comparison mercurial/patch.py @ 14451:c78d41db6f88

patch: refactor file creation/removal detection

The patcher has to know if a file is being created or removed to check if the
target already exists, or to actually unlink the file when a hunk emptying it
is applied. This was done by embedding the creation/removal information in the
first (and only) hunk attached to the file.

There are two problems with this approach:

- creation/removal is really a property of the file being patched and not its
  hunk.

- for regular patches, file creation cannot be deduced at parsing time: there
  are case where the *stripped* file paths must be compared. Modifying hunks
  after their creation is clumsy and prevent further refactorings related to
  copies handling.

Instead, we delegate this job to selectfile() which has all the relevant
information, and remove the hunk createfile() and rmfile() methods.

14450:d1a1578c5f78

unidesc = re.compile('@@ -(\d+)(,(\d+))? \+(\d+)(,(\d+))? @@')
contextdesc = re.compile('(---|\*\*\*) (\d+)(,(\d+))? (---|\*\*\*)')
eolmodes = ['strict', 'crlf', 'lf', 'auto']

class patchfile(object):
    def __init__(self, ui, fname, backend, mode, missing=False,
                 eolmode='strict'):
        self.fname = fname
        self.eolmode = eolmode
        self.eol = None
        self.backend = backend
        self.ui = ui
        self.lines = []
        self.exists = False
        self.missing = missing
        self.mode = mode
        if not missing:
            try:
                data, mode = self.backend.getfile(fname)
                if data:
                    self.lines = data.splitlines(True)

        if self.missing:
            self.rej.append(h)
            return -1

        if self.exists and h.createfile():
            self.ui.warn(_("file %s already exists\n") % self.fname)
            self.rej.append(h)
            return -1

        if isinstance(h, binhunk):
            if h.rmfile():
                self.backend.unlink(self.fname)
            else:
                self.lines[:] = h.new()
                self.offset += len(h.new())
                self.dirty = True

        orig_start = start
        # if there's skew we want to emit the "(offset %d lines)" even
        # when the hunk cleanly applies at start + skew, so skip the
        # fast case code
        if self.skew == 0 and diffhelpers.testhunk(old, self.lines, start) == 0:
            if h.rmfile():
                self.backend.unlink(self.fname)
            else:
                self.lines[start : start + h.lena] = h.new()
                self.offset += h.lenb - h.lena
                self.dirty = True

            self.writelines(self.fname, self.lines, self.mode)
        self.write_rej()
        return len(self.rej)

class hunk(object):
    def __init__(self, desc, num, lr, context, create=False, remove=False):
        self.number = num
        self.desc = desc
        self.hunk = [desc]
        self.a = []
        self.b = []

        if lr is not None:
            if context:
                self.read_context_hunk(lr)
            else:
                self.read_unified_hunk(lr)
        self.create = create
        self.remove = remove and not create

    def getnormalized(self):
        """Return a copy with line endings normalized to LF."""

        def normalize(lines):

                    line = line[:-2] + '\n'
                nlines.append(line)
            return nlines

        # Dummy object, it is rebuilt manually
        nh = hunk(self.desc, self.number, None, None, False, False)
        nh.number = self.number
        nh.desc = self.desc
        nh.hunk = self.hunk
        nh.a = normalize(self.a)
        nh.b = normalize(self.b)
        nh.starta = self.starta
        nh.startb = self.startb
        nh.lena = self.lena
        nh.lenb = self.lenb
        nh.create = self.create
        nh.remove = self.remove
        return nh

    def read_unified_hunk(self, lr):
        m = unidesc.match(self.desc)
        if not m:

            lr.push(l)

    def complete(self):
        return len(self.a) == self.lena and len(self.b) == self.lenb

    def createfile(self):
        return self.starta == 0 and self.lena == 0 and self.create

    def rmfile(self):
        return self.startb == 0 and self.lenb == 0 and self.remove

    def fuzzit(self, l, fuzz, toponly):
        # this removes context lines from the top and bottom of list 'l'.  It
        # checks the hunk to make sure only context lines are removed, and then
        # returns a new shortened list of lines.
        fuzz = min(fuzz, len(l)-1)

    def new(self, fuzz=0, toponly=False):
        return self.fuzzit(self.b, fuzz, toponly)

class binhunk:
    'A binary patch file. Only understands literals so far.'
    def __init__(self, gitpatch, lr):
        self.gitpatch = gitpatch
        self.text = None
        self.hunk = ['GIT binary patch\n']
        self._read(lr)

    def createfile(self):
        return self.gitpatch.op == 'ADD'

    def rmfile(self):
        return self.gitpatch.op == 'DELETE'

    def complete(self):
        return self.text is not None

    def new(self):

def selectfile(backend, afile_orig, bfile_orig, hunk, strip, gp):
    if gp:
        # Git patches do not play games. Excluding copies from the
        # following heuristic avoids a lot of confusion
        fname = pathstrip(gp.path, strip - 1)[1]
        missing = not hunk.createfile() and not backend.exists(fname)
        return fname, missing
    nulla = afile_orig == "/dev/null"
    nullb = bfile_orig == "/dev/null"
    abase, afile = pathstrip(afile_orig, strip)
    gooda = not nulla and backend.exists(afile)
    bbase, bfile = pathstrip(bfile_orig, strip)
    if afile == bfile:
        goodb = gooda
    else:
        goodb = not nullb and backend.exists(bfile)
    createfunc = hunk.createfile
    missing = not goodb and not gooda and not createfunc()

