Consider following POSIX standards when installing .sh scripts

[jonahbeckford]

I'd like to get your thoughts on doing the following for all .sh scripts installed into the build/ tree and into the OPAM_SWITCH_PREFIX tree:

• convert them to LF endings if CRLF. CR is not part of the POSIX standard, and some Windows shells (ie. MSYS2 dash.exe) are purist in adhering to POSIX standard. This avoids the need to have a .gitattributes file (ex. https://github.com/diskuv/dkml-installer-ocaml/blob/main/.gitattributes#L22) when checking out a project using git on Windows.
• If and only if the script has a shebang from the following list:
  • #!/bin/bash
  • #!/bin/sh
  • #!/usr/bin/env sh
  • #!/usr/bin/env bash
    replace the shebang with the output of command -v sh or command -v bash or follow the POSIX search algorithm on getconf PATH

The POSIX standard for installing scripts is at https://pubs.opengroup.org/onlinepubs/9699919799/utilities/sh.html#tag_20_117_16. The detection of the correct shell can be done at compile time (ex. opam$ ./configure --bash-shell /bin/bash --posix-shell /bin/sh) or be an opam global variable ... not sure. But the default of /bin/bash and /bin/sh will almost always work.

This:

• is related to the NixOS discussion at https://discuss.ocaml.org/t/installing-camlp5-on-ocaml-5-0-0-nixos/11482
• would help resolve problems for Windows users of sqoop and Chocolatey (confer: Scoop bash.exe (etc.) conflicts with v1.2.0-prerel7 Installer diskuv/dkml-installer-ocaml#34; that was an ocaml installation problem, but eventually the same shebang issue is going to happen with these same users with opam installed scripts)
• lack of shebang correctness pops up in weird places that I'm guessing is only sometimes visible to the opam team (see Enforce the macOS sandbox script to use /bin/bash instead of /usr/bin/env bash for a more consistent experience #5451)

I realize this may not be backwards compatible as proposed; perhaps there is a way to make it backwards compatible. For example, instead of "all .sh scripts" it could be an explicit .opam field like build_scripts: [ "configure" ].

Personally, if there was an opam global variable I'd set it to the full path to the MSYS2 dash.exe (that is the portable shell for MSYS2) and the full path to MSYS bash.exe so whatever weirdness a Diskuv OCaml Windows user has in their PATH (ex. Cygwin, Git Bash, Chocolatey, etc), the opam installed scripts just work.