Hi Jasen,

JB>GW> Bit odd really, as the two compilers I've looked at so far say clock()
JB>GW> measures time from program start.

JB>It doesn't even do that. help-pc says it measures elapsed processor-time
JB>for the process, so under an multitasking O/S it could well return a lower
JB>value... (my turbo-c manual agrees) but clock under turbo-c 2.0 isn't
JB>multitasker aware.

I said as much in other posts in the thread...

JB>After some experimentation under linux, it appears that a clock counts only
JB>the time that the process is actually executing, in a program that does a
JB>sleep(10) (10 seconds) clock returns 0 (no time elapsed)

Yippee, a system that has it implemented correctly...

JB>Two processes executing concurrently would share the
"clocks" and would each
JB>only see half the elapsed time.

JB>This suggests that the (portable) sleep in snippets is broken in that it
JB>doesn't stop clock(). And that uclock is broken in that it doesn't measure
JB>cpu-time.

I can't find a sleep() in SNIPPETS :-(. Which file?
The routines in uclock.c don't claim to give processor run time, but
absolute elapsed time to high resolution. They rely on direct access to
the hardware tick generator, and so are only valid for PCs and only
accurate under DOS. Under OS/2 the timer is virtualised and so not
really accurate (it does say so in the comments at the top of the file).

JB>I have no idea of what would happen under windows or OS/2 or on an Amiga or
JB>Mac.

It _should_ measure processor time. But what it actually does will
depend almost as much on the compiler library routines as the OS...
Watcom and IBM say it returns an approximation to processor time.
Borland C/C++ 3.1 expects to be the only process running, and so returns
system ticks since program start.
Borland C/C++ for OS/2 1.0 says it returns processor time, but
contradicts itself by saying "clock can be used to determine the time
interval between two events.". I suspect they've got it wrong (what a
suprise :-( ).

[later]

They have got it wrong :-(. No suprise there...
I checked clock() under BCOS/2 V1.0, and it returns elapsed time, not
processor time :-(. If you go to sleep() for 5 seconds, clock() shows 5
seconds time, rather than the 0 it should show if it measured processor
time...

George

 * SLMR 2.1a * Computers eliminate spare time.

--- Maximus/2 3.01