Change ConnectInfo to work with Maybe values

[sol]

Do you think it would be a good idea to change ConnectInfo to work with Maybe values, and get rid of the special treatment of "" and negative numbers in postgreSQLConnectionString? I'm asking, because I'm not sure if I like the code for yesodweb/persistent#38.

[lpsmith]

Well, to be honest in my own code I'm mostly using connectPostgreSQL, which is publicly exported and allows you to supply any postgresql connection string you want. The only code of mine that uses ConnectInfo is code that's been ported from Chris Done's pgsql-simple. ConnectInfo is taken from that, which is in turn a subset of the ConnectInfo record from mysql-simple.

I'm not necessarily opposed to changing the interface from the other *-simples, but I'd like to understand what I'm getting in return. In particular, I don't think that special-casing negative port numbers is that big of a deal, as port numbers must be positive. As for the rest, can you demonstrate a case where say, password='' leads to different behavior than omitting that argument altogether?

[lpsmith]

To be honest, I don't know what persistent-postgresql looked like before the port to postgresql-simple, but I would think it would be preferable to allow the user to supply their own connection string. This would expose the full range of connection and authentication options, such as SSL and GSSAPI, instead of the limited subset currently achievable through the ConnectInfo record.

[lpsmith]

What I think would make more sense here is to process the connection configuration object into a PostgreSQL connection string directly. Basically, turn every (key,value) pair in the JSON Object into a key='value' pair in the PostgreSQL connection string, with the values appropriately escaped of course. The one caveat is that if you are committed to supporting the database key, then you'll need to translate that to dbname. You can look at the libpq connection option documentation for more info.

If you decide to go this route, another possibility would be to extend the postgresql-libpq bindings to support PQconnectdbParams, which would avoid the need to build up the connection string and quoting, etc.

[sol]

Regarding your first reply:

My point about special-casing of negative port numbers / empty strings is that I only understand that special-casing if I look at the implementation. If I see a Maybe than it is immediately clear to me that this thing is optional.

Regarding your second reply:

I agree that it would be advantageous to use a plain connection string.

Regarding your third reply:

Actually the connection string was build up there, before it was ported to postgresql-simple. I'm not sure why this was changed. Porting persistent-postgresql to use connectPostgreSQL would be fine with me.

[sol]

As a side note: "dbname=foo user=" and "dbname=foo" give you different results. The first will not use the default, but fail.

[lpsmith]

Well, I'm going to close this issue, but feel free to reopen it if you aren't satisfied yet.