On Mon, 24 Jul 2023 at 21:16, Pavel Stehule <pavel.stehule@gmail.com> wrote:
> I don't understand how it can be possible to do it without. I need to process possible errors, and then I need to
readand synchronize protocol. I didn't inject
> this feature to some oher flow, so I need to implement a complete process.
I think I might be missing the reason for this then. Could you explain
a bit more why you didn't inject the feature into another flow?
Because it seems like it would be better to inserting the logic for
handling the new response packet in pqParseInput3(), and then wait on
the result with PQexecFinish(). This would allow sending these
messages in a pipelined mode.
> For example, PQsetClientEncoding does a PQexec call, which is much more expensive.
Yeah, but you'd usually only call that once for the life of the
connection. But honestly it would still be good if there was a
pipelined version of that function.
> Unfortunately, for this feature, I cannot change some local state variables, but I need to change the state of the
server.Own message is necessary, because we don't want to be limited by the current transaction state, and then we
cannotreuse PQexec.
I guess this is your reasoning for why it needs its own state machine,
but I don't think I understand the problem. Could you expand a bit
more on what you mean? Note that different message types use
PQexecFinish to wait for their result, e.g. PQdescribePrepared and
PQclosePrepared use PQexecFinish too and those wait for a
RowDescription and a Close message respectively. I added the logic for
PQclosePerpared recently, that patch might be some helpful example
code: https://github.com/postgres/postgres/commit/28b5726561841556dc3e00ffe26b01a8107ee654
> The API can be changed from PQlinkParameterStatus(PGconn *conn, const char *paramName)
>
> to
>
> PQlinkParameterStatus(PGconn *conn, int nParamNames, const char *paramNames)
That would definitely address the issue with the many round trips
being needed. But it would still leave the issue of introducing a
second state machine in the libpq code. So if it's possible to combine
this new code into the existing state machine, then that seems a lot
better.