import cStringIO
import errno
import re

from mercurial import patch
from mercurial import node
from mercurial import context
from mercurial import revlog

import svnwrap
import svnexternals
import util


binary_file_re = re.compile(r'''Index: ([^\n]*)
=*
Cannot display: file marked as a binary type.''')

property_exec_set_re = re.compile(r'''Property changes on: ([^\n]*)
_*
(?:Added|Name): svn:executable
   \+''')

property_exec_removed_re = re.compile(r'''Property changes on: ([^\n]*)
_*
(?:Deleted|Name): svn:executable
   -''')

empty_file_patch_wont_make_re = re.compile(r'''Index: ([^\n]*)\n=*\n(?=Index:)''')

any_file_re = re.compile(r'''^Index: ([^\n]*)\n=*\n''', re.MULTILINE)

property_special_set_re = re.compile(r'''Property changes on: ([^\n]*)
_*
(?:Added|Name): svn:special
   \+''')

property_special_removed_re = re.compile(r'''Property changes on: ([^\n]*)
_*
(?:Deleted|Name): svn:special
   \-''')


class BadPatchApply(Exception):
    pass


def print_your_svn_is_old_message(ui): #pragma: no cover
    ui.status("In light of that, I'll fall back and do diffs, but it won't do "
              "as good a job. You should really upgrade your server.\n")


def mempatchproxy(parentctx, files):
    # Avoid circular references patch.patchfile -> mempatch
    patchfile = patch.patchfile

    # TODO(durin42): make this a compat path for hg < 1.6.
    class mempatch(patchfile):
        def __init__(self, ui, fname, opener, missing=False, eolmode=None):
            patchfile.__init__(self, ui, fname, None, False, eolmode)

        def readlines(self, fname):
            if fname not in parentctx:
                raise IOError(errno.ENOENT, 'Cannot find %r to patch' % fname)
            fctx = parentctx[fname]
            data = fctx.data()
            if 'l' in fctx.flags():
                data = 'link ' + data
            return cStringIO.StringIO(data).readlines()

        def writelines(self, fname, lines):
            files[fname] = ''.join(lines)

        def unlink(self, fname):
            files[fname] = None

    return mempatch


def filteriterhunks(meta):
    iterhunks = patch.iterhunks
    def filterhunks(ui, fp, sourcefile=None, textmode=False):
        applycurrent = False
        # Passing False instead of textmode because we should never
        # be ignoring EOL type.
        if not iterhunks.func_defaults:
            # Since 1.7 (cfedc529e4a1)
            gen = iterhunks(ui, fp)
        elif len(iterhunks.func_defaults) == 1:
            gen = iterhunks(ui, fp, sourcefile)
        else:
            gen = iterhunks(ui, fp, sourcefile, textmode)
        for data in gen:
            if data[0] == 'file':
                if data[1][1] in meta.filemap:
                    applycurrent = True
                else:
                    applycurrent = False
            assert data[0] != 'git', 'Filtering git hunks not supported.'
            if applycurrent:
                yield data
    return filterhunks


def diff_branchrev(ui, svn, meta, branch, branchpath, r, parentctx):
    """Extract all 'branch' content at a given revision.

