# Copyright 2016-present Facebook. All Rights Reserved.

#

# context: context needed to annotate a file

#

# This software may be used and distributed according to the terms of the

# GNU General Public License version 2 or any later version.

from __future__ import absolute_import

import collections

import contextlib

import hashlib

import os

from mercurial.i18n import _

from mercurial import (

    error,

    linelog as linelogmod,

    lock as lockmod,

    mdiff,

    node,

    pycompat,

    scmutil,

    util,

)

from mercurial.utils import (

    stringutil,

)

from . import (

    error as faerror,

    revmap as revmapmod,

)

# given path, get filelog, cached

@util.lrucachefunc

def _getflog(repo, path):

    return repo.file(path)

# extracted from mercurial.context.basefilectx.annotate

def _parents(f, follow=True):

    # Cut _descendantrev here to mitigate the penalty of lazy linkrev

    # adjustment. Otherwise, p._adjustlinkrev() would walk changelog

    # from the topmost introrev (= srcrev) down to p.linkrev() if it

    # isn't an ancestor of the srcrev.

    f._changeid

    pl = f.parents()

    # Don't return renamed parents if we aren't following.

    if not follow:

        pl = [p for p in pl if p.path() == f.path()]

    # renamed filectx won't have a filelog yet, so set it

    # from the cache to save time

    for p in pl:

        if not '_filelog' in p.__dict__:

            p._filelog = _getflog(f._repo, p.path())

    return pl

# extracted from mercurial.context.basefilectx.annotate. slightly modified

# so it takes a fctx instead of a pair of text and fctx.

def _decorate(fctx):

    text = fctx.data()

    linecount = text.count('\n')

    if text and not text.endswith('\n'):

        linecount += 1

    return ([(fctx, i) for i in pycompat.xrange(linecount)], text)

# extracted from mercurial.context.basefilectx.annotate. slightly modified

# so it takes an extra "blocks" parameter calculated elsewhere, instead of

# calculating diff here.

def _pair(parent, child, blocks):

    for (a1, a2, b1, b2), t in blocks:

        # Changed blocks ('!') or blocks made only of blank lines ('~')

        # belong to the child.

        if t == '=':

            child[0][b1:b2] = parent[0][a1:a2]

    return child

# like scmutil.revsingle, but with lru cache, so their states (like manifests)

# could be reused

_revsingle = util.lrucachefunc(scmutil.revsingle)

def resolvefctx(repo, rev, path, resolverev=False, adjustctx=None):

    """(repo, str, str) -> fctx

    get the filectx object from repo, rev, path, in an efficient way.

    if resolverev is True, "rev" is a revision specified by the revset

    language, otherwise "rev" is a nodeid, or a revision number that can

    be consumed by repo.__getitem__.

    if adjustctx is not None, the returned fctx will point to a changeset

    that introduces the change (last modified the file). if adjustctx

    is 'linkrev', trust the linkrev and do not adjust it. this is noticeably

    faster for big repos but is incorrect for some cases.

    """

    if resolverev and not isinstance(rev, int) and rev is not None:

        ctx = _revsingle(repo, rev)

    else:

        ctx = repo[rev]

    # If we don't need to adjust the linkrev, create the filectx using the

    # changectx instead of using ctx[path]. This means it already has the

    # changectx information, so blame -u will be able to look directly at the

    # commitctx object instead of having to resolve it by going through the

    # manifest. In a lazy-manifest world this can prevent us from downloading a

    # lot of data.

    if adjustctx is None:

        # ctx.rev() is None means it's the working copy, which is a special

        # case.

        if ctx.rev() is None:

            fctx = ctx[path]

        else:

            fctx = repo.filectx(path, changeid=ctx.rev())

    else:

        fctx = ctx[path]

        if adjustctx == 'linkrev':

            introrev = fctx.linkrev()

        else:

            introrev = fctx.introrev()

        if introrev != ctx.rev():

            fctx._changeid = introrev

            fctx._changectx = repo[introrev]

    return fctx

# like mercurial.store.encodedir, but use linelog suffixes: .m, .l, .lock

def encodedir(path):

    return (path

            .replace('.hg/', '.hg.hg/')

            .replace('.l/', '.l.hg/')

            .replace('.m/', '.m.hg/')

