Mac: compute non-BOINC CPU usage in a more accurate way

[CharlieFenton]

Fixes #4327, #4328

Description of the Change
Implement #4859 for Macintosh computers

[AenBleidd]

There is no default value for scale.
In case processorCount is not equal zero, this value will be set to 0 by default, so later this function will always return 0 because the last statement is multiplication with scale.
If this is intended - then fine, but I believe this is a mistake

[AenBleidd]

P.S. If you want just to initialize this static variable once, then it's better to have either some flag or just verify if this value is zero (and then initialize it with a proper value) otherwise just continue execution

[CharlieFenton]

@AenBleidd I intended this. I have no better solution in case of such errors because the code by @davidpanderson ( line 495 in client/app.cpp) which calls this function has no ability to handle errors returned from total_cpu_time(). The UNIX / Linux version of total_cpu_time() created for #4859 in lib/procinfo_unix.cpp also returns 0 in case of any error opening or parsing /proc/stat. The MSW version of total_cpu_time() in lib/procinfo_win.cpp from #4859 does no error checking at all.

I agree that the code which calls this should take some sort of appropriate action if total_cpu_time() indicates an error by returning 0.

[AenBleidd]

Ok, code looks OK to me. Please let me know when it's tested and ready to merge

[CharlieFenton]

One possibility for handling errors from total_cpu_time() might be to revert to the old method of calculating non_boinc CPU usage (lines 502 - 523 in client/app.cpp).

[CharlieFenton]

P.S. If you want just to initialize this static variable once, then it's better to have either some flag or just verify if this value is zero (and then initialize it with a proper value) otherwise just continue execution

I'm not sure I understand what you are suggesting. processorCount is always updated by the host_processor_info() call; there is no way to avoid that. But I am having it do double duty, both as a flag allowing me to call sysconf(_SC_CLK_TCK) only once, and also to indicate an error by returning 0 if for some reason processorCount is 0 (as you pointed out in the first comment above.)

[AenBleidd]

This double duty is exactly that I don't like: it's an additional complication and fragile point, and in case you need to change the code, you still should remember that this variable serves as a flag too.
I believe in this case it's acceptable but in general case I'd recommend to use a separate variable as a flag.

[davidpanderson]

I suggest using
static bool first = true
if (first) {
first = false
... get CPU count etc.

Also: total_cpu_time() should return 0 if any error occurred.

[CharlieFenton]

This code is now ready to be merged.

I tested this by setting local preferences to suspend if non-BOINC CPU usage > 50%, setting the mem_usage_debug flag and letting BOINC run while telling Xcode to do a full build of BOINC. This computer has 4 CPUs, so the Xcode build runs hundreds of instances of the compiler (one per source file) 4 at a time. Each compiler instance runs for a fraction of a second.

With BOINC 7.20.2 (using the old code), BOINC failed to detect the high non-BOINC CPU usage. It reported mostly between 4% and 15%, with 25% a few times. The old code never suspended BOINC.

With the new code, it reported mostly 80% to 90% non-BOINC CPU usage and correctly suspended until the Xcode build finished. So the new code is accomplishing what was intended: better recognizing the CPU load when there are multiple short duration processes.