Allow `[` and `]` in the query component about modern-uri

[kirelagin]

I have to admit that the first part of my text ignores the elephant in the room: RFC 3986 was obsolete by WHATWG URL and the latter very explicitly prohibits [ and ] in the query component.

But the second part still stands and, I imagine, there are plenty of people who are not aware of the fact that their square brackets are now definitely illegal :/.

[kirelagin]

One alternative would be to redefine “modern URI” to mean “modern URL according to WHATWG”. Then disallowing the square brackets is the right behaviour indeed, but I am not sure if this library is actually in compliance. E.g. the parsing algorithms is almost certainly not. Here is the basic URL parser specification for reference.

[mrkkrp]

What do other packages do? Let's say network-uri and uri-bytestring?

[kirelagin]

I believe, network-uri implements the RFC naively (i.e. fails on []).

uri-bytestring is kinda weird in that it only parses &-separated key=value pairs, BUT it uses a customisable predicate for testing the validity of characters and ships two predefined tests, with laxURIParserOptions allowing unencoded brackets in queries.

[kirelagin]

(Here is the commit that implements this: Soostone/uri-bytestring@b893032.)

[kirelagin]

I realise this issue is a bit cumbersome, since I was doing the research as I was writing it and changed my opinion on the matter a couple of times in the process.

Currently, I think there are two independent actionable items:

• To justify “modern” in the name of this package, update it to follow WHATWG URL instead of the ancient RFC (this might or might not require renaming the package, since the new standard deprecates the term URI 🙂). As a side-effect, this will make it easier to motivate not allowing [].
• To account for the state of the real world, adjust the interface to allow the user to override the behaviour that disallows [], e.g. a function that returns m (Either Url Url) so that the users have a way of supporting legacy inputs with unencoded brackets but, at the same time, are forced to explicitly admit that this is happening and maybe warn their users.