Soft timeout notice

[Drup]

Currently, opam kills the solver when the timeout is exceeded.

As someone who really likes to do opam install ocaml-base-compiler <pkg> when installing a new switch for a package I want (which tends to take a long time for the solver), this is a bit annoying: it means that I have to restart from scratch with a bigger timeout value.

What I would prefer is that opam, when reaching the timeout value, would show a dialog option like this:

The solver has been running for X minutes (as set by OPAMSOLVERTIMEOUT). 
You might want to try to specify a simpler request, use a different solver. 
Press a key to interrupt it now. [] 

This way, I can let it run its course if I want to.

[AltGr]

Ok, I'll check ; I am still working on tweaking the solver for better results, and in my experiments, solving would either be less than a minute, or more than I would care to let it run for. This explains that this wouldn't have been an issue... but your report shows differently.

What you request would be hard to implement, though, because the interface with the solver is pretty limited and we run a single process (in the case of the builtin mccs+glpk, anyway) ; so I am hoping that it won't be needed.

How long on average does the solver need in your "difficult" cases ?

[Drup]

In my experience:

• either base is set in the switch, and solving is under 10 secondes
• base is not set (in my case, on purpose) and my request involves finding the "right" ocaml version, and it takes up to 2/3 minutes.

(I was going to say "isn't it what Lwt_preemptive is for ?", but ...)

[Drup]

And just in case you feel bad about having a solver that takes 3 minutes for complex queries, or for other people that read opam feature requests:

This kind of requests I'm talking about wouldn't even run before mccs. old builtin solver would just explode, and aspcud would consume an exponential amount of ram until it reaches heat death of the universe.

So, as far as I'm concerned, 3 minutes is really good. :3