    Return a tuple (files, filectxfn) where 'files' is the list of all files
    in the branch at the given revision, and 'filectxfn' is a memctx compatible
    callable to retrieve individual file information. Raise BadPatchApply upon
    error.
    """
    try:
        prev, pbranch, ppath = meta.get_source_rev(ctx=parentctx)
    except KeyError:
        prev, pbranch, ppath = None, None, None
    try:
        if prev is None or pbranch == branch:
            # letting patch handle binaries sounded
            # cool, but it breaks patch in sad ways
            d = svn.get_unified_diff(branchpath, r.revnum, deleted=False,
                                     ignore_type=False)
        else:
            d = svn.get_unified_diff(branchpath, r.revnum,
                                     other_path=ppath, other_rev=prev,
                                     deleted=True, ignore_type=True)
            if d:
                raise BadPatchApply('branch creation with mods')
    except svnwrap.SubversionRepoCanNotDiff:
        raise BadPatchApply('subversion diffing code is not supported')
    except svnwrap.SubversionException, e:
        if len(e.args) > 1 and e.args[1] != svnwrap.ERR_FS_NOT_FOUND:
            raise
        raise BadPatchApply('previous revision does not exist')
    if '\0' in d:
        raise BadPatchApply('binary diffs are not supported')
    files_data = {}
    # we have to pull each binary file by hand as a fulltext,
    # which sucks but we've got no choice
    binary_files = set(binary_file_re.findall(d))
    touched_files = set(binary_files)
    d2 = empty_file_patch_wont_make_re.sub('', d)
    d2 = property_exec_set_re.sub('', d2)
    d2 = property_exec_removed_re.sub('', d2)
    # Here we ensure that all files, including the new empty ones
    # are marked as touched. Content is loaded on demand.
    touched_files.update(any_file_re.findall(d))
    if d2.strip() and len(re.findall('\n[-+]', d2.strip())) > 0:
        try:
            oldpatchfile = patch.patchfile
            olditerhunks = patch.iterhunks
            patch.patchfile = mempatchproxy(parentctx, files_data)
            patch.iterhunks = filteriterhunks(meta)
            try:
                # We can safely ignore the changed list since we are
                # handling non-git patches. Touched files are known
                # by our memory patcher.
                patch_st = patch.applydiff(ui, cStringIO.StringIO(d2),
                                           {}, strip=0)
            finally:
                patch.patchfile = oldpatchfile
                patch.iterhunks = olditerhunks
        except patch.PatchError:
            # TODO: this happens if the svn server has the wrong mime
            # type stored and doesn't know a file is binary. It would
            # be better to do one file at a time and only do a
            # full fetch on files that had problems.
            raise BadPatchApply('patching failed')
        for x in files_data.iterkeys():
            ui.note('M  %s\n' % x)
        # if this patch didn't apply right, fall back to exporting the
        # entire rev.
        if patch_st == -1:
            assert False, ('This should only happen on case-insensitive'
                           ' volumes.')
        elif patch_st == 1:
            # When converting Django, I saw fuzz on .po files that was
            # causing revisions to end up failing verification. If that
            # can be fixed, maybe this won't ever be reached.
            raise BadPatchApply('patching succeeded with fuzz')
    else:
        ui.status('Not using patch for %s, diff had no hunks.\n' %
                  r.revnum)

    exec_files = {}
    for m in property_exec_removed_re.findall(d):
        exec_files[m] = False
    for m in property_exec_set_re.findall(d):
        exec_files[m] = True
    touched_files.update(exec_files)
    link_files = {}
    for m in property_special_set_re.findall(d):
        # TODO(augie) when a symlink is removed, patching will fail.
        # We're seeing that above - there's gotta be a better
        # workaround than just bailing like that.
        assert m in files_data
        link_files[m] = True
    for m in property_special_removed_re.findall(d):
        assert m in files_data
        link_files[m] = False

    unknown_files = set()
    for p in r.paths:
        action = r.paths[p].action
        if not p.startswith(branchpath) or action not in 'DR':
            continue
        if branchpath:
            p2 = p[len(branchpath)+1:].strip('/')
        else:
            p2 = p
        if p2 in parentctx:
            toucheds = [p2]
        else:
            # If this isn't in the parent ctx, it must've been a dir
            toucheds = [f for f in parentctx if f.startswith(p2 + '/')]
        if action == 'R':
            # Files were replaced, we don't know if they still exist
            unknown_files.update(toucheds)
        else:
            files_data.update((f, None) for f in toucheds)

    touched_files.update(files_data)
    touched_files.update(unknown_files)

    copies = getcopies(svn, meta, branch, branchpath, r, touched_files,
                       parentctx)