            .replace('.lock/', '.lock.hg/'))

def hashdiffopts(diffopts):

    diffoptstr = stringutil.pprint(sorted(

        (k, getattr(diffopts, k))

        for k in mdiff.diffopts.defaults

    ))

    return node.hex(hashlib.sha1(diffoptstr).digest())[:6]

_defaultdiffopthash = hashdiffopts(mdiff.defaultopts)

class annotateopts(object):

    """like mercurial.mdiff.diffopts, but is for annotate

    followrename: follow renames, like "hg annotate -f"

    followmerge: follow p2 of a merge changeset, otherwise p2 is ignored

    """

    defaults = {

        'diffopts': None,

        'followrename': True,

        'followmerge': True,

    }

    def __init__(self, **opts):

        opts = pycompat.byteskwargs(opts)

        for k, v in self.defaults.iteritems():

            setattr(self, k, opts.get(k, v))

    @util.propertycache

    def shortstr(self):

        """represent opts in a short string, suitable for a directory name"""

        result = ''

        if not self.followrename:

            result += 'r0'

        if not self.followmerge:

            result += 'm0'

        if self.diffopts is not None:

            assert isinstance(self.diffopts, mdiff.diffopts)

            diffopthash = hashdiffopts(self.diffopts)

            if diffopthash != _defaultdiffopthash:

                result += 'i' + diffopthash

        return result or 'default'

defaultopts = annotateopts()

class _annotatecontext(object):

    """do not use this class directly as it does not use lock to protect

    writes. use "with annotatecontext(...)" instead.

    """

    def __init__(self, repo, path, linelogpath, revmappath, opts):

        self.repo = repo

        self.ui = repo.ui

        self.path = path

        self.opts = opts

        self.linelogpath = linelogpath

        self.revmappath = revmappath

        self._linelog = None

        self._revmap = None

        self._node2path = {} # {str: str}

    @property

    def linelog(self):

        if self._linelog is None:

            if os.path.exists(self.linelogpath):

                with open(self.linelogpath, 'rb') as f:

                    try:

                        self._linelog = linelogmod.linelog.fromdata(f.read())

                    except linelogmod.LineLogError:

                        self._linelog = linelogmod.linelog()

            else:

                self._linelog = linelogmod.linelog()

        return self._linelog

    @property

    def revmap(self):

        if self._revmap is None:

            self._revmap = revmapmod.revmap(self.revmappath)

        return self._revmap

    def close(self):

        if self._revmap is not None:

            self._revmap.flush()

            self._revmap = None

        if self._linelog is not None:

            with open(self.linelogpath, 'wb') as f:

                f.write(self._linelog.encode())

            self._linelog = None

    __del__ = close

    def rebuild(self):

        """delete linelog and revmap, useful for rebuilding"""

        self.close()

        self._node2path.clear()

        _unlinkpaths([self.revmappath, self.linelogpath])

    @property

    def lastnode(self):

        """return last node in revmap, or None if revmap is empty"""

        if self._revmap is None:

            # fast path, read revmap without loading its full content

            return revmapmod.getlastnode(self.revmappath)

        else:

            return self._revmap.rev2hsh(self._revmap.maxrev)

    def isuptodate(self, master, strict=True):

        """return True if the revmap / linelog is up-to-date, or the file

        does not exist in the master revision. False otherwise.

        it tries to be fast and could return false negatives, because of the

        use of linkrev instead of introrev.

        useful for both server and client to decide whether to update

        fastannotate cache or not.

        if strict is True, even if fctx exists in the revmap, but is not the

        last node, isuptodate will return False. it's good for performance - no

        expensive check was done.

        if strict is False, if fctx exists in the revmap, this function may

        return True. this is useful for the client to skip downloading the

        cache if the client's master is behind the server's.

        """

        lastnode = self.lastnode

        try:

            f = self._resolvefctx(master, resolverev=True)

            # choose linkrev instead of introrev as the check is meant to be

            # *fast*.

            linknode = self.repo.changelog.node(f.linkrev())

            if not strict and lastnode and linknode != lastnode:

                # check if f.node() is in the revmap. note: this loads the

                # revmap and can be slow.

                return self.revmap.hsh2rev(linknode) is not None

            # avoid resolving old manifest, or slow adjustlinkrev to be fast,

            # false negatives are acceptable in this case.

            return linknode == lastnode

        except LookupError:

            # master does not have the file, or the revmap is ahead

            return True

    def annotate(self, rev, master=None, showpath=False, showlines=False):

        """incrementally update the cache so it includes revisions in the main

        branch till 'master'. and run annotate on 'rev', which may or may not be

        included in the main branch.

        if master is None, do not update linelog.

        the first value returned is the annotate result, it is [(node, linenum)]

        by default. [(node, linenum, path)] if showpath is True.

        if showlines is True, a second value will be returned, it is a list of

        corresponding line contents.

