Le 08/09/11 04:30, Gordon Woodhull a écrit :

On Sep 7, 2011, at 5:35 PM, "Vicente J. Botet Escriba"<vicente.botet@wanadoo.fr> wrote:

...

Do you think that it is worth proposing an addition to Boost.Chrono that provides only stopwatches (without reporting)? It would be really nice to have a replacement for Boost.Timer ASAP. Perhaps it would make sense to introduce something simple as a Chrono example, and then decide how to expand it into a library later?

I don't know what do you consider a simple example. Do this simple example report on standard output the time slapsed? Could you give one, please? This will help me a lot.

Not sure if this answers your question. I haven't looked closely.

I will be interested in your opinion. Let me know what do you think is you have time to look closely. Best, Vicente

Le 10/09/11 10:02, Gordon Woodhull a écrit :

On Sep 9, 2011, at 3:10 PM, "Vicente J. Botet Escriba"<vicente.botet@wanadoo.fr> wrote:

...

Le 08/09/11 04:30, Gordon Woodhull a écrit :

...

On Sep 7, 2011, at 5:35 PM, "Vicente J. Botet Escriba"<vicente.botet@wanadoo.fr> wrote:

...

Do you think that it is worth proposing an addition to Boost.Chrono that provides only stopwatches (without reporting)? It would be really nice to have a replacement for Boost.Timer ASAP. Perhaps it would make sense to introduce something simple as a Chrono example, and then decide how to expand it into a library later? I don't know what do you consider a simple example. Oops, I meant: yes please make stopwatches a sublibrary of chrono and then see if it needs to set off on its own.

Yes, this is a good idea.

All I want is a replacement for boost::timer, for writing benchmarks. I never used the progress_timer so I don't know about the reporting side of Boost.Timer, never wanted it. I want a simple class that keeps track of elapsed time so I don't have to even do one subtraction.

In one quick pass of the Stopwatches documentation, I can't find anything that doesn't do automatic reporting. I figure there must be a class in there that looks a lot like boost::timer or the one in Chrono's "key is struck" example, which is what your original question must be proposing. I guess this is due to the fact I used everytime the reporting facilities. The equivalent of the timer class is stopwatch. Here follows un example

using namespace boost::chrono; using namespace boost::stopwatches; int f1(long j) { stopwatch|| <mailbox:///Users/viboes/Library/Thunderbird/Profiles/7k7xdbkm.default/Mail/pop.orange.fr/Sent?number=319244&part=1.2.2&filename=stopwatches.html><system_clock> sw; for ( long i = 0; i < j; ++i ) std::sqrt( 123.456L ); // burn some time return milliseconds(sw.elapsed()).count(); // retun the number of milliseconds } elapsed() returns the elapsed time since the stopwatch was created. Does this respond to your expectations?

I find Boost.Timer kind of buggy. It was reporting CPU time on Linux, so I couldn't benchmark a parallelized program. I feel it has other problems. I thought Chrono was supposed to replace it, but it's still a little verbose for this very simple task.

My first proposal of Boost.Chrono included Stopwatches, but some on this list sugested that i would be better to split the library :( If no body is agains I will move Boost.Stopwatches to the namespace boost::chrono, remove the reporting facilities, and find a date for a review. Best, Vicente

...

My first proposal of Boost.Chrono included Stopwatches, but some on this list sugested that i would be better to split the library :(

That design still has merit, but -

...

If no body is agains I will move Boost.Stopwatches to the namespace boost::chrono, remove the reporting facilities, and find a date for a review.

My point was that since you own Chrono you can add whatever you want to it. So you could get something distributed now and see about a review later.

But perhaps I'm the only one thinking the need for a Boost.Timer replacement is urgent.

Nod. Sounds like an important addition to me. IMO a small addition of a stopwatch could be done without a formal review - maybe just post the design and get feedback? But something like: template <class Clock> struct stopwatch { void reset(); double elapsed(); }; would seem hard to go wrong with? OK arguably the result of elapsed() should be a duration, but that makes it harder to use.... BTW I spotted a typo in your docs: "The standard defines tree system-wide clocks" ^^ John.