Upgrade the "lockfile" with all the most up-to-date dependencies available

[kit-ty-kate]

Changes:

• dune ¯\_(ツ)_/¯ however it cannot be updated to its current latest version due to dune 3.23.1 fails to build on the cygwin backend dune#15340 EDIT: not updated. There are complications with OCaml 5.6
• base64 removed support for OCaml < 4.07
• spdx_licenses upgraded to Version 3.28.0 of the SPDX License List
• menhir removed support for OCaml < 4.08 and fixed a couple of bugs that do not impact us
• patch fixed an issue arising with handwritten patches fix: ---/+++ should take precedence over spurious git metadata hannesm/patch#42
• checkseum fixed some C warnings on recent clang versions
• decompress sped up compression EDIT: not updated. 32bit support is missing

I noticed the C warnings fixed in the checkseum release while building the macOS binaries for 2.6.0~beta2

[hannesm]

excellent work!

[kit-ty-kate]

@hannesm thanks for commenting, i had almost forgotten to update the OCaml compiler versions to 4.14.4 :D

  File "src_ext/decompress/lib/de.ml", line 4055, characters 20-30:
  4055 |   let _hash_magic = 0x9e3779b1 (* xxHash *)
                             ^^^^^^^^^^
  Error: Integer literal exceeds the range of representable integers of type int

@dinosaure does decompress not support 32bit platforms anymore?

[dinosaure]

@dinosaure does decompress not support 32bit platforms anymore?

It's indeed the case due to my last upgrade (decompress.1.6.0), I can figure out (with optint) if it's really required to keep the support on 32-bits (I was probably too optimistic).

[kit-ty-kate]

let _hash_magic = 0x9e3779b1 (* xxHash *)

doing a quick search yesterday on the why of this magic number i found out that it seems to be used exactly because it fits into 32bit ints. However since OCaml has 31bit ints it doesn't fit anymore. Why not use Int32.t or simply reduce the precision to something like 0x4f1bbcdc (aka. int_of_float ((sqrt 5. -. 1.) /. 2.) *. (2. ** 31.)) ?

[dinosaure]

Why not use Int32.t or simply reduce the precision to something like 0x4f1bbcdc (aka. int_of_float ((sqrt 5. -. 1.) /. 2.) *. (2. ** 31.)) ?

This part of decompress is about the deflation, use a boxed int32 value will probably slow down the deflation process (and how we fill/search on our internal hash table with patterns). We can probably give a try to an int32 and see if we can have a regression about the decompress's throughput.

For the global perspective, mirage/decompress#171 is basically the difference between the old zlib (which would like to keep a certain compatibility with really old computers) and zlib-ng. Tell me if you are interested by such experiment (we can also take 0x4f1bbcdc also).

[rjbou]

Thanks!
A tiny comment on first commit message. Proposal

- dune is not updated due to complications with OCaml 5.6 and
- ocaml/dune#15340
+ dune is not updated due to complications with OCaml 5.6 and
+ build on Cygwin issue (ocaml/dune#15340)

[kit-ty-kate]

Ignoring the CI failure on Windows (#6890)