Warn if num installed to a broken system compiler

[dra27]

This follows on from #11207.

If the num library was installed in a system switch where the user had write permissions to the system compiler's lib directory before #11207 was merged, then fresh installations of the ocamlfind package will detect this "system" installation of num and create a META file for it.

This in turn will cause the num package installation to fail. Annoyingly, this failure only appears once because opam then executes ocamlfind remove num so a subsequent attempt to install the num package will appear to have worked (however, in reality the system-installed files will conflict with the ones in the opam directory).

This patch displays a comprehensive error message the first time, strongly suggesting that the user manually delete the files which were incorrectly installed previously.

If you have a system compiler where num has been manually installed to ocamlc -where (e.g. by running sudo make install from the GitHub sources) and a fresh installation of the ocamlfind package, then you will now see the following if you attempt to install the num package:

dra@zesty64:~/num/src$ opam install num
The following actions will be performed:
  ∗  install num 1.1 

=-=- Gathering sources =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
[num.1.1] found in cache

=-=- Processing actions -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
[ERROR] The installation of num failed at "make findlib-install".
Processing  2/2: [num: make findlib-uninstall]
#=== ERROR while installing num.1.1 ===========================================#
# context      2.0.0~beta6 | linux/x86_64 | ocaml-system.4.06.0 | file:///home/dra/opam-repository
# path         ~/.opam/temp/.opam-switch/build/num.1.1
# command      /usr/bin/make findlib-install
# exit-code    2
# env-file     ~/.opam/log/num-21717-a00279.env
# output-file  ~/.opam/log/num-21717-a00279.out
### output ###
# [ERROR] It appears that the num library was previously installed to your system
#         compiler's lib directory, probably by a faulty opam package.
#         You will need to remove arith_flags.*, arith_status.*, big_int.*,
#         int_misc.*, nat.*, num.*, ratio.*, nums.*, libnums.* and
#         stublibs/dllnums.* from /usr/local/lib/ocaml.
# Makefile:46: recipe for target 'findlib-install' failed
# make: *** [findlib-install] Error 1



=-=- Error report -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
┌─ The following actions failed
│ ∗  install num 1.1
└─ 
╶─ No changes have been performed

cc @avsm

[samoht]

I haven't followed the discussion on this issue so I'll let an other maintainer merge this PR.

[dra27]

@samoht - are you able to kick the DataKit CI and Camelus?

[samoht]

I've restarted the Datakit CI runs, but I don't control Camelus. /cc @AltGr to see if he can do something about it :-)

[AltGr]

Kicked it, but a "Could not complete" error is probably reproducible...
(NB: you need to go to config→webhooks→opam-ci, find the corresponding request and replay it)

[camelus]

✅ All lint checks passed eba59ad

• These packages passed lint tests: num.1.0, num.1.1

---

✅ Installability check (8230 → 8230)

[kit-ty-kate]

Sorry for the long delay. Thanks !

[Calvin-L]

The guard implemented in this PR is causing me a lot of headaches. (Although, maybe fewer headaches than I would be experiencing without it? Unclear.)

After running opam install num, I have arrived in a totally stuck state (the advice in the message is irrelevant). Here's what I'm seeing:

$ ocamlc -where
/usr/local/lib/ocaml

$ ls -l /usr/local/lib/ocaml/{arith_flags.,arith_status.,big_int.,int_misc.,nat.,num.,ratio.,nums.,libnums.,stublibs/dllnums.}*
ls: /usr/local/lib/ocaml/arith_flags.*: No such file or directory
ls: /usr/local/lib/ocaml/arith_status.*: No such file or directory
ls: /usr/local/lib/ocaml/big_int.*: No such file or directory
ls: /usr/local/lib/ocaml/int_misc.*: No such file or directory
ls: /usr/local/lib/ocaml/libnums.*: No such file or directory
ls: /usr/local/lib/ocaml/nat.*: No such file or directory
ls: /usr/local/lib/ocaml/num.*: No such file or directory
ls: /usr/local/lib/ocaml/nums.*: No such file or directory
ls: /usr/local/lib/ocaml/ratio.*: No such file or directory
ls: /usr/local/lib/ocaml/stublibs/dllnums.*: No such file or directory

$ opam install num
The following actions will be performed:
  ∗ install num 1.2

<><> Gathering sources ><><><><><><><><><><><><><><><><><><><><><><><><><><>  🐫 
[num.1.2] found in cache

<><> Processing actions <><><><><><><><><><><><><><><><><><><><><><><><><><>  🐫 
[ERROR] The installation of num failed at "make findlib-install".

