Changeset - 1f4d4b8d72f5
[Not reviewed]
beta
0 2 0
Marcin Kuzminski - 13 years ago 2012-07-30 22:45:43
marcin@python-works.com
switched git_command to subprocession for non-blocking Popen.
2 files changed with 25 insertions and 13 deletions:
0 comments (0 inline, 0 general)
rhodecode/lib/subprocessio.py
Show inline comments
 
@@ -245,161 +245,164 @@ class BufferedGenerator():
 
        '''
 
        return self.worker.EOF.is_set()
 

	
 
    @property
 
    def length(self):
 
        '''
 
        returns int.
 

	
 
        This is the lenght of the que of chunks, not the length of
 
        the combined contents in those chunks.
 

	
 
        __len__() cannot be meaningfully implemented because this
 
        reader is just flying throuh a bottomless pit content and
 
        can only know the lenght of what it already saw.
 

	
 
        If __len__() on WSGI server per PEP 3333 returns a value,
 
        the responce's length will be set to that. In order not to
 
        confuse WSGI PEP3333 servers, we will not implement __len__
 
        at all.
 
        '''
 
        return len(self.data)
 

	
 
    def prepend(self, x):
 
        self.data.appendleft(x)
 

	
 
    def append(self, x):
 
        self.data.append(x)
 

	
 
    def extend(self, o):
 
        self.data.extend(o)
 

	
 
    def __getitem__(self, i):
 
        return self.data[i]
 

	
 

	
 
class SubprocessIOChunker(object):
 
    '''
 
    Processor class wrapping handling of subprocess IO.
 

	
 
    In a way, this is a "communicate()" replacement with a twist.
 

	
 
    - We are multithreaded. Writing in and reading out, err are all sep threads.
 
    - We support concurrent (in and out) stream processing.
 
    - The output is not a stream. It's a queue of read string (bytes, not unicode)
 
      chunks. The object behaves as an iterable. You can "for chunk in obj:" us.
 
    - We are non-blocking in more respects than communicate()
 
      (reading from subprocess out pauses when internal buffer is full, but
 
       does not block the parent calling code. On the flip side, reading from
 
       slow-yielding subprocess may block the iteration until data shows up. This
 
       does not block the parallel inpipe reading occurring parallel thread.)
 

	
 
    The purpose of the object is to allow us to wrap subprocess interactions into
 
    and interable that can be passed to a WSGI server as the application's return
 
    value. Because of stream-processing-ability, WSGI does not have to read ALL
 
    of the subprocess's output and buffer it, before handing it to WSGI server for
 
    HTTP response. Instead, the class initializer reads just a bit of the stream
 
    to figure out if error ocurred or likely to occur and if not, just hands the
 
    further iteration over subprocess output to the server for completion of HTTP
 
    response.
 

	
 
    The real or perceived subprocess error is trapped and raised as one of
 
    EnvironmentError family of exceptions
 

	
 
    Example usage:
 
    #    try:
 
    #        answer = SubprocessIOChunker(
 
    #            cmd,
 
    #            input,
 
    #            buffer_size = 65536,
 
    #            chunk_size = 4096
 
    #            )
 
    #    except (EnvironmentError) as e:
 
    #        print str(e)
 
    #        raise e
 
    #
 
    #    return answer
 

	
 

	
 
    '''
 
    def __init__(self, cmd, inputstream=None, buffer_size=65536,
 
                 chunk_size=4096, starting_values=[], **kwargs):
 
        '''
 
        Initializes SubprocessIOChunker
 

	
 
        :param cmd: A Subprocess.Popen style "cmd". Can be string or array of strings
 
        :param inputstream: (Default: None) A file-like, string, or file pointer.
 
        :param buffer_size: (Default: 65536) A size of total buffer per stream in bytes.
 
        :param chunk_size: (Default: 4096) A max size of a chunk. Actual chunk may be smaller.
 
        :param starting_values: (Default: []) An array of strings to put in front of output que.
 
