Review of attachment 219840 [details] [review]:

everything except the gvdb symbols leakage looks good to me; this stuff should be caught by the abicheck.sh script during 'make check' on release, so it may be good to investigate what failed there.

::: gio/gio.symbols
@@ +1754,3 @@
+gvdb_table_write_contents
+gvdb_hash_table_insert_string
+gvdb_item_set_hash_table

gvdb is private and internal, and its symbols are annotated by G_GNUC_INTERNAL as well.

Comment on attachment 219840 [details] [review]
win32: More symbols to be exported from gio

Those symbols (other than the gvdb ones) are all already in gio.symbols. There must be some bug in the Makefile rule that generates gio.def from that?

(In reply to comment #1)
> Review of attachment 219840 [details] [review]:
> 
> everything except the gvdb symbols leakage looks good to me; 
e.g. gvdb_table_write_contents is used by glib-compile-resources.c - how is that supposed to work without exporting?

(In reply to comment #2)
> Those symbols (other than the gvdb ones) are all already in gio.symbols. There
> must be some bug in the Makefile rule that generates gio.def from that?
Good catch, probably some (missing) merge conflict on my side.

(In reply to comment #3)
> (In reply to comment #1)
> > Review of attachment 219840 [details] [review] [details]:
> > 
> > everything except the gvdb symbols leakage looks good to me; 
> e.g. gvdb_table_write_contents is used by glib-compile-resources.c - how is
> that supposed to work without exporting?

glib-compile-resources.c is compiled against gvdb explicitly, so it's not going through the internal symbols of libgio; this works because the visibility of the symbols is kept internal.

Thanks. I've updated my build files accordingly which leaves this not-a-bug.