Hooks on errors

[cbandy]

Is there a particular reason hooks are not called when the underlying driver returns an error?

I can't quite think of a use case for statistics, but I can imagine logging all database interactions.

[cgilling]

Hmmm thats a good point. I can see that being useful to be able to get that information.

So the question is how much context would we want around these errors? We could have a generic error callback for all errors that occur? Or we would want an error callback for connections, statements, transactions, queries, and execs all separately.

[cbandy]

I think we would want the highest resolution possible at the core. Other wrappers could aggregate or filter as desired.

My initial thought was that each hook take an error argument, but the names don't sound quite right. For example when Open returns an error, no "Conn [was] Opened."

[cgilling]

Sorry been really busy lately.

Having more granularity at the core does seem like the way to go. I was considering having a separate ErrorHook interface that could optionally be implemented by a hook have error calls corresponding to the various Hook calls.

I do like the suggestion of having it just be part of the various Hook calls already in place but it does bring in the little bit of strangeness that you are talking about. I think I might be able to live with that if it gives an easy way to return the errors when they occur.

Any thoughts about having a second ErrorHook interface?

Also I'm wondering about if we want to do anything special to handle ErrBadConn or just leave that up to the various implementors. I'm assuming most of the time those would want to be handled separately from other errors because its not so much something wrong with the Query or Transaction, but some network problem or timeout or something on the server side of things.

[cbandy]

I was considering having a separate ErrorHook interface that could optionally be implemented by a hook have error calls corresponding to the various Hook calls.

Another interface sounds reasonable. Rather than "optionally implemented," there would need to be another Add method and another list, I think.

Also I'm wondering about if we want to do anything special to handle ErrBadConn or just leave that up to the various implementors.

I don't have any thoughts on this at the moment.

[cgilling]

I implemented my proposed change for this in #4