Meaning of time/1, extra choice point

[UWN]

?- 9=I,N is 2^I, time((true;true)),length(L,N).
   % CPU time: 0.000 seconds
   I = 9, L = [_A,_B,_C,_D,_E,_F,_G,_H,_I,_J|...], N = 512
;  % CPU time: 4.199 seconds
   I = 9, L = [_A,_B,_C,_D,_E,_F,_G,_H,_I,_J|...], N = 512
;  % CPU time: 8.396 seconds
   false.
?- time((true;true)).
   % CPU time: 0.000 seconds
   true
;  % CPU time: 0.019 seconds
   true
;  % CPU time: 0.040 seconds
   false.

There should not be a big difference between these two queries, as time/1 only measures (true;true). However, it seems that time/1 also includes the time spent in predicates thereafter and the top level loop which takes a lot with longer lists.

Further, time/1 adds an extra CP which changes performance substantially by retaining data that would otherwise be collected (in the long run, indeed)

[triska]

The CP is due to format/2 using maplist/N, please see #192 for the core reason that should be addressed to resolve this once and for all, without ad hoc workarounds.

Resolved via #732!

[ghost]

On Linux, the following happens:

$ scryer-prolog
?- use_module(library(time)).
   true.
?- time(sleep(2)).
   % CPU time: 0.000 seconds
   % CPU time: 0.001 seconds
   true.
?- halt
$ time -p scryer-prolog -g 'use_module(library(time))' -g 'sleep(2)' -g halt      
real 2.56
user 0.47
sys 0.08
$

Sleeping time isn't measured. What is time/1 measuring?

[triska]

time/1 reports the CPU time, this is how long the CPU was active. sleep/1 suspends the process, so the CPU does nothing in that time (for that process).

The elapsed time is referred to as the "wall time".

[UWN]

?- time(true).
   % CPU time: 0.010 seconds
   % CPU time: 0.044 seconds
   true.

And what does the second time mean?

[triska]

After ten thousand attempts, I think I resolved both issues!

Nested invocations of time/1 require correction of #1123. We currently have:

?- time(time(true)).
   % CPU time: 0.000 seconds
   % CPU time: 0.000 seconds
   true.
?- time(time(true)).
unexpected loop

Once #1123 is solved, this should also work with the present implementation of time/1.