Обсуждение: unknown libpq service entries ignored
When using the "service" parameter in a libpq connection (e.g., psql service=foo), and the service name is not defined in pg_service.conf, then the setting is silently ignored and the connection proceeds with whatever other settings and defaults apply. That does not look very robust. Shouldn't there be an error when a specified service name does not exist?
Peter Eisentraut wrote: > When using the "service" parameter in a libpq connection (e.g., psql > service=foo), and the service name is not defined in pg_service.conf, > then the setting is silently ignored and the connection proceeds with > whatever other settings and defaults apply. That does not look very > robust. Shouldn't there be an error when a specified service name does > not exist? I agree. It leads to strange error messages at least (I didn't want to connect to port 5432 on localhost?!?), and you could end up connecting to a different database than you wanted without noticing it. Yours, Laurenz Albe
On tor, 2009-11-26 at 10:57 +0200, Peter Eisentraut wrote: > When using the "service" parameter in a libpq connection (e.g., psql > service=foo), and the service name is not defined in pg_service.conf, > then the setting is silently ignored and the connection proceeds with > whatever other settings and defaults apply. That does not look very > robust. Shouldn't there be an error when a specified service name does > not exist? Possible patch for this issue.
Вложения
Peter Eisentraut <peter_e@gmx.net> writes: > + printfPQExpBuffer(errorMessage, > + libpq_gettext("ERROR: service \"%s\" not found\n"), service); Please make the message consistent with the rest of libpq. AFAICS none of the other messages in that file use an ERROR: prefix. regards, tom lane
On fre, 2009-11-27 at 10:22 -0500, Tom Lane wrote: > Peter Eisentraut <peter_e@gmx.net> writes: > > + printfPQExpBuffer(errorMessage, > > + libpq_gettext("ERROR: service \"%s\" not found\n"), service); > > Please make the message consistent with the rest of libpq. AFAICS none > of the other messages in that file use an ERROR: prefix. Well, that style is used four times in that same function, but no where else in the file. So the existing uses should probably also be cleaned up.