Refine constraints on lwt6

[raphael-proust]

let the CI run, I'll add constraints if needs arises

[raphael-proust]

broad categories of incompatibilities:

• Lwt_unix.notification type mismatch: type has changed to abstract (was int), simple fix (if there is arithmetics done on the notification then we should burn the package down rather than fix), picos is already fixed but needs release, not sure about others
• Lwt_main.enter_iter_hooks and other hooks: those long-deprecated values have been removed and the code should just call Lwt_main.Enter_iter_hooks.* functions to manipulate hooks
• Lwt_engine has abstract method: add constraint to lwt<6.1.0, fix in package by adding an lwt_engine id, see WIP compatibility with lwt.6 ocaml-multicore/lwt_eio#31

[raphael-proust]

• Lwt_unix.run: long-depercated, now-removed, use Lwt_main.run instead

[raphael-proust]

idk what to do about eio:
the tests break, but it's tests about the order of mixed lwt+eio scheduling, and the order is "better" now (for some notion of better anyway)

I think it's ok to use eio_lwt with lwt6, but the tests do fail so i'll think about what's the right thing to do

[jmid]

With lwt.6.1.1 there are 440 occurrences of this kind of error:
https://opam.ci.ocaml.org/github/ocaml/opam-repository/commit/3af7d0a25910b3d20c909e559efb484eb6a78128/variant/compilers,4.14,lambda-term.3.3.0

#=== ERROR while compiling lwt_log.1.1.1 ======================================#
# context              2.5.0 | linux/x86_64 | ocaml-base-compiler.4.14.2 | file:///home/opam/opam-repository
# path                 ~/.opam/4.14/.opam-switch/build/lwt_log.1.1.1
# command              ~/.opam/opam-init/hooks/sandbox.sh build dune build -p lwt_log -j 71
# exit-code            1
# env-file             ~/.opam/log/lwt_log-7-afc58f.env
# output-file          ~/.opam/log/lwt_log-7-afc58f.out
### output ###
# (cd _build/default && /home/opam/.opam/4.14/bin/ocamlc.opt -w -40 -w +A -g -bin-annot -I src/unix/.lwt_log.objs/byte -I /home/opam/.opam/4.14/lib/bytes -I /home/opam/.opam/4.14/lib/lwt -I /home/opam/.opam/4.14/lib/lwt/unix -I /home/opam/.opam/4.14/lib/ocaml/threads -I /home/opam/.opam/4.14/lib/ocplib-endian -I /home/opam/.opam/4.14/lib/ocplib-endian/bigstring -I src/core/.lwt_log_core.objs/byte -intf-suffix .ml -no-alias-deps -o src/unix/.lwt_log.objs/byte/lwt_daemon.cmo -c -impl src/unix/lwt_daemon.ml)
# File "src/unix/lwt_daemon.ml", line 64, characters 49-68:
# 64 |     Lwt_sequence.iter_node_l Lwt_sequence.remove Lwt_main.exit_hooks [@ocaml.warning "-3"];
#                                                       ^^^^^^^^^^^^^^^^^^^
# Error: Unbound value Lwt_main.exit_hooks
# (cd _build/default && /home/opam/.opam/4.14/bin/ocamlopt.opt -w -40 -w +A -g -I src/unix/.lwt_log.objs/byte -I src/unix/.lwt_log.objs/native -I /home/opam/.opam/4.14/lib/bytes -I /home/opam/.opam/4.14/lib/lwt -I /home/opam/.opam/4.14/lib/lwt/unix -I /home/opam/.opam/4.14/lib/ocaml/threads -I /home/opam/.opam/4.14/lib/ocplib-endian -I /home/opam/.opam/4.14/lib/ocplib-endian/bigstring -I src/core/.lwt_log_core.objs/byte -I src/core/.lwt_log_core.objs/native -intf-suffix .ml -no-alias-deps -o src/unix/.lwt_log.objs/native/lwt_daemon.cmx -c -impl src/unix/lwt_daemon.ml)
# File "src/unix/lwt_daemon.ml", line 64, characters 49-68:
# 64 |     Lwt_sequence.iter_node_l Lwt_sequence.remove Lwt_main.exit_hooks [@ocaml.warning "-3"];
#                                                       ^^^^^^^^^^^^^^^^^^^
# Error: Unbound value Lwt_main.exit_hooks

Is it a good idea to lift this upper bound - or are extra conflicts needed? 🤔

[kit-ty-kate]

note that to have a real sense of which packages actually need lwt5 vs not, you would need to temporarily modify the lwt.6.1.1 package to force its revdeps to build. Otherwise some package might still choose lwt5 based on other factors and you'll miss that data.

[kit-ty-kate]

but overall this is probably more of a job for opam-health-check-ng. I can help you set it up locally if you need.

[raphael-proust]

thanks for the help and the suggestions and all

I'll use opam grep to find occurrences of exit_hooks and other such deprecated-now-removed functions, once this is done i'll see about doing a local health-check

[raphael-proust]

in the case of the broken lwt_eio tests: eio tests the scheduling order of mixed eio+lwt programs, i'm still unclear why the order has changed but

• it has changed to a less surprising scheduling
• lwt doesn't make guarantees about scheduling order
• i think it'd be very wrong to rely on the scheduling order of a mixed lwt+eio program

i'll just remove the tests altogether methinks