Changeset - 08d439bfbd8c
[Not reviewed]
beta
0 2 0
Marcin Kuzminski - 13 years ago 2013-05-09 13:29:03
marcin@python-works.com
fixed handling shell argument in subprocess calls, it always was hardcoded even when passed properly in arguments
2 files changed with 4 insertions and 3 deletions:
0 comments (0 inline, 0 general)
rhodecode/lib/vcs/backends/git/repository.py
Show inline comments
 
@@ -32,196 +32,197 @@ from rhodecode.lib.vcs.exceptions import
 
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, get_user_home
 

	
 
from rhodecode.lib.vcs.utils.hgcompat import (
 
    hg_url, httpbasicauthhandler, httpdigestauthhandler
 
)
 

	
 
from .changeset import GitChangeset
 
from .config import ConfigFile
 
from .inmemory import GitInMemoryChangeset
 
from .workdir import GitWorkdir
 

	
 
SHA_PATTERN = re.compile(r'^[[0-9a-fA-F]{12}|[0-9a-fA-F]{40}]$')
 

	
 
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)
 
        repo = self._get_repo(create, src_url, update_after_clone, bare)
 
        self.bare = repo.bare
 

	
 
    @property
 
    def _config_files(self):
 
        return [
 
            self.bare and abspath(self.path, 'config')
 
                      or abspath(self.path, '.git', 'config'),
 
             abspath(get_user_home(), '.gitconfig'),
 
         ]
 

	
 
    @property
 
    def _repo(self):
 
        return Repo(self.path)
 

	
 
    @property
 
    def head(self):
 
        try:
 
            return self._repo.head()
 
        except KeyError:
 
            return None
 

	
 
    @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()
 

	
 
    @classmethod
 
    def _run_git_command(cls, cmd, **opts):
 
        """
 
        Runs given ``cmd`` as git command and returns tuple
 
        (stdout, stderr).
 

	
 
        :param cmd: git command to be executed
 
        :param opts: env options to pass into Subprocess command
 
        """
 

	
 
        if '_bare' in opts:
 
            _copts = []
 
            del opts['_bare']
 
        else:
 
            _copts = ['-c', 'core.quotepath=false', ]
 
        safe_call = False
 
        if '_safe' in opts:
 
            #no exc on failure
 
            del opts['_safe']
 
            safe_call = True
 

	
 
        _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'
 

	
 
        _git_path = settings.GIT_EXECUTABLE_PATH
 
        cmd = [_git_path] + _copts + cmd
 
        if _str_cmd:
 
            cmd = ' '.join(cmd)
 

	
 
        try:
 
            _opts = dict(
 
                env=gitenv,
 
                shell=False,
 
                shell=True,
 
            )
 
            _opts.update(opts)
 
            p = subprocessio.SubprocessIOChunker(cmd, **_opts)
 
        except (EnvironmentError, OSError), err:
 
            tb_err = ("Couldn't run git command (%s).\n"
 
                      "Original error was:%s\n" % (cmd, err))
 
            log.error(tb_err)
 
            if safe_call:
 
                return '', err
 
            else:
 
                raise RepositoryError(tb_err)
 

	
 
        return ''.join(p.output), ''.join(p.error)
 

	
 
    def run_git_command(self, cmd):
 
        opts = {}
 
        if os.path.isdir(self.path):
 
            opts['cwd'] = self.path
 
        return self._run_git_command(cmd, **opts)
 

	
 
    @classmethod
 
    def _check_url(cls, 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
 
        """
 

	
 
        # check first if it's not an local url
 
        if os.path.isdir(url) or url.startswith('file:'):
 
            return True
 

	
 
        if('+' in url[:url.find('://')]):
 
            url = url[url.find('+') + 1:]
 

	
 
        handlers = []
 
        test_uri, authinfo = hg_url(url).authinfo()
 
        if not test_uri.endswith('info/refs'):
 
            test_uri = test_uri.rstrip('/') + '/info/refs'
 
        if authinfo:
 
            #create a password manager
 
            passmgr = urllib2.HTTPPasswordMgrWithDefaultRealm()
 
            passmgr.add_password(*authinfo)
 

	
 
            handlers.extend((httpbasicauthhandler(passmgr),
 
                             httpdigestauthhandler(passmgr)))
 

	
 
        o = urllib2.build_opener(*handlers)
 
        o.addheaders = [('User-Agent', 'git/1.7.8.0')]  # fake some git
 

	
 
        q = {"service": 'git-upload-pack'}
 
        qs = '?%s' % urllib.urlencode(q)
 
        cu = "%s%s" % (test_uri, qs)
 
        req = urllib2.Request(cu, None, {})
 

	
 
        try:
 
            resp = o.open(req)
 
            return resp.code == 200
 
        except Exception, e:
 
            # means it cannot be cloned
 
            raise urllib2.URLError("[%s] %s" % (url, e))
 

	
 
    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:
 
                GitRepository._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 self._repo
 
        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 []
rhodecode/lib/vcs/subprocessio.py
Show inline comments
 
@@ -249,167 +249,167 @@ 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
 

	
 
        _shell = kwargs.get('shell', True)
 
        if isinstance(cmd, (list, tuple)):
 
            cmd = ' '.join(cmd)
 

	
 
        _shell = kwargs.get('shell') or True
 
        kwargs['shell'] = _shell
 
        _p = subprocess.Popen(cmd,
 
            bufsize=-1,
 
            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 = '%s' % ''.join(bg_err)
 
            if err:
 
                raise EnvironmentError("Subprocess exited due to an error:\n" + err)
 
            raise EnvironmentError("Subprocess exited with non 0 ret code:%s" % _returncode)
 

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

	
 
    def __iter__(self):
 
        return self
 

	
 
    def next(self):
 
        if self.process.poll():
 
            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()
0 comments (0 inline, 0 general)