    # some diff programs apparently produce patches where the afile is
    # not /dev/null, but afile starts with bfile
    abasedir = afile[:afile.rfind('/') + 1]
    bbasedir = bfile[:bfile.rfind('/') + 1]
    if missing and abasedir == bbasedir and afile.startswith(bfile):
        # this isn't very pretty
        hunk.create = True
        if createfunc():
            missing = False
        else:
            hunk.create = False

    # If afile is "a/b/foo" and bfile is "a/b/foo.orig" we assume the
    # diff is between a file and its backup. In this case, the original
    # file should be patched (see original mpatch code).
    isbackup = (abase == bbase and bfile.startswith(afile))

        elif not nulla:
            fname = afile
        else:
            raise PatchError(_("undefined source and destination files"))

    return fname, missing

def scangitpatch(lr, firstline):
    """
    Git patches can emit:
    - rename a to b

            or x.startswith('GIT binary patch')):
            gp = None
            if gitpatches and gitpatches[-1][0] == bfile:
                gp = gitpatches.pop()[1]
            if x.startswith('GIT binary patch'):
                h = binhunk(gp, lr)
            else:
                if context is None and x.startswith('***************'):
                    context = True
                create = afile == '/dev/null' or gp and gp.op == 'ADD'
                remove = bfile == '/dev/null' or gp and gp.op == 'DELETE'
                h = hunk(x, hunknum + 1, lr, context, create, remove)
            hunknum += 1
            if emitfile:
                emitfile = False
                yield 'file', (afile, bfile, h, gp)
            yield 'hunk', h

                    backend.setfile(path, data, gp.mode)
            if not first_hunk:
                continue
            try:
                mode = gp and gp.mode or None
                current_file, missing = selectfile(backend, afile, bfile,
                                                   first_hunk, strip, gp)
                current_file = patcher(ui, current_file, backend, mode,
                                       missing=missing, eolmode=eolmode)
            except PatchError, inst:
                ui.warn(str(inst) + '\n')
                current_file = None
                rejects += 1
                continue

                    changed.add(pathstrip(gp.path, strip - 1)[1])
                    if gp.op == 'RENAME':
                        changed.add(pathstrip(gp.oldpath, strip - 1)[1])
                if not first_hunk:
                    continue
                current_file, missing = selectfile(backend, afile, bfile,
                                                   first_hunk, strip, gp)
                changed.add(current_file)
            elif state not in ('hunk', 'git'):
                raise util.Abort(_('unsupported parser state: %s') % state)
        return changed
    finally:

14451:c78d41db6f88

unidesc = re.compile('@@ -(\d+)(,(\d+))? \+(\d+)(,(\d+))? @@')
contextdesc = re.compile('(---|\*\*\*) (\d+)(,(\d+))? (---|\*\*\*)')
eolmodes = ['strict', 'crlf', 'lf', 'auto']

class patchfile(object):
    def __init__(self, ui, fname, backend, mode, create, remove, missing=False,
                 eolmode='strict'):
        self.fname = fname
        self.eolmode = eolmode
        self.eol = None
        self.backend = backend
        self.ui = ui
        self.lines = []
        self.exists = False
        self.missing = missing
        self.mode = mode
        self.create = create
        self.remove = remove
        if not missing:
            try:
                data, mode = self.backend.getfile(fname)
                if data:
                    self.lines = data.splitlines(True)

        if self.missing:
            self.rej.append(h)
            return -1

        if self.exists and self.create:
            self.ui.warn(_("file %s already exists\n") % self.fname)
            self.rej.append(h)
            return -1

        if isinstance(h, binhunk):
            if self.remove:
                self.backend.unlink(self.fname)
            else:
                self.lines[:] = h.new()
                self.offset += len(h.new())
                self.dirty = True

        orig_start = start
        # if there's skew we want to emit the "(offset %d lines)" even
        # when the hunk cleanly applies at start + skew, so skip the
        # fast case code
        if self.skew == 0 and diffhelpers.testhunk(old, self.lines, start) == 0:
            if self.remove:
                self.backend.unlink(self.fname)
            else:
                self.lines[start : start + h.lena] = h.new()
                self.offset += h.lenb - h.lena
                self.dirty = True

            self.writelines(self.fname, self.lines, self.mode)
        self.write_rej()
        return len(self.rej)

class hunk(object):
    def __init__(self, desc, num, lr, context):
        self.number = num
        self.desc = desc
        self.hunk = [desc]
        self.a = []
        self.b = []

        if lr is not None:
            if context:
                self.read_context_hunk(lr)
            else:
                self.read_unified_hunk(lr)

    def getnormalized(self):
        """Return a copy with line endings normalized to LF."""

