Commits on Dec 6, 2010
  1. README update

    jonashaag committed Dec 6, 2010
  2. Add bjoern.version attribute

    jonashaag committed Dec 6, 2010
  3. Make bjoern PyPI ready

    jonashaag committed Dec 6, 2010
Commits on Nov 20, 2010
Commits on Nov 3, 2010
  1. Use libev's default loop

    jonashaag committed Nov 3, 2010
Commits on Nov 2, 2010
  1. fix checkout for http-parser

    kkris committed Nov 2, 2010
Commits on Oct 25, 2010
  1. http-parser is now a submodule

    jonashaag committed Oct 25, 2010
Commits on Oct 24, 2010
  1. Make non-gcc happy

    jonashaag committed Oct 24, 2010
  2. Fixed multiple-calls-to-start_response reference count bug.

    jonashaag committed Oct 24, 2010
    When doing multiple calls to start_response without the 'exc_info'
    argument, bjoern updated the 'request->{status,headers}' fields but did
    not INCREF them. The correct behaviour is to not update them, which was
    implemented in this patch.
  3. Cleanup

    jonashaag committed Oct 24, 2010
Commits on Oct 22, 2010
  1. make gcc happy

    jonashaag committed Oct 22, 2010
Commits on Oct 17, 2010
  1. docstring

    jonashaag committed Oct 17, 2010
Commits on Oct 15, 2010
  1. Fixed Control+C: bjoern now calls PyErr_SetInterrupt() on SIGINT whic…

    jonashaag committed Oct 15, 2010
    …h shuts down the current interpreter thread
Commits on Oct 10, 2010
  1. improve kkris' changes

    jonashaag committed Oct 10, 2010
Commits on Oct 9, 2010
  1. strip off extra output

    kkris committed Oct 9, 2010
  2. added

    kkris committed Oct 9, 2010
Commits on Oct 8, 2010
Commits on Oct 6, 2010
Commits on Oct 5, 2010
  1. Fixed some memory leaks and crashes related to exceptions in the WSGI

    jonashaag committed Oct 5, 2010
    - DECREF old status and headers on second call of 'start_response'
    - INCREF items of the exc_info tuple passed as 3rd argument to
      'start_response' if that exception is restored. (cleaning up
      exceptions/setting an exception DECREFs the old one)
  2. Fixed `start_response` object deallocation bug: I shouldn't call

    jonashaag committed Oct 5, 2010
    `PyObject_FREE` manually but use `Py_DECREF`.
    If the WSGI application, called by `wsgi_call_application`, keeps a
    reference to `start_response` (passed as argument), the `start_response`
    object must not be deallocated before the WSGI application is, because
    otherwise that would lead into a `Py_DECREF` on a deallocated object.
    Additionally, every type has to be intialized with `PyType_Ready(type)`
    to fill skipped fields (that are set to NULL by the compiler);
    otherwise, doing unexpected operations with objects of that type (for
    example, trying to create a new instance of their type) will end in
    segmentation faults or memory corruption.
    See this thread in the comp.lang.python group/mailing list for more:
Commits on Oct 2, 2010
  1. Make DBG possible anywhere.

    jonashaag committed Oct 2, 2010
Commits on Sep 10, 2010
  1. foo

    jonashaag committed Sep 10, 2010