GitLab

This fixes the issue of wrongly ignoring new services.

Each service keeps a pointer to its node in the device description doc so
we could easily end-up service accessing an invalid node pointer in
situations where the service in question is not unrefed BEFORE the dispose
of the associated device is run.

This was the cause of rygel crashing if you hit CTRL^C at its initialization.

Check that a GUPnPServiceProxyAction is used with the
GUPnPServiceProxy object that it was created for.

The code that sorts the Accept-Language header by quality leaked
languages under certain circumstances and was overly complex.
Replace it by a simpler implementation that sorts the returned
list instead of trying to sort the string array in place.

This reverts commit 9da38618.

This is no longer needed as the bug in DLNA CTT has been fixed and
hopefully be available in the next release. The issue turned out to that
CTT *was* replying to all the HTTP requests but in one of them, it didn't
set any headers at all and libsoup was still expecting more data on the
socket and hence didn't close the socket since from its POV the socket was
still active.

Although DLNA requirement 7.2.8.7 specifically states that we need to close the
socket if there is no pending requests on that, the DLNA CTT test case for
7.2.8.7 doesn't reply to our request and then wrongly assumes that we should be
closing the socket after 60 seconds even though there is a pending request on
the socket. Astonishingly the same test case for clients does the right thing.

Until the bug is resolved in DLNA CTT, we can temporarily work around the issue
by setting-up a timeout on blocking I/O as well.

This drop the dbus-glib dependency but bumps the gio dependency to 2.26 (latest
stable release) if network-manager is to be used as the context manager. One
"issue" here is that GDBus (and gio in general it seems) simply uses the main
context associated with the current thread's context and do not allow us to set
the main context we are told to use by the application. Its not really an issue
since applications can use g_main_context_push_thread_default() instead of
passing the main context to gupnp, however this is still a fundamental change in
the behavior of gupnp.

The gzip-compression feature needs gio >= 2.24 and
libsoup-2.4 >= 2.28.2.
Signed-off-by: Sven Neumann <s.neumann@raumfeld.com>

Enable transparent decoding of gzip compressed HTTP bodies in
libsoup and let GUPnPServiceProxy set the Accept-Encoding header
to "gzip".

Let GUPnPService check for the Accept-Encoding header and
optionally compress large (> 1024 bytes) repsonse bodies. This
saves some bandwidth and gives a noticeable speedup on slow
connections.

The shared-mime-info that we use to detect mime-type of the contents we
are serving reports "application/xml" for XML, which seems to break
compatibility with some crappy XML parsers (See
https://bugzilla.gnome.org/show_bug.cgi?id=626196#c41). According to UPnP
Device Architecture 1.1, we MUST use the following value:

text/xml; charset="utf-8"

Instead of calling g_hash_table_foreach_remove() with a callback
function that always returns TRUE, just call g_hash_table_remove_all().

Read and write floats and doubles using the '.' as decimal
separator, independent of the locale that is currently in use.

The code used to use atoi() to parse unsigned integer values
from XML strings. Use the proper functions instead.

Avoid 64bit problems if glib hands out GTypes > INT_MAX

Don't use deprecated glib, gobject and libsoup APIs.

Based on a patch by Nader Morshed.

soup_message_headers_get() -> soup_message_headers_get_one().

soup_message_headers_get_content_length() only returns non-0 if the
encoding in use is 'Content-Length'. Although I'm not sure we are
correctly handling 'chunked' encoding but at least now we pass the DLNA
CTT test 7.2.5.7.

Signed-off-by: Sven Neumann <s.neumann@raumfeld.com>

GUPnPService should remove all handlers installed on the
shared SoupServer instance when it is disposed.
Signed-off-by: Sven Neumann <s.neumann@raumfeld.com>