        """

        # the fast path test requires commit hash, convert rev number to hash,

        # so it may hit the fast path. note: in the "fctx" mode, the "annotate"

        # command could give us a revision number even if the user passes a

        # commit hash.

        if isinstance(rev, int):

            rev = node.hex(self.repo.changelog.node(rev))

        # fast path: if rev is in the main branch already

        directly, revfctx = self.canannotatedirectly(rev)

        if directly:

            if self.ui.debugflag:

                self.ui.debug('fastannotate: %s: using fast path '

                              '(resolved fctx: %s)\n'

                              % (self.path,

                                 stringutil.pprint(util.safehasattr(revfctx,

                                                                    'node'))))

            return self.annotatedirectly(revfctx, showpath, showlines)

        # resolve master

        masterfctx = None

        if master:

            try:

                masterfctx = self._resolvefctx(master, resolverev=True,

                                               adjustctx=True)

            except LookupError: # master does not have the file

                pass

            else:

                if masterfctx in self.revmap: # no need to update linelog

                    masterfctx = None

        #                  ... - @ <- rev (can be an arbitrary changeset,

        #                 /                not necessarily a descendant

        #      master -> o                 of master)

        #                |

        #     a merge -> o         'o': new changesets in the main branch

        #                |\        '#': revisions in the main branch that

        #                o *            exist in linelog / revmap

        #                | .       '*': changesets in side branches, or

        # last master -> # .            descendants of master

        #                | .

        #                # *       joint: '#', and is a parent of a '*'

        #                |/

        #     a joint -> # ^^^^ --- side branches

        #                |

        #                ^ --- main branch (in linelog)

        # these DFSes are similar to the traditional annotate algorithm.

        # we cannot really reuse the code for perf reason.

        # 1st DFS calculates merges, joint points, and needed.

        # "needed" is a simple reference counting dict to free items in

        # "hist", reducing its memory usage otherwise could be huge.

        initvisit = [revfctx]

        if masterfctx:

            if masterfctx.rev() is None:

                raise error.Abort(_('cannot update linelog to wdir()'),

                                  hint=_('set fastannotate.mainbranch'))

            initvisit.append(masterfctx)

        visit = initvisit[:]

        pcache = {}

        needed = {revfctx: 1}

        hist = {} # {fctx: ([(llrev or fctx, linenum)], text)}

        while visit:

            f = visit.pop()

            if f in pcache or f in hist:

                continue

            if f in self.revmap: # in the old main branch, it's a joint

                llrev = self.revmap.hsh2rev(f.node())

                self.linelog.annotate(llrev)

                result = self.linelog.annotateresult

                hist[f] = (result, f.data())

                continue

            pl = self._parentfunc(f)

            pcache[f] = pl

            for p in pl:

                needed[p] = needed.get(p, 0) + 1

                if p not in pcache:

                    visit.append(p)

        # 2nd (simple) DFS calculates new changesets in the main branch

        # ('o' nodes in # the above graph), so we know when to update linelog.

        newmainbranch = set()

        f = masterfctx

        while f and f not in self.revmap:

            newmainbranch.add(f)

            pl = pcache[f]

            if pl:

                f = pl[0]

            else:

                f = None

                break

        # f, if present, is the position where the last build stopped at, and

        # should be the "master" last time. check to see if we can continue

        # building the linelog incrementally. (we cannot if diverged)

        if masterfctx is not None:

            self._checklastmasterhead(f)

        if self.ui.debugflag:

            if newmainbranch:

                self.ui.debug('fastannotate: %s: %d new changesets in the main'

                              ' branch\n' % (self.path, len(newmainbranch)))

            elif not hist: # no joints, no updates

                self.ui.debug('fastannotate: %s: linelog cannot help in '

                              'annotating this revision\n' % self.path)

        # prepare annotateresult so we can update linelog incrementally

        self.linelog.annotate(self.linelog.maxrev)

