Accidental setting of XLogReaderState.private_data ?
StartupDecodingContext() initializes ctx->reader->private_data with ctx, and
it even does so twice. I couldn't find a place in the code where the
(LogicalDecodingContext *) pointer is retrieved from the reader, and a simple
test of logical replication works if the patch below is applied. Thus I assume
that assignment is a thinko, isn't it?
--
Antonin Houska
Web: https://www.cybertec-postgresql.com
Attachments:
xlogreader_private_data.difftext/x-diffDownload+1-3
Antonin Houska <ah@cybertec.at> writes:
StartupDecodingContext() initializes ctx->reader->private_data with ctx, and
it even does so twice. I couldn't find a place in the code where the
(LogicalDecodingContext *) pointer is retrieved from the reader, and a simple
test of logical replication works if the patch below is applied. Thus I assume
that assignment is a thinko, isn't it?
Hmm. The second, duplicate assignment is surely pointless, but I think
that setting the ctx as the private_data is a good idea. It hardly seems
out of the question that it might be needed in future.
regards, tom lane
On Mon, Apr 15, 2019 at 11:06:18AM -0400, Tom Lane wrote:
Hmm. The second, duplicate assignment is surely pointless, but I think
that setting the ctx as the private_data is a good idea. It hardly seems
out of the question that it might be needed in future.
Agreed that we should keep the assignment done with
XLogReaderAllocate(). I have committed the patch which removes the
useless assignment though.
--
Michael