On Mon, Jul 11, 2011 at 05:02:58PM +0200, Landry Breuil wrote: > On Mon, Jul 11, 2011 at 08:51:12AM -0500, Marco Peereboom wrote: > > So the following patch seems to make webkit 1.4.2 mostly usable. > > > > This doesn't fix all issues with epiphany or xxxterm. I have been > > making updates to xxxterm to work around some issues. I started working > > with the upstream guys but they are less than interested because "linux > > works". The core of the issue is that they reset ulimit values despite > > being set properly by the browser. > > Why not fixing that code instead ? Do you have references to the discussion > with upstream ?
http://comments.gmane.org/gmane.os.opendarwin.webkit.gtk/575 That is the public debate. > > > Secondly web sockets keep half open > > connections around longer because they never issue a close and it is > > trivially simple to run out of file descriptors. > > What about adding that close() properly instead, through a proper bug > report ? I have been told this is a function of libsoup + web sockets that can't be fixed. When the remote site sends a close libsoup does (can?) not receive an event so the connection lingers using up a file descriptor. I am unaware of any awesomeness web sockets brings us btw. My non scientific observations seem to confirm that disabling websockets makes the browser a little more snappy. > > You're trying to slip disable-link-prefetch in, wasnt it supposed to be > a pref exposed through the api ? No. It is disabled by default what I am trying to do is to keep it that way. > And what is the other patch ? Is it reported somewhere upstream ? Sent it to the mailing list. Haven't heard back. I am not sure this is an OpenBSD only issue either. FWIW, without this patch crashing webkit is trivially simple. In fact so simple that xxxterm is unusable without and other browsers that have automation built in (it is a race when opening tons of web connections). A warning in the package to up ulimits would also be a good idea. > > Landry >
