[new release] tcpip (6.1.0)

[dinosaure]

OCaml TCP/IP networking stack, used in MirageOS

• Project page: https://github.com/mirage/mirage-tcpip
• Documentation: https://mirage.github.io/mirage-tcpip/

CHANGES:

• Ensure that listen really binds the given socket before
  creating a task on tcpip.stack-socket (@dinosaure, @hannesm, Ensure that a call of listen really binds the given socket to the INET_ADDR mirage/mirage-tcpip#439)
• Add ppx_cstruct as a dependency (@hannesm, @dinosaure, Ensure that a call of listen really binds the given socket to the INET_ADDR mirage/mirage-tcpip#439)
• Upgrade to ocamlformat.0.17.0 (@dinosaure, Upgrade travis ci and ocamlformat mirage/mirage-tcpip#442)
• Drop the support of OCaml 4.08.0 (@dinosaure, Upgrade travis ci and ocamlformat mirage/mirage-tcpip#442)
• Use the usual layout to compile freestanding C stubs and link them to
  a Solo5 unikernel (@dinosaure, @hannesm, Use the usual layout to compile freestanding C stubs mirage/mirage-tcpip#441)
  breaking changes
  C stubs are prepended by mirage_. Symbols such as checksum's
  symbols are caml_mirage_tcpip_* instead of caml_tcpip_*
tcpip.unix is a fake sub-package and user does not it anymore, he can
  safely remove it from its project.
• Conflict with < ocaml-freestanding.0.4.1 (@hannesm, Use the usual layout to compile freestanding C stubs mirage/mirage-tcpip#441)

[camelus]

Commit: dbfddf8

A pull request by opam-seasoned @dinosaure.

☀️ All lint checks passed dbfddf8

• These packages passed lint tests: tcpip.6.1.0

---

☀️ Installability check (+1)

• new installable packages (1): tcpip.6.1.0

[mseri]

A test failed on the latest Ubuntu with

# ┌──────────────────────────────────────────────────────────────────────────────┐
# │ [FAIL]        socket                0   two sockets connect via TCP.         │
# └──────────────────────────────────────────────────────────────────────────────┘
# test.exe: [INFO] IPv4 socket stack: connect
# test.exe: [INFO] IPv4 socket stack: connect
# [exception] Unix.Unix_error(Unix.EADDRINUSE, "bind", "")
#             Raised by primitive operation at Tcpip_stack_socket.V4.listen_tcpv4 in file "src/stack-unix/tcpip_stack_socket.ml", line 70, characters 6-88
#             Called from Dune__exe__Test_socket.two_connect_tcp.(fun) in file "test/test_socket.ml", line 40, characters 2-60
#             Called from Dune__exe__Test.run in file "test/test.ml" (inlined), line 37, characters 15-24
#             Called from Dune__exe__Test.(fun) in file "test/test.ml", line 51, characters 42-47
#             Called from Alcotest_engine__Core.Make.protect_test.(fun) in file "src/alcotest-engine/core.ml", line 277, characters 17-23
#             Called from Alcotest_engine__Monad.Identity.catch in file "src/alcotest-engine/monad.ml", line 34, characters 31-35
#             

Is this a known Heisenbug of the tests or needs to be investigated?

[dinosaure]

I think it's related to mirage/mirage-tcpip#437, let me look into it and try to find a solution.

[dinosaure]

I re-released mirage-tcpip with a fix about cancelation (thx @hannesm & @talex5), this release should be fine now.

[hannesm]

please hold on, on Wednesday we discussed to avoid the caml_ prefix in the C stub names -- would you mind to take a look at mirage/mirage-tcpip#445 and include this in the release?

[dinosaure]

Release was redone including the patch of @hannesm 👍.

[mseri]

Should we worry about those three EADDRINUSE failures? (Seems they have not yet gone)

[dinosaure]

Should we worry about those three EADDRINUSE failures? (Seems they have not yet gone)

Hmmmhmm, it's really strange that in our CI, this bug does not appear and it appears on the opam-ci. Then, we delayed the release in order to fix this bug... I don't really know what happens and don't have enough time to continue to deeply look on it 😕 (/cc @mirage/core)

[sternenseemann]

On NixOS never hit these failures (to the point where I accidentally updated tcpip to 6.1.0 at the first github release because I forgot to check if the opam PR was merged already 😬).

[mseri]

I don’t mind merging, you just have to give me the thumbs up for doing it

[dinosaure]

For my point of view, it's not acceptable to release mirage-tcpip as is 😕 . I hope that someone more confident with the code understand such error and I believe that it will take a long time. So, it's better to close I think.

[mseri]

Do you know if it has ever happened before? (The error I mean)

[dinosaure]

As far as I can tell, no but it seems serious when opam-ci reproduced it several times.

[kit-ty-kate]

Do you know if it has ever happened before? (The error I mean)

This sort of failure is rather common with mirage software (I'm looking into the past logs to get a list of the ones i remember about). Last version used -j1 for the tests, testing with that for now.

[kit-ty-kate]

ok nevermind, it still fails with the same failure. Closing then.
I might have remembered old similar failures (e.g. #18002 (comment) or mirage/mirage-tcpip#435), I can't seem to find similar failures in my logs (though similar non-deterministic failures have been seen in the past)

[samoht]

It's not super surprising that this fails If concurrent tests try to open the same port. /cc @MagnusS

[MagnusS]

@samoht looks like it may be an old bug that didn't fail the tests before this release, I followed up in mirage/mirage-tcpip#446