IO.read should not gobble errors silently

[avsm]

IO.read currently returns an option, with no ability to comment on the sort of error it's returning. This is a fairly big revision, but it needs to move to a model where errors and end-of-files are distinguished.

See #207

[Chris00]

What is the preferred scheme for error handling? Using something like Or_error, using Lwt.fail (nor really portable to async), polymorphic variants,…?

[talex5]

Currently undecided, but I'm collecting comments and proposals here: mirage/mirage-www#274

(also, there's lots of discussion on the mailing list at present)

[avsm]

Gobbling errors like is is really, really harmful in larger scale uses of Cohttp. Running https://github.com/avsm/opam-mirror is causing undebuggable failures when running a large number of threads. This must be fixed before a 1.0 despite the interface breaking change I think /cc @rgrinberg

[SGrondin]

Agreed. I'm running a busy cohttp server and managing its crashes and errors is a pain.

[rgrinberg]

@avsm I'm +1231241 on this change but why do you consider this a breaking change? I don't consider S.IO part of our stable API.

@SGrondin Can you please describe in a bit more detail. If it's graphic enough, it would encourage me to crack my knuckles and finally tackle this.

[yomimono]

src/s.ml now has a module type Http_io with

val read : IO.ic -> [ `Eof | `Invalid of string | `Ok of t ] IO.t

, which seems to fulfill this request, so I'll close this issue. If this is incomplete or incorrect, please feel free to reopen.