Update to dune 2

[shonfeder]

Closes #1034

Everything here should be a purely mechanical change required to use dune 2.0 syntax.

[shonfeder]

It looks like the ocaml-ci failures here are the same 5 that are failing on master https://github.com/ocsigen/lwt/runs/31355838204 (FreeBSD and MacOs).

[raphael-proust]

We might be preventing some projects to update to more recent Lwt with this PR. I don't think it's a problem because it's project that have a dune<2.0 constraint. They already can't update much of their deps, probably innactive projects.

Still, for good measure I've grepped through the opam-repository and I found multiple packages which have dependency to dune with a version constraint < "2.0".

• 0install.2.14 and 0install.2.14.1: it's fine because they have a constraints for lwt to be < "5.0.0"
• base_quickcheck.v0.12.0: maybe fine because it depends on base < "V0.13" which is somewhat dated?
• blake2.0.1 and blake2.0.2: probably fine because it's not widely used (it's a tezos thing) and is somewhat dated (version 0.3 is out)
• bun.0.3.3: probably fine because there is a more recent releases, and because it's a binary intended to be used in the CI, so it doesn't mess-up users' dependency cones
• bytearray.1.0,0: probably fine because it has ocaml<4.10, several newer point releases, no "used by" in the opam-repo
• core_kernel.v12.*: same as base_quickcheck
• csv.2.1: that one is used a lot and has a csv-lwt companion package, although there are newer releases and this version is from 2018
• dokeysto.3.0.0 and dokeysto_*.3.0.0: probably fine because there are point-releases with looser constraints as well as newer releases
• dolmen.0.4: probably ok because there are point releases following
• encore.0.2: probably breaks users of angstrom-lwt-unix.0.14.0+dune<2.0 but maybe ok because there are newer releases for all of those.
• ezsqlite.0.41 and ezsqlite.0.4.1but notezsqlite.0.4.2` so probably ok, especially because this doesn't seem to be an Lwt-aware library.
• fluent_logger probably ok because is not Lwt-aware
• gemini.0.1 and gemini.0.2.0: ok because uses async
• get_line<=6.0 probably ok because it's a binary
• hacl<=0.2 probably ok because not widely used (tezos only?)
• herdtools7.7.54 probably ok because there are newer releases and it seems to be for the purpose of running some simulations
• mccs.1.1+[5-9] probably ok because newer "plus" releases
• minicli.5.0.0 and minicli.5.0.1: probably ok because there is minicli.5.0.2
• minisat.0.2 probably ok because there are newer point releases
• npy.0.0.8 maybe ok because there is a 0.0.9
• ocaml-migrate-parsetree.1.{1.0,0.11} (use jbuilder) but not the other versions so most likely ok
• odoc.1.3.0 probably fine because other versions
• open maybe ok because it's a binary rather than a library?
• orsvm_e1071.3.0.2, potentially breaking because the next version is 4.0.0 so a major release, on the other hand the deps seem to suggest use of parmap so unlikely to be used in lwt?
• some parany versions but I don't think it's recommended to mix with lwt
• patdiff.v0.12.0 most likely ok because newer versions and binary
• ppx_deriving_cmdliner.0.4.1 probs ok because newer releases
• ppxlib.0.2.1 and ppxlib.0.3.0 most likely ok because there are so many newer releases
• qtest.2.10 probs ok because there's a point release following and some more releases after that
• re2.v0.12.0 point release following plus more releases after
• some reason version but nothing too recent
• relit-reason all versions
• rosetta.0.1.0 maybe ok because further versions
• some rtop versions but nothing too recent
• setcore.1.0.1 maybe ok because there's one point release afterwards
• tensorflow.0.0.11 possibly breaking because it's the only version that even supports dune,. but quite dated
• uecc.0.1 but has further point release and seem to be tezos only
• uri.2.0.0 and uri.2.1.0 but there are much newer versions

At this point I timed out, but I think it's fine. There are a couple of things that could potentially be maybe interesting to look into… but nothing screams for attention.

[raphael-proust]

Maybe there's a better way to handle the conditional preprocessing than the current

let preprocess =
   match Sys.getenv "BISECT_ENABLE" with
   | "yes" -> "(preprocess (pps bisect_ppx))"
   | _ -> ""
   | exception _ -> ""
let () = Jbuild_plugin.V1.send @@

dune envs maybe?

Anyway, that's orthogonal, I'll look into it later.

[raphael-proust]

I'm trying to fix issues with the CI

[raphael-proust]

I removed some tests that were not used

• packaging tests: not ran from anywhere unless implicitely by dune @runtest and then kind of incorrectly anyway, uneeded because opam-repository CI runs revdeps checks which ensures we don't break packaging for users
• ppx_expect tests: not ran from anywhere, had some type errors when I tried to activate them

The CI is much greener now. I think we can merge this. @smorimoto let me know if you disagree.

[MisterDA]

Why restrict Windows support? Does this mean Lwt on Windows is only supported on 5.2?