Re: [racket-users] Racket 6.4 very slow

[Robby Findler]

On Sat, Mar 5, 2016 at 8:09 AM, Matthew Flatt  wrote:
> If you start DrRacket fresh or create a new tab and paste
>
> (define (total n)
>   (for/sum ([x (in-range (+ 1 n))]) x))
> (time (total 10))
>
> into the interactions window, then it runs at non-debugging speed.
> After clicking "Run" (with an empty program), the interactions window
> runs the example with debugging speed.
>
> That difference seems confusing, and it might explain why some had
> trouble seeing the example run slowly.

I think this is a bug in the behavior of DrRacket, separate from any
lack of transparency. I've made an attempt at a fix and pushed it.

On Wed, Mar 2, 2016 at 10:25 PM, Neil Van Dyke  wrote:
> Alex Harsanyi wrote on 03/02/2016 11:19 PM:
>>
>> If it cannot be improved, perhaps a warning message should be printed in
>> the eval window...
>
> I like this idea.  Maybe add to the DrRacket REPL banner, the
> debugging/instrumentation options that are enabled.

I've pushed something that uses these strings, put onto the name of
the language as shown in the REPL. Is that what you had in mind? Does
it seem helpful?

  (module-language-repl-no-annotations "")
  (module-language-repl-debug-annotations ", with debugging")
  (module-language-repl-debug/profile-annotations ", with debugging
and profiling")
  (module-language-repl-test-annotations ", with test coverage")

Robby

-- 
You received this message because you are subscribed to the Google Groups 
"Racket Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to racket-users+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Re: [racket-users] Racket 6.4 very slow

[Matthew Flatt]

Here's a recap of a few points, now that I've spent some time
investigating the issue.


Replicating the problem and detecting debugging mode:

If you start DrRacket fresh or create a new tab and paste

(define (total n)
  (for/sum ([x (in-range (+ 1 n))]) x))
(time (total 10))

into the interactions window, then it runs at non-debugging speed.
After clicking "Run" (with an empty program), the interactions window
runs the example with debugging speed.

That difference seems confusing, and it might explain why some had
trouble seeing the example run slowly.

In any case, I agree with the suggest that making DrRacket more clearly
display its current mode could be helpful.


Performance with debugging enabled:

I've made some compiler improvements that cut the debugging-mode time
for the example in half.

I don't think a factor of 10 is completely out of line for a program
that counts up to a billion and adds a pair of small numbers along the
way. Those are such small and optimizable steps that demanding extra
transparency is bound to get in the way. More typical/useful programs
see a smaller cost, because they use larger primitives and spend more
time in libraries.

Even for traditional Scheme benchmarks --- which are written using
minimal bits of the language so that there's little opportunity to hide
behind macro expansions or libraries --- most programs run in debug
mode with slowing factors in the x1.5 to x4 range. That seems
consistent with my experience running unoptimized C for debugging.


Improving the way debugging instrumentation works:

Since the example uses `for/sum`, it should benefit from spending its
time in a library abstraction that is not in debugging mode, as it did
in v6.2 and earlier.

I have an idea for restoring v6.2-like debugging, and I've replied to
Vincent's `syntax-original?` question on the dev list. (Everyone is
welcome to participate there, of course).

-- 
You received this message because you are subscribed to the Google Groups 
"Racket Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to racket-users+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Re: [racket-users] Racket 6.4 very slow

[Vincent St-Amour]

On Wed, 02 Mar 2016 22:23:29 -0600,
Matthew Flatt wrote:
> Instead of using the existence of a source location to determine where
> to add instrumentation, debugging should be based on the details of the
> source location. I'm not immediately sure of the right rule, but I'll
> work on it.

Would `syntax-original?` help here?

Vincent

-- 
You received this message because you are subscribed to the Google Groups 
"Racket Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to racket-users+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Re: [racket-users] Racket 6.4 very slow

[Neil Van Dyke]

Alex Harsanyi wrote on 03/02/2016 11:19 PM:
If it cannot be improved, perhaps a warning message should be printed 
in the eval window...


I like this idea.  Maybe add to the DrRacket REPL banner, the 
debugging/instrumentation options that are enabled.



Welcome to DrRacket, version 6.4 [3m].
Language: racket; memory limit: 128 MB.
>


Neil V.

--
You received this message because you are subscribed to the Google Groups "Racket 
Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to racket-users+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Re: [racket-users] Racket 6.4 very slow

[Matthew Flatt]

I see that there's a big difference in the effect of debugging mode for
this example in v6.3-v6.4 compared to earlier versions.

On my machine:

   Racket   DrRacket
 with debugging
 v6.2 ~2500 ms  ~3800 ms
 v6.4 ~2500 ms ~63000 ms

In both versions, debugging instrumentation is applied at the
granularity of terms that have source locations. The idea was that
having a source location is a good indication that the code is
something the user wrote, and not in a library.

