One might wonder why their CI builds takes ages

yes that's a bit annoying. One of the great thing about the current behavior is that the failure mode is pretty clear. I am wondering if we should try to compare version in bash and do the switch only if that's below 4.03 (I am not sure I really like that this solution either).

I'm not super fond of making those scripts more complex but as you pointed out the current error is better than silently falling back to switch creation unexpectedly!

If you want to add a check in the script I think it would need to be more accurate than < 4.03 since I guess some people could be interested in having variants (as in non base-compilers) in their CI.

Maybe adding a case on the actual pre-installed compiler versions to convert them to the 4.0x form would do the trick. That still means the script has to be updated when those change but I guess that's ok.

Your call!

Same here, I'm not super fond of making the script more complex.

Maybe the right solution is to control that behavior via an OPAM_SWITCH_CREATE env variable. It's a bit ugly, but at least it's explicit.

One advantage of your patch is that we can now specify a particular patch release (e.g. OCAML_VERSION=4.07.0).

How about just printing something to stdout to indicate that a custom switch is getting compiled with a note that things might be faster if the final .0 is dropped? That would be a simple change and not add yet another environment dial into the mix.

Ok let's merge this, we can see if we can make things better later.