        # 3rd DFS does the actual annotate

        visit = initvisit[:]

        progress = self.ui.makeprogress(('building cache'),

                                        total=len(newmainbranch))

        while visit:

            f = visit[-1]

            if f in hist:

                visit.pop()

                continue

            ready = True

            pl = pcache[f]

            for p in pl:

                if p not in hist:

                    ready = False

                    visit.append(p)

            if not ready:

                continue

            visit.pop()

            blocks = None # mdiff blocks, used for appending linelog

            ismainbranch = (f in newmainbranch)

            # curr is the same as the traditional annotate algorithm,

            # if we only care about linear history (do not follow merge),

            # then curr is not actually used.

            assert f not in hist

            curr = _decorate(f)

            for i, p in enumerate(pl):

                bs = list(self._diffblocks(hist[p][1], curr[1]))

                if i == 0 and ismainbranch:

                    blocks = bs

                curr = _pair(hist[p], curr, bs)

                if needed[p] == 1:

                    del hist[p]

                    del needed[p]

                else:

                    needed[p] -= 1

            hist[f] = curr

            del pcache[f]

            if ismainbranch: # need to write to linelog

                progress.increment()

                bannotated = None

                if len(pl) == 2 and self.opts.followmerge: # merge

                    bannotated = curr[0]

                if blocks is None: # no parents, add an empty one

                    blocks = list(self._diffblocks('', curr[1]))

                self._appendrev(f, blocks, bannotated)

            elif showpath: # not append linelog, but we need to record path

                self._node2path[f.node()] = f.path()

        progress.complete()

        result = [

            ((self.revmap.rev2hsh(fr) if isinstance(fr, int) else fr.node()), l)

            for fr, l in hist[revfctx][0]] # [(node, linenumber)]

        return self._refineannotateresult(result, revfctx, showpath, showlines)

    def canannotatedirectly(self, rev):

        """(str) -> bool, fctx or node.

        return (True, f) if we can annotate without updating the linelog, pass

        f to annotatedirectly.

        return (False, f) if we need extra calculation. f is the fctx resolved

        from rev.

        """

        result = True

        f = None

        if not isinstance(rev, int) and rev is not None:

            hsh = {20: bytes, 40: node.bin}.get(len(rev), lambda x: None)(rev)

            if hsh is not None and (hsh, self.path) in self.revmap:

                f = hsh

        if f is None:

            adjustctx = 'linkrev' if self._perfhack else True

            f = self._resolvefctx(rev, adjustctx=adjustctx, resolverev=True)

            result = f in self.revmap

            if not result and self._perfhack:

                # redo the resolution without perfhack - as we are going to

                # do write operations, we need a correct fctx.

                f = self._resolvefctx(rev, adjustctx=True, resolverev=True)

        return result, f

    def annotatealllines(self, rev, showpath=False, showlines=False):

        """(rev : str) -> [(node : str, linenum : int, path : str)]

        the result has the same format with annotate, but include all (including

        deleted) lines up to rev. call this after calling annotate(rev, ...) for

        better performance and accuracy.

        """

        revfctx = self._resolvefctx(rev, resolverev=True, adjustctx=True)

        # find a chain from rev to anything in the mainbranch

        if revfctx not in self.revmap:

            chain = [revfctx]

            a = ''

            while True:

                f = chain[-1]

                pl = self._parentfunc(f)

                if not pl:

                    break

                if pl[0] in self.revmap:

                    a = pl[0].data()

                    break

                chain.append(pl[0])

            # both self.linelog and self.revmap is backed by filesystem. now

            # we want to modify them but do not want to write changes back to

            # files. so we create in-memory objects and copy them. it's like

            # a "fork".

            linelog = linelogmod.linelog()

            linelog.copyfrom(self.linelog)

            linelog.annotate(linelog.maxrev)

            revmap = revmapmod.revmap()

            revmap.copyfrom(self.revmap)

            for f in reversed(chain):

                b = f.data()

                blocks = list(self._diffblocks(a, b))

                self._doappendrev(linelog, revmap, f, blocks)

                a = b

        else:

            # fastpath: use existing linelog, revmap as we don't write to them

            linelog = self.linelog

            revmap = self.revmap

        lines = linelog.getalllines()

        hsh = revfctx.node()

        llrev = revmap.hsh2rev(hsh)