Some bugs have been found in ASAN builds using a compiled-in
malloc-only memory context  (see commits 54ab74865, 1fe5a347e,
ccc8194e4 for the bugs fixed). The patch used was first mentioned here
(attachment: malloc_allocator.patch):

https://www.postgresql.org/message-id/[email protected]

Do we want something like this in core? I don't think we'd need the
ENABLE_ASAN symbol, since that seems orthogonal -- we can configure
ASAN separately how we wish, and I don't see why we'd need to tie the
GUC default for the special context to whether we use ASAN. Aside from
that, the basic idea seems useful.

-- 
John Naylor
Amazon Web Services


Reply via email to