (r+1).ext, (r+1).empty)
            for r in range(len(rlog) - 1)
           )) is 64.

    Where "r.delta", "r.ext", and "r.empty" are booleans matching properties
    mentioned above.

    Return expected [(text, rawtext)].
    '''
    result = []
    for i, x in enumerate(genbits(3)):
        isdelta, isext, isempty = bool(x & 1), bool(x & 2), bool(x & 4)
        if isempty:
            text = b''
        else:
            text = gentext(i)
        rev = appendrev(rlog, text, tr, isext=isext, isdelta=isdelta)

        # Verify text, rawtext, and rawsize
        if isext:
            rawtext = writeprocessor(None, text)[0]
        else:
            rawtext = text
        if rlog.rawsize(rev) != len(rawtext):
            abort('rev %d: wrong rawsize' % rev)
        if rlog.revision(rev, raw=False) != text:
            abort('rev %d: wrong text' % rev)
        if rlog.revision(rev, raw=True) != rawtext:
            abort('rev %d: wrong rawtext' % rev)
        result.append((text, rawtext))

        # Verify flags like isdelta, isext work as expected
        # isdelta can be overridden to False if this or p1 has isext set
        if bool(rlog.deltaparent(rev) > -1) and not isdelta:
            abort('rev %d: isdelta is unexpected' % rev)
        if bool(rlog.flags(rev)) != isext:
            abort('rev %d: isext is ineffective' % rev)
    return result

# Main test and checking

def checkrevlog(rlog, expected):
    '''Check if revlog has expected contents. expected is [(text, rawtext)]'''
    # Test using different access orders. This could expose some issues
    # depending on revlog caching (see revlog._cache).
    for r0 in range(len(rlog) - 1):
        r1 = r0 + 1
        for revorder in [[r0, r1], [r1, r0]]:
            for raworder in [[True], [False], [True, False], [False, True]]:
                nlog = newrevlog()
                for rev in revorder:
                    for raw in raworder:
                        t = nlog.revision(rev, raw=raw)
                        if t != expected[rev][int(raw)]:
                            abort('rev %d: corrupted %stext'
                                  % (rev, raw and 'raw' or ''))

def maintest():
    expected = rl = None
    with newtransaction() as tr:
        rl = newrevlog(recreate=True)
        expected = writecases(rl, tr)
        checkrevlog(rl, expected)
        print('local test passed')
        # Copy via revlog.addgroup
        rl1 = addgroupcopy(rl, tr)
        checkrevlog(rl1, expected)
        rl2 = addgroupcopy(rl, tr, optimaldelta=False)
        checkrevlog(rl2, expected)
        print('addgroupcopy test passed')
        # Copy via revlog.clone
        rl3 = newrevlog(name=b'_destrevlog3.i', recreate=True)
        rl.clone(tr, rl3)
        checkrevlog(rl3, expected)
        print('clone test passed')
        # Copy via low-level revlog._addrevision
        rl4 = lowlevelcopy(rl, tr)
        checkrevlog(rl4, expected)
        print('lowlevelcopy test passed')

try:
    maintest()
except Exception as ex:
    abort('crashed: %s' % ex)