Re: [PATCH] psql: tab completion for ALTER ROLE ... IN DATABASE ...

Hi,

Thanks for taking a closer look — that’s a fair question.

The guard is not meant to protect exec_query(), but rather the calls to
PQescapeLiteral(), which are executed before COMPLETE_WITH_QUERY_PLUS
and therefore before exec_query() is reached.

In this code path we construct the query string by calling
PQescapeLiteral(pset.db, ...) directly, and that function assumes a
non-NULL, valid PGconn. If pset.db is NULL or the connection is not in
CONNECTION_OK state, calling PQescapeLiteral() would be unsafe and could
lead to a crash before exec_query() has a chance to handle anything.

The fallback to COMPLETE_WITH("ALL") is intended as a safe default, matching
existing RESET completion behavior when no catalog information can be
retrieved.

That said, I agree it’s worth double-checking whether pset.db is always
guaranteed to be valid in this context, and I’m happy to refine or document
this further if you think a different approach would be better.

Thanks for flagging this .

Regards,
Vasuki M
C-DAC,Chennai