    def filectxfn(repo, memctx, path):
        if path in files_data and files_data[path] is None:
            raise IOError(errno.ENOENT, '%s is deleted' % path)

        if path in binary_files or path in unknown_files:
            pa = path
            if branchpath:
                pa = branchpath + '/' + path
            data, mode = svn.get_file(pa, r.revnum)
            isexe = 'x' in mode
            islink = 'l' in mode
        else:
            isexe = exec_files.get(path, 'x' in parentctx.flags(path))
            islink = link_files.get(path, 'l' in parentctx.flags(path))
            data = ''
            if path in files_data:
                data = files_data[path]
                if islink:
                    data = data[len('link '):]
            elif path in parentctx:
                data = parentctx[path].data()

        copied = copies.get(path)
        # TODO this branch feels like it should not be required,
        # and this may actually imply a bug in getcopies
        if copied not in parentctx.manifest():
            copied = None
        return context.memfilectx(path=path, data=data, islink=islink,
                                  isexec=isexe, copied=copied)

    return list(touched_files), filectxfn

def makecopyfinder(meta, r, branchpath):
    """Return a function detecting copies.

    Returned copyfinder(path) returns None if no copy information can
    be found or ((source, sourcerev), sourcepath) where "sourcepath" is the
    copy source path, "sourcerev" the source svn revision and "source" is the
    copy record path causing the copy to occur. If a single file was copied
    "sourcepath" and "source" are the same, while file copies dectected from
    directory copies return the copied source directory in "source".
    """
    # cache changeset contexts and map them to source svn revisions
    ctxs = {}
    def getctx(branch, svnrev):
        if svnrev in ctxs:
            return ctxs[svnrev]
        changeid = meta.get_parent_revision(svnrev + 1, branch, True)
        ctx = None
        if changeid != revlog.nullid:
            ctx = meta.repo.changectx(changeid)
        ctxs[svnrev] = ctx
        return ctx

    # filter copy information for current branch
    branchpath = (branchpath and branchpath + '/') or ''
    copies = []
    for path, e in r.paths.iteritems():
        if not e.copyfrom_path:
            continue
        if not path.startswith(branchpath):
            continue
        # compute converted source path and revision
        frompath, frombranch = meta.split_branch_path(e.copyfrom_path)[:2]
        if frompath is None:
            continue
        fromctx = getctx(frombranch, e.copyfrom_rev)
        if fromctx is None:
            continue
        destpath = path[len(branchpath):]
        copies.append((destpath, (frompath, fromctx)))

    copies.sort(reverse=True)
    exactcopies = dict(copies)

    def finder(path):
        if path in exactcopies:
            return exactcopies[path], exactcopies[path][0]
        # look for parent directory copy, longest first
        for dest, (source, sourcectx) in copies:
            dest = dest + '/'
            if not path.startswith(dest):
                continue
            sourcepath = source + '/' + path[len(dest):]
            return (source, sourcectx), sourcepath
        return None

    return finder

def getcopies(svn, meta, branch, branchpath, r, files, parentctx):
    """Return a mapping {dest: source} for every file copied into r.
    """
    if parentctx.node() == revlog.nullid:
        return {}

    # Extract svn copy information, group them by copy source.
    # The idea is to duplicate the replay behaviour where copies are
    # evaluated per copy event (one event for all files in a directory copy,
    # one event for single file copy). We assume that copy events match
    # copy sources in revision info.
    svncopies = {}
    finder = makecopyfinder(meta, r, branchpath)
    for f in files:
        copy = finder(f)
        if copy:
            svncopies.setdefault(copy[0], []).append((f, copy[1]))
    if not svncopies:
        return {}

    # check svn copies really make sense in mercurial
    hgcopies = {}
    for (sourcepath, sourcectx), copies in svncopies.iteritems():
        for k, v in copies:
            if not util.issamefile(sourcectx, parentctx, v):
                continue
            hgcopies.update({k: v})
    return hgcopies

def fetch_externals(ui, svn, branchpath, r, parentctx):
    """Extract svn:externals for the current revision and branch

