Re: Reducing the chunk header sizes on all memory context types

Tomas Vondra <tomas.vondra@enterprisedb.com> writes:
> I can reproduce it on my system (rpi4 running 32-bit raspbian).

Yeah, more or less same as what I'm testing on.

Seeing that the complaint is about pfree'ing a non-maxaligned
ReorderBufferChange pointer, I tried adding

diff --git a/src/backend/replication/logical/reorderbuffer.c b/src/backend/replication/logical/reorderbuffer.c
index 89cf9f9389..dfa9b6c9ee 100644
--- a/src/backend/replication/logical/reorderbuffer.c
+++ b/src/backend/replication/logical/reorderbuffer.c
@@ -472,6 +472,8 @@ ReorderBufferGetChange(ReorderBuffer *rb)
        change = (ReorderBufferChange *)
                MemoryContextAlloc(rb->change_context, sizeof(ReorderBufferChange));

+       Assert(change == (void *) MAXALIGN(change));
+
        memset(change, 0, sizeof(ReorderBufferChange));
        return change;
 }

and this failed!

(gdb) f 3
#3  0x003cb888 in ReorderBufferGetChange (rb=0x24ed820) at reorderbuffer.c:475
475             Assert(change == (void *) MAXALIGN(change));
(gdb) p change
$1 = (ReorderBufferChange *) 0x24aaa14

So the bug is in fact in David's changes, and it consists in palloc
sometimes handing back non-maxaligned pointers.  I find it mildly
astonishing that we managed to get through core regression tests
without such a fault surfacing, but there you have it.

This machine has MAXALIGN 8 but 4-byte pointers, so there's something
wrong with that situation.

            regards, tom lane