#=== ERROR while installing num.1.2 ===========================================#
# context     2.0.5 | macos/x86_64 | ocaml-system.4.07.1 | https://opam.ocaml.org#506eb174
# path        ~/.opam/default/.opam-switch/build/num.1.2
# command     ~/.opam/opam-init/hooks/sandbox.sh install make findlib-install
# exit-code   2
# env-file    ~/.opam/log/num-98674-ad8886.env
# output-file ~/.opam/log/num-98674-ad8886.out
### output ###
# [ERROR] It appears that the num library was previously installed to your system
#         compiler's lib directory, probably by a faulty opam package.
#         You will need to remove arith_flags.*, arith_status.*, big_int.*,
#         int_misc.*, nat.*, num.*, ratio.*, nums.*, libnums.* and
#         stublibs/dllnums.* from /usr/local/lib/ocaml.
# make: *** [findlib-install] Error 1



<><> Error report <><><><><><><><><><><><><><><><><><><><><><><><><><><><><>  🐫 
┌─ The following actions failed
│ ∗ install num 1.2
└─ 
╶─ No changes have been performed

$ ocamlfind query num
/Users/loncaric/.opam/default/lib/num

A few more details about how we got here:

• My package manager (Homebrew) installs a system-wide version of num for a program that I need.
• The suggestion here (manually remove various files) is a bad one, since those files are managed by Homebrew.
• What you are seeing above happens after a previous failed install for num. and after telling Homebrew to uninstall its version of num.
• The suggestion here (manually remove various files) is totally irrelevant in my current state.
• The only way I have found to successfully install this package is to uninstall Homebrew's version (undesirable since I lose the downstream program I need), nuke ~/.opam, opam init, and then reinstall num.

I desperately wish that Homebrew's num and opam's num could coexist. Failing that, I desperately wish that there was a clear way out of the "stuck" state I arrived in after running opam install num.

[dra27]

@Calvin-L - opam's support for packages installed by a system compiler is very limited at the moment, so I'd expect the solution to be not installing the homebrew ocaml-num package, at least for the foreseeable future.

I'm possibly puzzled about the ocamlfind query num result - what's in /Users/loncaric/.opam/default/lib/num and, if there's a META there, what's its content?

[Calvin-L]

Thanks for the quick follow-up! I did get everything working by removing Homebrew's system-wide installation of num (although I had to nuke ~/.opam as well).

Here are the contents of the META file from my earlier corrupted state:

$ cat /Users/loncaric/.opam/default/lib/num/META 
# Specification for the "num" library:
requires = "num.core"
requires(toploop) = "num.core,num-top"
version = "[distributed with Ocaml]"
description = "Arbitrary-precision rational arithmetic"
package "core" (
  directory = "^"
  version = "[internal]"
  browse_interfaces = " Unit name: Arith_flags Unit name: Arith_status Unit name: Big_int Unit name: Int_misc Unit name: Nat Unit name: Num Unit name: Ratio "
  archive(byte) = "nums.cma"
  archive(native) = "nums.cmxa"
  plugin(byte) = "nums.cma"
  plugin(native) = "nums.cmxs"
)

[dra27]

@Calvin-L - that META looks correct for an opam-installed ocamlfind which has detected the homebrew-installed num package. I'm not clear what you mean by corrupt - was it that you couldn't install anything else which depends on the opam num package?

[Calvin-L]

That's correct. I was using opam to install sexplib which depends on num. However, opam insisted on building num, which would then fail. (Sorry for using "corrupt" incorrectly; I have zero familiarity with opam's directory structure and I assumed that something was broken in ~/.opam.)

[dra27]

@Calvin-L - no problem with terminology, I just wanted to be sure I'd correctly understood what had gone wrong from your perspective 🙂