    Return an externalsfile instance or None if there are no externals
    to convert and never were.
    """
    externals = svnexternals.parse(ui, parentctx)
    # Detect property additions only, changes are handled by checking
    # existing entries individually. Projects are unlikely to store
    # externals on many different root directories, so we trade code
    # duplication and complexity for a constant lookup price at every
    # revision in the common case.
    dirs = set(externals)
    if parentctx.node() == revlog.nullid:
        dirs.update([p for p,k in svn.list_files(branchpath, r.revnum) if k == 'd'])
        dirs.add('')
    else:
        branchprefix = (branchpath and branchpath + '/') or branchpath
        for path, e in r.paths.iteritems():
            if e.action == 'D':
                continue
            if not path.startswith(branchprefix) and path != branchpath:
                continue
            kind = svn.checkpath(path, r.revnum)
            if kind != 'd':
                continue
            path = path[len(branchprefix):]
            dirs.add(path)
            if e.action == 'M' or (e.action == 'A' and e.copyfrom_path):
                # Do not recurse in copied directories, changes are marked
                # as 'M', except for the copied one.
                continue
            for child, k in svn.list_files(branchprefix + path, r.revnum):
                if k == 'd':
                    dirs.add((path + '/' + child).strip('/'))

    # Retrieve new or updated values
    for dir in dirs:
        try:
            dpath = (branchpath and branchpath + '/' + dir) or dir
            values = svn.list_props(dpath, r.revnum)
            externals[dir] = values.get('svn:externals', '')
        except IOError:
            externals[dir] = ''
    return externals


def fetch_branchrev(svn, meta, branch, branchpath, r, parentctx):
    """Extract all 'branch' content at a given revision.

    Return a tuple (files, filectxfn) where 'files' is the list of all files
    in the branch at the given revision, and 'filectxfn' is a memctx compatible
    callable to retrieve individual file information.
    """
    files = []
    if parentctx.node() == revlog.nullid:
        # Initial revision, fetch all files
        for path, kind in svn.list_files(branchpath, r.revnum):
            if kind == 'f':
                files.append(path)
    else:
        branchprefix = (branchpath and branchpath + '/') or ''
        for path, e in r.paths.iteritems():
            if not path.startswith(branchprefix):
                continue
            if not meta.is_path_valid(path):
                continue
            kind = svn.checkpath(path, r.revnum)
            path = path[len(branchprefix):]
            if kind == 'f':
                files.append(path)
            elif kind == 'd':
                if e.action == 'M':
                    continue
                dirpath = branchprefix + path
                for child, k in svn.list_files(dirpath, r.revnum):
                    if k == 'f':
                        if path:
                            childpath = '%s/%s' % (path, child)
                        else:
                            childpath = child
                        files.append(childpath)
                if e.action == 'R':
                    # Check all files in replaced directory
                    path = path + '/'
                    files += [f for f in parentctx if f.startswith(path)]
            else:
                if path in parentctx:
                    files.append(path)
                    continue
                # Assume it's a deleted directory
                path = path + '/'
                deleted = [f for f in parentctx if f.startswith(path)]
                files += deleted

    copies = getcopies(svn, meta, branch, branchpath, r, files, parentctx)

    def filectxfn(repo, memctx, path):
        svnpath = path
        if branchpath:
            svnpath = branchpath + '/' + path
        data, mode = svn.get_file(svnpath, r.revnum)
        isexec = 'x' in mode
        islink = 'l' in mode
        copied = copies.get(path)
        # TODO this branch feels like it should not be required,
        # and this may actually imply a bug in getcopies
        if copied not in parentctx.manifest():
            copied = None
        return context.memfilectx(path=path, data=data, islink=islink,
                                  isexec=isexec, copied=copied)

