Hi,
I'm seeing several significant memory leaks in MesaCVS-Sep05 at the
`repeat: create context, draw, destroy context' level.
The most significant ones seems to be from gl_immediate_alloc().
I haven't yet managed to unterstand the refcounting for
`struct immediate's. However, the appended patch for context.c
seems to be necessary in any case:
--- context.c 1999/09/09 21:14:50 1.1
+++ context.c 1999/09/11 11:48:09
@@ -1355,7 +1355,11 @@
free( ctx->PB );
- free( ctx->VB );
+
+ if (ctx->input != ctx->VB->IM)
+ gl_immediate_free(ctx->input);
+
+ gl_vb_free( ctx->VB );
ctx->Shared->RefCount--;
assert(ctx->Shared->RefCount>=0);
@@ -1412,6 +1416,14 @@
free( ctx->EvalMap.Map2Texture3.Points );
if (ctx->EvalMap.Map2Texture4.Points)
free( ctx->EvalMap.Map2Texture4.Points );
+
+ /* Free cache of immediate buffers. */
+ while (ctx->nr_im_queued-- > 0) {
+ struct immediate* next = ctx->freed_im_queue->next;
+ /*printf("free_im(%p)\n", ctx->freed_im_queue);*/
+ free(ctx->freed_im_queue);
+ ctx->freed_im_queue = next;
+ }
free( (void *) ctx );
There is another possibility for struct immediate to leak when it is
put in a compiled display list. The following is obviously _not_ the
correct fix, the refcounting should be fixed instead:
--- dlist.c 1999/09/11 09:16:42 1.1
+++ dlist.c 1999/09/11 11:34:37
@@ -348,11 +348,14 @@
while (!done) {
switch (n[0].opcode) {
/* special cases first */
- case OPCODE_VERTEX_CASSETTE:
- if ( ! -- ((struct immediate *) n[1].data)->ref_count )
- gl_immediate_free( (struct immediate *) n[1].data );
+ case OPCODE_VERTEX_CASSETTE: {
+ struct immediate* im = (struct immediate *) n[1].data;
+ if ( 1 || ! -- im->ref_count )
+ gl_immediate_free( im );
n += InstSize[n[0].opcode];
- break;
+ break; }
case OPCODE_MAP1:
gl_free_control_points( ctx, n[1].e, (GLfloat *) n[6].data );
n += InstSize[n[0].opcode];
I'll keep looking to understand the refcounting, but others may
be quicker. I'll also report the more minor leaks once this is
fixed.
Regards,
Wolfram.
--
`Surf the sea, not double-u three...'
[EMAIL PROTECTED]
_______________________________________________
Mesa-dev maillist - [EMAIL PROTECTED]
http://lists.mesa3d.org/mailman/listinfo/mesa-dev
