Don't ignore error message on PQ.consumeInput

[steve-chavez]

Hey @diogob,

On PostgREST/postgrest#3313 (reply in thread), a user reports the LISTEN channel is getting killed and we're getting a generic error message from hasql-notifications which comes from:

hasql-notifications/src/Hasql/Notifications.hs

Line 167 in b072182

Nothing -> panic "Error checking for PostgreSQL notifications"

There might be more details on PQ.consumeInput, since according to pg docs:

PQconsumeInput normally returns 1 indicating “no error”, but returns 0 if there was some kind of trouble (in which case PQerrorMessage can be consulted).

However PQ.consumeInput result is getting ignored on:

hasql-notifications/src/Hasql/Notifications.hs

Line 175 in b072182

void $ PQ.consumeInput pqCon

[diogob]

Hi @steve-chavez , would a panic with some custom message work on your side? I can get the string from PQerrorMessage and default to the generic one when no string is available (comes inside a Maybe).

It could be the case that the call to PQsocket is failing during the connection setup (I think that panic should not be there, perhaps just using a delay will fix the issue).

I'll work on a PR to fix this.

[diogob]

@steve-chavez I just put together a quick attempt to address the issue on #21 but haven't tested it. Are you able to use the version from that branch and verify if the behaviour change would satisfy your case?

[diogob]

@steve-chavez just published 0.2.1.0