    return files, filectxfn

def checkbranch(meta, r, branch):
    branchedits = meta.revmap.branchedits(branch, r)
    if not branchedits:
        return None
    branchtip = branchedits[0][1]
    for child in meta.repo[branchtip].children():
        b = child.branch() != 'default' and child.branch() or None
        if b == branch and child.extra().get('close'):
            return None
    return branchtip

def branches_in_paths(meta, tbdelta, paths, revnum, checkpath, listdir):
    '''Given a list of paths, return mapping of all branches touched
    to their branch path.
    '''
    branches = {}
    paths_need_discovery = []
    for p in paths:
        relpath, branch, branchpath = meta.split_branch_path(p)
        if relpath is not None:
            branches[branch] = branchpath
        elif paths[p].action == 'D' and not meta.get_path_tag(p):
            ln = meta.localname(p)
            # must check in branches_to_delete as well, because this runs after we
            # already updated the branch map
            if ln in meta.branches or ln in tbdelta['branches'][1]:
                branches[ln] = p
        else:
            paths_need_discovery.append(p)

    if not paths_need_discovery:
        return branches

    paths_need_discovery = [(len(p), p) for p in paths_need_discovery]
    paths_need_discovery.sort()
    paths_need_discovery = [p[1] for p in paths_need_discovery]
    actually_files = []
    while paths_need_discovery:
        p = paths_need_discovery.pop(0)
        path_could_be_file = True
        ind = 0
        while ind < len(paths_need_discovery) and not paths_need_discovery:
            if op.startswith(p):
                path_could_be_file = False
            ind += 1
        if path_could_be_file:
            if checkpath(p, revnum) == 'f':
                actually_files.append(p)
            # if there's a copyfrom_path and there were files inside that copyfrom,
            # we need to detect those branches. It's a little thorny and slow, but
            # seems to be the best option.
            elif paths[p].copyfrom_path and not p.startswith('tags/'):
                paths_need_discovery.extend(['%s/%s' % (p,x[0])
                                             for x in listdir(p, revnum)
                                             if x[1] == 'f'])

        if not actually_files:
            continue

        filepaths = [p.split('/') for p in actually_files]
        filepaths = [(len(p), p) for p in filepaths]
        filepaths.sort()
        filepaths = [p[1] for p in filepaths]
        while filepaths:
            path = filepaths.pop(0)
            parentdir = '/'.join(path[:-1])
            filepaths = [p for p in filepaths if not '/'.join(p).startswith(parentdir)]
            branchpath = meta.normalize(parentdir)
            if branchpath.startswith('tags/'):
                continue
            branchname = meta.localname(branchpath)
            if branchpath.startswith('trunk/'):
                branches[meta.localname('trunk')] = 'trunk'
                continue
            if branchname and branchname.startswith('../'):
                continue
            branches[branchname] = branchpath
    return branches

def convert_rev(ui, meta, svn, r, tbdelta, firstrun):
    # this server fails at replay

    branches = branches_in_paths(meta, tbdelta, r.paths, r.revnum,
                                 svn.checkpath, svn.list_files)
    brpaths = branches.values()
    bad_branch_paths = {}
    for br, bp in branches.iteritems():
        bad_branch_paths[br] = []

        # This next block might be needed, but for now I'm omitting it until it
        # can be proven necessary.
        # for bad in brpaths:
        #     if bad.startswith(bp) and len(bad) > len(bp):
        #         bad_branch_paths[br].append(bad[len(bp)+1:])

