On Mon, 17 Mar 2014, Donovan Watteau wrote:
> Hi,
>
> rev 1.76 of gnome.port.mk made it possible to build a full Xfce desktop
> on loongson (yay!). Unfortunately, Thunar crashes at startup with a
> SIGBUS.
>
> I do have the latest glib2-2.38.2p6 update. thunar-1.6.3 and
> gvfs-1.18.3p5 were rebuilt with DEBUG=-g.
>
> Please let me know if I should report this to upstream instead.
>
> [snip]
Some progress...
Here's a hack found in upstream's bugzilla. Works for me. But it's
a hack.
Only tested on loongson, don't have access to any other arch ATM.
Index: Makefile
===================================================================
RCS file: /cvs/ports/x11/gnome/gvfs/Makefile,v
retrieving revision 1.112
diff -u -p -r1.112 Makefile
--- Makefile 9 Mar 2014 19:46:35 -0000 1.112
+++ Makefile 18 Mar 2014 19:29:59 -0000
@@ -11,7 +11,7 @@ PKGNAME-main= ${DISTNAME}
PKGNAME-smb= ${GNOME_PROJECT}-smb-${GNOME_VERSION}
PKGNAME-goa= ${GNOME_PROJECT}-goa-${GNOME_VERSION}
-REVISION-main= 5
+REVISION-main= 6
SHARED_LIBS += gvfscommon 4.0 # unknown
Index: patches/patch-metadata_metatree_c
===================================================================
RCS file: patches/patch-metadata_metatree_c
diff -N patches/patch-metadata_metatree_c
--- /dev/null 1 Jan 1970 00:00:00 -0000
+++ patches/patch-metadata_metatree_c 18 Mar 2014 19:29:59 -0000
@@ -0,0 +1,54 @@
+$OpenBSD$
+
+XXX Hack for unaligned access
+
+https://bugzilla.gnome.org/show_bug.cgi?id=726456
+
+--- metadata/metatree.c.orig Thu Nov 7 14:40:18 2013
++++ metadata/metatree.c Tue Mar 18 19:43:37 2014
+@@ -132,6 +132,27 @@ struct _MetaTree {
+ MetaJournal *journal;
+ };
+
++/* Unfortunately the journal entries are only aligned to 32 bit boundaries
++ but on some 64-bit RISC architectures (e.g. Alpha) this is insufficient
++ to guarantee correct alignment of 64-bit accesses. This is not a show
++ stopper but does cause inefficient traps to the kernel and pollution of
++ kernel logs. Rather than fix the alignment we provide a helper function,
++ dependent on features specific to gcc, to correctly access a 64-bit datum
++ that may be misaligned. This causes no loss of inefficiency on
++ architectures where alignment is not an issue provided we are compiled
++ with compiler optimisation turned on. */
++#ifdef __GNUC__
++struct una_u64 { guint64 x __attribute__((packed)); };
++static inline guint64 ldq_u(guint64 *p)
++{
++ const struct una_u64 *ptr = (const struct una_u64 *) p;
++ return ptr->x;
++}
++#else
++#define ldq_u(x) (*(x))
++#endif
++
++
+ static void meta_tree_refresh_locked (MetaTree *tree,
+ gboolean force_reread);
+ static MetaJournal *meta_journal_open (MetaTree *tree,
+@@ -1230,7 +1251,7 @@ meta_journal_iterate (MetaJournal *journal,
+ break;
+ }
+
+- mtime = GUINT64_FROM_BE (entry->mtime);
++ mtime = GUINT64_FROM_BE( ldq_u(&(entry->mtime)) );
+ journal_path = &entry->path[0];
+
+ if (journal_entry_is_key_type (entry) &&
+@@ -2214,7 +2235,7 @@ apply_journal_to_builder (MetaTree *tree,
+ entry = journal->first_entry;
+ while (entry < journal->last_entry)
+ {
+- mtime = GUINT64_FROM_BE (entry->mtime);
++ mtime = GUINT64_FROM_BE( ldq_u(&(entry->mtime)) );
+ journal_path = &entry->path[0];
+
+ switch (entry->entry_type)
