http-io-streams test failure

[mihaimaruseac]

           Opening a connection                                                                                                                                                                                     
             properly caches hostname and port [✘]                                                                                                                                                                  
           Request, when serialized                                                                                                                                                                                 
             has a properly formatted request line [✔]                                                                                                                                                              
             handles empty request path [✔]                                                                                                                                                                         
             terminates with a blank line [✘]
             has a properly formatted Host header [✔]
             properly formats Accept header [✔]
             properly formats Authorization header [✔]
           Parsing responses
             parses a simple 200 response [✔]
             parses response when HTTP version doesn't match [✔]
             handles padded Content-Length [✔]
             recognizes chunked transfer encoding and decodes [✘]
             recognzies fixed length message [✘]
             doesn't choke if server neglects Content-Length [✔]
             reads body without Content-Length or Transfer-Encoding [✔]
             handles 204 No Content response without Content-Length [✔]
             recognizes gzip content encoding and decompresses [✘]
             repeated response headers are properly concatonated [✘]
           Expectation handling
             sends expectation and handles 100 response [✘]
           Convenience API
             PUT correctly chunks known size entity body [✘]
             POST correctly chunks a fileBody [✘]
             POST with form data correctly encodes parameters [✘]
             GET internal handler follows redirect on 307 [✘]
             check splitURI for local redirects [✔]
             GET internal handler follows local redirect on 307 [✘]
             GET includes a properly formatted request path [✘]
             too many redirects result in an exception [✘]
             GET with general purpose handler throws exception on 404 [✘]
             public establish function behaves correctly [✘]
           Corner cases in protocol compliance
             Sends a request body for PUT [✘]
             Sends a request body for DELETE [✘]
             Sends a request body for PATCH [✘]

Unsure what changed but with latest nightly we're seeing these failures. Will mark tests as expected to fail.

CC @andreasabel

[andreasabel]

I am also seeing these in the CI: E.g. https://github.com/haskell-hvr/http-io-streams/actions/runs/12582552298/job/35068402844?pr=20

Opening a connection properly caches hostname and port
uncaught exception: IOException of type NoSuchThing
Network.Socket.connect: <socket: 7>: does not exist (Connection refused)

[andreasabel]

Locally on my machine the tests run fine, so maybe it is a case of tightened permissions in the github action runners.
@mihaimaruseac Where (on what machines) does stackage run the tests?

[mihaimaruseac]

It's running on a docker image, built from https://github.com/commercialhaskell/stackage/blob/master/Dockerfile.

See https://github.com/commercialhaskell/stackage/actions/runs/12231107228 for the last version of the image

[andreasabel]

On a different CI, the tests succeed: https://github.com/haskell-hvr/http-io-streams/actions/runs/12655700933/job/35266609463?pr=20#step:11:11

So it is certainly not http-io-streams itself to be blamed, must be in the test environment(s).

[mihaimaruseac]

I was seeing the failure on 3 consecutive runs (I always rerun on test breakages).

If it is stable now, we can probably re-enable. I'm outside my duty week, but could still approve a PR (though will leave the current curator to merge)

[andreasabel]

If it is stable now, we can probably re-enable. I'm outside my duty week, but could still approve a PR (though will leave the current curator to merge)

Probably not, since we have not found the cause of failure.
I haven't changed any byte of the http-io-streams library.

Upstream issue:

• Tests fail in Haskell CI and stackage haskell-hvr/http-io-streams#22