        # We've go a branch that contains other branches. We have to be careful
        # to get results similar to real replay in this case.
        for existingbr in meta.branches:
            bad = meta.remotename(existingbr)
            if bad.startswith(bp) and len(bad) > len(bp):
                bad_branch_paths[br].append(bad[len(bp)+1:])

    deleted_branches = {}
    for p in r.paths:
        tag = meta.get_path_tag(p)
        if tag and tag not in meta.tags:
            continue
        branch = meta.localname(p)
        if not (r.paths[p].action == 'R' and branch in meta.branches):
            continue
        closed = checkbranch(meta, r, branch)
        if closed is not None:
            deleted_branches[branch] = closed

    date = meta.fixdate(r.date)
    check_deleted_branches = set(tbdelta['branches'][1])
    for b in branches:
        parentctx = meta.repo[meta.get_parent_revision(r.revnum, b)]
        tag = meta.get_path_tag(meta.remotename(b))
        kind = svn.checkpath(branches[b], r.revnum)
        if kind != 'd':
            if not tag:
                # Branch does not exist at this revision. Get parent
                # revision and remove everything.
                deleted_branches[b] = parentctx.node()
            continue

        incremental = (meta.revmap.oldest > 0)

        if incremental:
            try:
                files_touched, filectxfn2 = diff_branchrev(
                    ui, svn, meta, b, branches[b], r, parentctx)
            except BadPatchApply, e:
                # Either this revision or the previous one does not exist.
                ui.note("Fetching entire revision: %s.\n" % e.args[0])
                incremental = False
        if not incremental:
            files_touched, filectxfn2 = fetch_branchrev(
                svn, meta, b, branches[b], r, parentctx)

        externals = {}
        if meta.layout != 'single':
            externals = fetch_externals(ui, svn, branches[b], r, parentctx)
            externals = svnexternals.getchanges(ui, meta.repo, parentctx,
                                                externals)
            files_touched.extend(externals)

        def filectxfn(repo, memctx, path):
            if path in externals:
                if externals[path] is None:
                    raise IOError(errno.ENOENT, 'no externals')
                return context.memfilectx(path=path, data=externals[path],
                                          islink=False, isexec=False,
                                          copied=None)
            for bad in bad_branch_paths[b]:
                if path.startswith(bad):
                    raise IOError(errno.ENOENT, 'Path %s is bad' % path)
            return filectxfn2(repo, memctx, path)

        if '' in files_touched:
            files_touched.remove('')
        excluded = [f for f in files_touched if f not in meta.filemap]
        for f in excluded:
            files_touched.remove(f)

        if b:
            # Regular tag without modifications, it will be committed by
            # svnmeta.committag(), we can skip the whole branch for now
            if (tag and tag not in meta.tags and
                b not in meta.branches
                and b not in meta.repo.branchtags()
                and not files_touched):
                continue

        if parentctx.node() == node.nullid and not files_touched:
            meta.repo.ui.debug('skipping commit since parent is null and no files touched.\n')
            continue

        for f in files_touched:
            if f:
                # this is a case that really shouldn't ever happen, it means
                # something is very wrong
                assert f[0] != '/'

        extra = meta.genextra(r.revnum, b)
        if tag:
            if parentctx.node() == node.nullid:
                continue
            extra.update({'branch': parentctx.extra().get('branch', None),
                          'close': 1})

        origbranch = extra.get('branch', None)
        meta.mapbranch(extra)
        current_ctx = context.memctx(meta.repo,
                                     [parentctx.node(), revlog.nullid],
                                     r.message or util.default_commit_msg(ui),
                                     files_touched,
                                     filectxfn,
                                     meta.authors[r.author],
                                     date,
                                     extra)
        ha = meta.repo.commitctx(current_ctx)

        if not tag:
            if (not origbranch in meta.branches
                and not meta.get_path_tag(meta.remotename(origbranch))):
                meta.branches[origbranch] = None, 0, r.revnum
            meta.revmap[r.revnum, b] = ha
        else:
            meta.movetag(tag, ha, r, date)
            meta.addedtags.pop(tag, None)
        util.describe_commit(ui, ha, b)

    # These are branches with an 'R' status in svn log. This means they were
    # replaced by some other branch, so we need to verify they get marked as closed.
    for branch in check_deleted_branches:
        closed = checkbranch(meta, r, branch)
        if closed is not None:
            deleted_branches[branch] = closed

    return deleted_branches