        '''
 

	
 
        if inputstream:
 
            input_streamer = StreamFeeder(inputstream)
 
            input_streamer.start()
 
            inputstream = input_streamer.output
 

	
 
        if isinstance(cmd, (list, tuple)):
 
            cmd = ' '.join(cmd)
 

	
 
        _p = subprocess.Popen(cmd,
 
            bufsize=-1,
 
            shell=True,
 
            stdin=inputstream,
 
            stdout=subprocess.PIPE,
 
            stderr=subprocess.PIPE,
 
            **kwargs
 
            )
 

	
 
        bg_out = BufferedGenerator(_p.stdout, buffer_size, chunk_size, starting_values)
 
        bg_err = BufferedGenerator(_p.stderr, 16000, 1, bottomless=True)
 

	
 
        while not bg_out.done_reading and not bg_out.reading_paused and not bg_err.length:
 
            # doing this until we reach either end of file, or end of buffer.
 
            bg_out.data_added_event.wait(1)
 
            bg_out.data_added_event.clear()
 

	
 
        # at this point it's still ambiguous if we are done reading or just full buffer.
 
        # Either way, if error (returned by ended process, or implied based on
 
        # presence of stuff in stderr output) we error out.
 
        # Else, we are happy.
 
        _returncode = _p.poll()
 
        if _returncode or (_returncode == None and bg_err.length):
 
            try:
 
                _p.terminate()
 
            except:
 
                pass
 
            bg_out.stop()
 
            bg_err.stop()
 
            err = '%r' % ''.join(bg_err)
 
            raise EnvironmentError("Subprocess exited due to an error.\n" + err)
 
            err = '%s' % ''.join(bg_err)
 
            raise EnvironmentError("Subprocess exited due to an error:\n" + err)
 

	
 
        self.process = _p
 
        self.output = bg_out
 
        self.error = bg_err
 

	
 
    def __iter__(self):
 
        return self
 

	
 
    def next(self):
 
        if self.process.poll():
 
            err = '%r' % ''.join(self.error)
 
            err = '%s' % ''.join(self.error)
 
            raise EnvironmentError("Subprocess exited due to an error:\n" + err)
 
        return self.output.next()
 

	
 
    def throw(self, type, value=None, traceback=None):
 
        if self.output.length or not self.output.done_reading:
 
            raise type(value)
 

	
 
    def close(self):
 
        try:
 
            self.process.terminate()
 
        except:
 
            pass
 
        try:
 
            self.output.close()
 
        except:
 
            pass
 
        try:
 
            self.error.close()
 
        except:
 
            pass
 

	
 
    def __del__(self):
 
        self.close()
rhodecode/lib/vcs/backends/git/repository.py
Show inline comments
 
# -*- coding: utf-8 -*-
 
"""
 
    vcs.backends.git
 
    ~~~~~~~~~~~~~~~~
 

	
 
    Git backend implementation.
 

	
 
    :created_on: Apr 8, 2010
 
    :copyright: (c) 2010-2011 by Marcin Kuzminski, Lukasz Balcerzak.
 
"""
 

	
 
import os
 
import re
 
import time
 
import posixpath
 
import logging
 
import traceback
 
from dulwich.repo import Repo, NotGitRepository
 
#from dulwich.config import ConfigFile
 
from string import Template
 
from subprocess import Popen, PIPE
 
from rhodecode.lib.vcs.backends.base import BaseRepository
 
from rhodecode.lib.vcs.exceptions import BranchDoesNotExistError
 
from rhodecode.lib.vcs.exceptions import ChangesetDoesNotExistError
 
from rhodecode.lib.vcs.exceptions import EmptyRepositoryError
 
from rhodecode.lib.vcs.exceptions import RepositoryError
 
from rhodecode.lib.vcs.exceptions import TagAlreadyExistError
 
from rhodecode.lib.vcs.exceptions import TagDoesNotExistError
 
from rhodecode.lib.vcs.utils import safe_unicode, makedate, date_fromtimestamp
 
from rhodecode.lib.vcs.utils.lazy import LazyProperty
 
from rhodecode.lib.vcs.utils.ordered_dict import OrderedDict
 
from rhodecode.lib.vcs.utils.paths import abspath
 
from rhodecode.lib.vcs.utils.paths import get_user_home
 
from .workdir import GitWorkdir
 
from .changeset import GitChangeset
 
from .inmemory import GitInMemoryChangeset
 
from .config import ConfigFile
 
from rhodecode.lib import subprocessio
 

	
 

	
 
log = logging.getLogger(__name__)
 

	
 

	
 
class GitRepository(BaseRepository):
 
    """
 
    Git repository backend.
 