        def normalize(lines):

                    line = line[:-2] + '\n'
                nlines.append(line)
            return nlines

        # Dummy object, it is rebuilt manually
        nh = hunk(self.desc, self.number, None, None)
        nh.number = self.number
        nh.desc = self.desc
        nh.hunk = self.hunk
        nh.a = normalize(self.a)
        nh.b = normalize(self.b)
        nh.starta = self.starta
        nh.startb = self.startb
        nh.lena = self.lena
        nh.lenb = self.lenb
        return nh

    def read_unified_hunk(self, lr):
        m = unidesc.match(self.desc)
        if not m:

            lr.push(l)

    def complete(self):
        return len(self.a) == self.lena and len(self.b) == self.lenb

    def fuzzit(self, l, fuzz, toponly):
        # this removes context lines from the top and bottom of list 'l'.  It
        # checks the hunk to make sure only context lines are removed, and then
        # returns a new shortened list of lines.
        fuzz = min(fuzz, len(l)-1)

    def new(self, fuzz=0, toponly=False):
        return self.fuzzit(self.b, fuzz, toponly)

class binhunk:
    'A binary patch file. Only understands literals so far.'
    def __init__(self, lr):
        self.text = None
        self.hunk = ['GIT binary patch\n']
        self._read(lr)

    def complete(self):
        return self.text is not None

    def new(self):

def selectfile(backend, afile_orig, bfile_orig, hunk, strip, gp):
    if gp:
        # Git patches do not play games. Excluding copies from the
        # following heuristic avoids a lot of confusion
        fname = pathstrip(gp.path, strip - 1)[1]
        create = gp.op == 'ADD'
        remove = gp.op == 'DELETE'
        missing = not create and not backend.exists(fname)
        return fname, missing, create, remove
    nulla = afile_orig == "/dev/null"
    nullb = bfile_orig == "/dev/null"
    create = nulla and hunk.starta == 0 and hunk.lena == 0
    remove = nullb and hunk.startb == 0 and hunk.lenb == 0
    abase, afile = pathstrip(afile_orig, strip)
    gooda = not nulla and backend.exists(afile)
    bbase, bfile = pathstrip(bfile_orig, strip)
    if afile == bfile:
        goodb = gooda
    else:
        goodb = not nullb and backend.exists(bfile)
    missing = not goodb and not gooda and not create

    # some diff programs apparently produce patches where the afile is
    # not /dev/null, but afile starts with bfile
    abasedir = afile[:afile.rfind('/') + 1]
    bbasedir = bfile[:bfile.rfind('/') + 1]
    if (missing and abasedir == bbasedir and afile.startswith(bfile)
        and hunk.starta == 0 and hunk.lena == 0):
        create = True
        missing = False

    # If afile is "a/b/foo" and bfile is "a/b/foo.orig" we assume the
    # diff is between a file and its backup. In this case, the original
    # file should be patched (see original mpatch code).
    isbackup = (abase == bbase and bfile.startswith(afile))

        elif not nulla:
            fname = afile
        else:
            raise PatchError(_("undefined source and destination files"))

    return fname, missing, create, remove

def scangitpatch(lr, firstline):
    """
    Git patches can emit:
    - rename a to b

            or x.startswith('GIT binary patch')):
            gp = None
            if gitpatches and gitpatches[-1][0] == bfile:
                gp = gitpatches.pop()[1]
            if x.startswith('GIT binary patch'):
                h = binhunk(lr)
            else:
                if context is None and x.startswith('***************'):
                    context = True
                h = hunk(x, hunknum + 1, lr, context)
            hunknum += 1
            if emitfile:
                emitfile = False
                yield 'file', (afile, bfile, h, gp)
            yield 'hunk', h

                    backend.setfile(path, data, gp.mode)
            if not first_hunk:
                continue
            try:
                mode = gp and gp.mode or None
                current_file, missing, create, remove = selectfile(
                    backend, afile, bfile, first_hunk, strip, gp)
                current_file = patcher(ui, current_file, backend, mode,
                                       create, remove, missing=missing,
                                       eolmode=eolmode)
            except PatchError, inst:
                ui.warn(str(inst) + '\n')
                current_file = None
                rejects += 1
                continue

                    changed.add(pathstrip(gp.path, strip - 1)[1])
                    if gp.op == 'RENAME':
                        changed.add(pathstrip(gp.oldpath, strip - 1)[1])
                if not first_hunk:
                    continue
                current_file, missing, create, remove = selectfile(
                    backend, afile, bfile, first_hunk, strip, gp)
                changed.add(current_file)
            elif state not in ('hunk', 'git'):
                raise util.Abort(_('unsupported parser state: %s') % state)
        return changed
    finally: