Connection string parsing corner case

[johto]

We currently don't fail on a connection string like this: "postgre://marko@internet". From a quick scan of the code it looks like this is because we parse the connection string the same way as libpq does, but we miss the last step: if the connection string is one big blob, libpq uses it as the database name:

psql "postgre://marko@internet"
psql: FATAL:  database "postgre://marko@internet" does not exist

Silently ignoring the blob seems like a bad idea, but I'm not sure we want to use it as the database name either. Any thoughts?

[kisielk]

Interesting. I tried to replicate the libpq parsing behaviour as closely as possible but missed that bit. So you're saying that lib/pq currently silently ignores the connection string? or what happens?

[johto]

Interesting. I tried to replicate the libpq parsing behaviour as closely as possible but missed that bit. So you're saying that lib/pq currently silently ignores the connection string?

Correct.

[kisielk]

The behaviour in libpq is to ignore the trailing unused parts of the options string, as covered by the test cases here:

https://github.com/lib/pq/blob/master/conn_test.go#L685

I guess if there have been no options parsed at all, they just use that as the database name.

We could either reproduce that behaviour, or change this to be an error.

[johto]

Ugh. I need to stop using psql to test libpq behaviour.

It looks like libpq actually does this:

% ./libpq "postgre://marko@internet"
Connection to database failed: missing "=" after "postgre://marko@internet" in connection info string

Also, the test you pointed at seems to not work in libpq:

% ./libpq "user=foo blah"
Connection to database failed: missing "=" after "blah" in connection info string

[kisielk]

hm ok, well that needs to be fixed, but it's good to see that libpq doesn't do anything weird.

[johto]

hm ok, well that needs to be fixed, but it's good to see that libpq doesn't do anything weird.

Right. It's psql doing the weirdness in that case. Sorry about the confusion.

[kisielk]

Ok, I think I figured out the problem and have some changes ready. There's just a couple more things I wanted to compare. What is this libpq program you are using to test? Just something you hooked up, or is it available in the sources? If it's something you put together, can you share it in a gist?

[johto]

I just took example #1 from the libpq examples and removed all the unnecessary parts: https://gist.github.com/johto/6923085