    """
 
    DEFAULT_BRANCH_NAME = 'master'
 
    scm = 'git'
 

	
 
    def __init__(self, repo_path, create=False, src_url=None,
 
                 update_after_clone=False, bare=False):
 

	
 
        self.path = abspath(repo_path)
 
        self._repo = self._get_repo(create, src_url, update_after_clone, bare)
 
        #temporary set that to now at later we will move it to constructor
 
        baseui = None
 
        if baseui is None:
 
            from mercurial.ui import ui
 
            baseui = ui()
 
        # patch the instance of GitRepo with an "FAKE" ui object to add
 
        # compatibility layer with Mercurial
 
        setattr(self._repo, 'ui', baseui)
 

	
 
        try:
 
            self.head = self._repo.head()
 
        except KeyError:
 
            self.head = None
 

	
 
        self._config_files = [
 
            bare and abspath(self.path, 'config') or abspath(self.path, '.git',
 
                'config'),
 
            abspath(get_user_home(), '.gitconfig'),
 
        ]
 
        self.bare = self._repo.bare
 

	
 
    @LazyProperty
 
    def revisions(self):
 
        """
 
        Returns list of revisions' ids, in ascending order.  Being lazy
 
        attribute allows external tools to inject shas from cache.
 
        """
 
        return self._get_all_revisions()
 

	
 
    def run_git_command(self, cmd):
 
        """
 
        Runs given ``cmd`` as git command and returns tuple
 
        (returncode, stdout, stderr).
 

	
 
        .. note::
 
           This method exists only until log/blame functionality is implemented
 
           at Dulwich (see https://bugs.launchpad.net/bugs/645142). Parsing
 
           os command's output is road to hell...
 

	
 
        :param cmd: git command to be executed
 
        """
 

	
 
        _copts = ['-c', 'core.quotepath=false', ]
 
        _str_cmd = False
 
        if isinstance(cmd, basestring):
 
            cmd = [cmd]
 
            _str_cmd = True
 

	
 
        gitenv = os.environ
 
        # need to clean fix GIT_DIR !
 
        if 'GIT_DIR' in gitenv:
 
            del gitenv['GIT_DIR']
 
        gitenv['GIT_CONFIG_NOGLOBAL'] = '1'
 

	
 
        cmd = ['git'] + _copts + cmd
 
        if _str_cmd:
 
            cmd = ' '.join(cmd)
 
        try:
 
            opts = dict(
 
                shell=isinstance(cmd, basestring),
 
                stdout=PIPE,
 
                stderr=PIPE,
 
                env=gitenv,
 
            )
 
            if os.path.isdir(self.path):
 
                opts['cwd'] = self.path
 
            p = Popen(cmd, **opts)
 
        except OSError, err:
 
            p = subprocessio.SubprocessIOChunker(cmd, **opts)
 
        except (EnvironmentError, OSError), err:
 
            log.error(traceback.format_exc())
 
            raise RepositoryError("Couldn't run git command (%s).\n"
 
                "Original error was:%s" % (cmd, err))
 
        so, se = p.communicate()
 
        if not se.startswith("fatal: bad default revision 'HEAD'") and \
 
            p.returncode != 0:
 
            raise RepositoryError("Couldn't run git command (%s).\n"
 
                "stderr:\n%s" % (cmd, se))
 

	
 
        so = ''.join(p)
 
        se = None
 
        return so, se
 

	
 
    def _check_url(self, url):
 
        """
 
        Functon will check given url and try to verify if it's a valid
 
        link. Sometimes it may happened that mercurial will issue basic
 
        auth request that can cause whole API to hang when used from python
 
        or other external calls.
 

	
 
        On failures it'll raise urllib2.HTTPError
 
        """
 

	
 
        #TODO: implement this
 
        pass
 

	
 
    def _get_repo(self, create, src_url=None, update_after_clone=False,
 
            bare=False):
 
        if create and os.path.exists(self.path):
 
            raise RepositoryError("Location already exist")
 
        if src_url and not create:
 
            raise RepositoryError("Create should be set to True if src_url is "
 
                                  "given (clone operation creates repository)")
 
        try:
 
            if create and src_url:
 
                self._check_url(src_url)
 
                self.clone(src_url, update_after_clone, bare)
 
                return Repo(self.path)
 
            elif create:
 
                os.mkdir(self.path)
 
                if bare:
 
                    return Repo.init_bare(self.path)
 
                else:
 
                    return Repo.init(self.path)
 
            else:
 
                return Repo(self.path)
 
        except (NotGitRepository, OSError), err:
 
            raise RepositoryError(err)
 

