Hi, I've been thinking about this problem:
http://www.mail-archive.com/[email protected]/msg00065.html An easy, simple and foolproof way to fix this would be to make all oneways block while sending data(*). Another fix would be to put a ref on the connection if the send is a oneway, but that fix will be more involved and not as straightforward. Please comment. Thanks, jules (*) It doesn't look like link_connection_writev() even looks at the link options before eventually returning LINK_IO_QUEUED_DATA at line 1180... Index: src/orb/GIOP/giop-send-buffer.c =================================================================== --- src/orb/GIOP/giop-send-buffer.c (revision 2025) +++ src/orb/GIOP/giop-send-buffer.c (working copy) @@ -445,11 +445,15 @@ gboolean blocking) { int retval; + gboolean oneway; LinkConnection *lcnx = LINK_CONNECTION (cnx); - static LinkWriteOpts *non_block = NULL; + static LinkWriteOpts opts; - if (!non_block) - non_block = link_write_options_new (FALSE); + oneway = giop_send_buffer_is_oneway (buf); + if (oneway || blocking) + opts.block_on_write = TRUE; + else + opts.block_on_write = FALSE; /* FIXME: if a FRAGMENT, assert the 8 byte tail align, &&|| giop_send_buffer_align (buf, 8); */ @@ -457,16 +461,16 @@ if (g_thread_supported () && lcnx->timeout_msec && !lcnx->timeout_source_id - && !giop_send_buffer_is_oneway (buf)) { + && !oneway) { giop_timeout_add (cnx); } + retval = link_connection_writev (lcnx, buf->iovecs, buf->num_used, - blocking ? NULL : non_block); - - if (!blocking && retval == LINK_IO_QUEUED_DATA) + &opts); + if (!opts.block_on_write && (retval == LINK_IO_QUEUED_DATA)) retval = 0; /* FIXME: we need to flag the connection disconnected on fatal error */ Index: ChangeLog =================================================================== --- ChangeLog (revision 2029) +++ ChangeLog (working copy) @@ -1,3 +1,10 @@ +2007-09-26 Jules Colding <[EMAIL PROTECTED]> + + * src/orb/GIOP/giop-send-buffer.c (giop_send_buffer_write): + Make oneway conenctions block until all data has been send. Will + fix this: + http://www.mail-archive.com/[email protected]/msg00065.html + 2007-09-25 Jules Colding <[EMAIL PROTECTED]> * test/timeout-server.c (main): Do not write more into the _______________________________________________ orbit-list mailing list [email protected] http://mail.gnome.org/mailman/listinfo/orbit-list
