I can use PQcmdTuples to simulate it. If PQcmdTuples returns 0, I run
the command, if not, I give a error message. ok?

There use to be a readonly=(0|1) option, but it seems postgresql does
not support for it any more.
---------

PQcmdTuples Returns the number of rows affected by the SQL command.

char * PQcmdTuples(const PGresult *res);

If the SQL command that generated the PGresult was INSERT, UPDATE or
DELETE, this returns a string containing the number of rows affected.
If the command was anything else, it returns the empty string.

no, that won't work, since PQcmdTuples works AFTER the command was
executed. So, it is not the solution. We want to check if it is a
read-only command or not before executing it.

For instance, if there were a 'DROP ALL TABLES' command, using
PQcmdTuples you would execute it, and then return the error, which
will cause all the tables to be dropped. In that case, we must stop
the execution of the command before it gets sent to the DB server.

Well, as the connection must be read-only, my plan was to forbid
transactions and issue a BEGIN before actually running the command,
then call to PQCmdTuples () and if no tuples, ok. Else ROLLBACK.

But as you say, that won't work with DROP...

I'm not in charge of this anymore...

PostgreSQL can have read-only transactions:
http://www.postgresql.org/docs/9.1/static/sql-set-transaction.html

And you can set this as the default for all transactions in the connection:
http://www.postgresql.org/docs/9.1/static/runtime-config-client.html#GUC-DEFAULT-TRANSACTION-READ-ONLY

However, I guess that is only useful if you make sure that everything happens in a transaction.

You might also just make sure that the connection's user has only read access, though maybe that is not generically possible from gnumeric.

Anyway, we should probably close this bug report as obsolete. Please reopen if someone is still interested.