	
 
    def _get_all_revisions(self):
 
        # we must check if this repo is not empty, since later command
 
        # fails if it is. And it's cheaper to ask than throw the subprocess
 
        # errors
 
        try:
 
            self._repo.head()
 
        except KeyError:
 
            return []
 
        cmd = 'rev-list --all --reverse --date-order'
 
        try:
 
            so, se = self.run_git_command(cmd)
 
        except RepositoryError:
 
            # Can be raised for empty repositories
 
            return []
 
        return so.splitlines()
 

	
 
    def _get_all_revisions2(self):
 
        #alternate implementation using dulwich
 
        includes = [x[1][0] for x in self._parsed_refs.iteritems()
 
                    if x[1][1] != 'T']
 
        return [c.commit.id for c in self._repo.get_walker(include=includes)]
 

	
 
    def _get_revision(self, revision):
 
        """
 
        For git backend we always return integer here. This way we ensure
 
        that changset's revision attribute would become integer.
 
        """
 
        pattern = re.compile(r'^[[0-9a-fA-F]{12}|[0-9a-fA-F]{40}]$')
 
        is_bstr = lambda o: isinstance(o, (str, unicode))
 
        is_null = lambda o: len(o) == revision.count('0')
 

	
 
        if len(self.revisions) == 0:
 
            raise EmptyRepositoryError("There are no changesets yet")
 

	
 
        if revision in (None, '', 'tip', 'HEAD', 'head', -1):
 
            revision = self.revisions[-1]
 

	
 
        if ((is_bstr(revision) and revision.isdigit() and len(revision) < 12)
 
            or isinstance(revision, int) or is_null(revision)):
 
            try:
 
                revision = self.revisions[int(revision)]
 
            except:
 
                raise ChangesetDoesNotExistError("Revision %r does not exist "
 
                    "for this repository %s" % (revision, self))
 

	
 
        elif is_bstr(revision):
 
            # get by branch/tag name
 
            _ref_revision = self._parsed_refs.get(revision)
 
            _tags_shas = self.tags.values()
 
            if _ref_revision:  # and _ref_revision[1] in ['H', 'RH', 'T']:
 
                return _ref_revision[0]
 

	
 
            # maybe it's a tag ? we don't have them in self.revisions
 
            elif revision in _tags_shas:
 
                return _tags_shas[_tags_shas.index(revision)]
 

	
 
            elif not pattern.match(revision) or revision not in self.revisions:
 
                raise ChangesetDoesNotExistError("Revision %r does not exist "
 
                    "for this repository %s" % (revision, self))
 

	
 
        # Ensure we return full id
 
        if not pattern.match(str(revision)):
 
            raise ChangesetDoesNotExistError("Given revision %r not recognized"
 
                % revision)
 
        return revision
 

	
 
    def _get_archives(self, archive_name='tip'):
 

	
 
        for i in [('zip', '.zip'), ('gz', '.tar.gz'), ('bz2', '.tar.bz2')]:
 
                yield {"type": i[0], "extension": i[1], "node": archive_name}
 

	
 
    def _get_url(self, url):
 
        """
 
        Returns normalized url. If schema is not given, would fall to
 
        filesystem (``file:///``) schema.
 
        """
 
        url = str(url)
 
        if url != 'default' and not '://' in url:
 
            url = ':///'.join(('file', url))
 
        return url
 

	
 
    @LazyProperty
 
    def name(self):
 
        return os.path.basename(self.path)
 

	
 
    @LazyProperty
 
    def last_change(self):
 
        """
 
        Returns last change made on this repository as datetime object
 
        """
 
        return date_fromtimestamp(self._get_mtime(), makedate()[1])
 

	
 
    def _get_mtime(self):
 
        try:
 
            return time.mktime(self.get_changeset().date.timetuple())
 
        except RepositoryError:
 
            idx_loc = '' if self.bare else '.git'
 
            # fallback to filesystem
 
            in_path = os.path.join(self.path, idx_loc, "index")
 
            he_path = os.path.join(self.path, idx_loc, "HEAD")
 
            if os.path.exists(in_path):
 
                return os.stat(in_path).st_mtime
 
            else:
 
                return os.stat(he_path).st_mtime
0 comments (0 inline, 0 general)