Starting with v6.3, however, compiled bytecode preserves source
locations for syntax literals --- so, code in a macro implementation
has a source location when it appears in an expansion. With v6.3 and
later, debugging instrumentation is added at many more program points
in the expansion of the `for` macro.

Instead of using the existence of a source location to determine where
to add instrumentation, debugging should be based on the details of the
source location. I'm not immediately sure of the right rule, but I'll
work on it.

Another direction is to improve the overhead of the debugging
instumentation, even when it is applied at a fine granularity. I'm not
clear on why it's so expensive in this case, and I think it should be
easy to improve (but I might be wrong about that).


At Wed, 2 Mar 2016 22:06:54 -0500, Matthias Felleisen wrote:
> 
> 1. Don’t ever measure anything in drracket (version 6.4.1 or earlier). 
> 
> 2. Are you comparing two different installations of Racket like the OP does? 
> Or are you just saying something is slow sometimes? 
> 
> 
> 
> 
> > On Mar 2, 2016, at 8:21 PM, Alex Harsanyi  wrote:
> > 
> > I have the same problem with Racket 6.4 64Bit Windows (running on Windows 
> > 7):
> > 
> >> (time (total 10))
> >  cpu time: 85520 real time: 88070 gc time: 204
> > 
> > BTW, my "Choose Language" dialog box (Ctrl+L) has the following selected:
> > 
> >  * Debugging
> >  * Populate "compiled directories"
> >  * Preserve stack trace
> >  * Enforce constant definitions
> > 
> > When disabling the "Debugging", the time goes down significantly:
> > 
> >> (time (total 10))
> >  cpu time: 4555 real time: 5036 gc time: 63
> > 
> > If I also disable the "Preserve stack trace option", the time goes down a 
> bit more:
> > 
> >> (time (total 10))
> >  cpu time: 3619 real time: 3662 gc time: 31
> > 
> > I did not expect the debugging option to have such a huge impact.  I don't 
> even know what it does, as there is a separate "Debug" button in the toolbar, 
> which seems to work even when the option is disabled.
> > 
> > BTW, I ran raco setup as suggested in a previous message and after that, 
> DrRacket would not start up anymore [1] and I had to do a fresh install. 
> Racket was installed from the default installer package from racket-lang.org. 
> No additional packages added and the installation was not modified in any 
> way. 
> > 
> > Best Regards,
> > Alex.
> > 
> > [1] 
> https://drive.google.com/file/d/0B5h4XOdkim72bm04eDBGNWpFZ0E/view?usp=sharing
> > 
> > -- 
> > You received this message because you are subscribed to the Google Groups 
> "Racket Users" group.
> > To unsubscribe from this group and stop receiving emails from it, send an 
> email to racket-users+unsubscr...@googlegroups.com.
> > For more options, visit https://groups.google.com/d/optout.
> 
> -- 
> You received this message because you are subscribed to the Google Groups 
> "Racket Users" group.
> To unsubscribe from this group and stop receiving emails from it, send an 
> email to racket-users+unsubscr...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

-- 
You received this message because you are subscribed to the Google Groups 
"Racket Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to racket-users+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Re: [racket-users] Racket 6.4 very slow

[Alex Harsanyi]

On Thursday, March 3, 2016 at 11:06:56 AM UTC+8, Matthias Felleisen wrote:

> 1. Don’t ever measure anything in drracket (version 6.4.1 or earlier). 

Perhaps this explains why you could not reproduce the problem :-)

DrRacket is the tool most people use to write Racket programs, and I suspect a 
lot of the time the only place where they run their programs.  Most people will 
judge the speed of Racket by how fast programs run inside DrRacket.

Also, the OP was not talking about shaving a few milliseconds, or even a 2x 
slowdown, the time difference is huge (20x).  This slowdown is not expected by 
anyone who has some familiarity with other programming environments.  If it 
cannot be improved, perhaps a warning message should be printed in the eval 
window...

> 2. Are you comparing two different installations of Racket like the OP does? 
> Or are you just saying something is slow sometimes? 

My timings are from the same racket installation, Racket 6.4 64 bit Windows 7, 
default install.  Just to recap, for the same code example as the OP presented, 
running in DrRacket:

  * "Debug" and "Preserve stack trace" enabled -> 85 seconds
  * "Debug" disabled, "Preserve stack trace" enabled -> 4.5 seconds
  * "Debug" and "Preserve stack trace" disabled -> 3.6 seconds.

> 
> > On Mar 2, 2016, at 8:21 PM, Alex Harsanyi  wrote:
> > 
> > I have the same problem with Racket 6.4 64Bit Windows (running on Windows 
> > 7):
> > 
> >> (time (total 10))
> >  cpu time: 85520 real time: 88070 gc time: 204
> > 
> > BTW, my "Choose Language" dialog box (Ctrl+L) has the following selected:
> > 
> >  * Debugging
> >  * Populate "compiled directories"
> >  * Preserve stack trace
> >  * Enforce constant definitions
> > 
> > When disabling the "Debugging", the time goes down significantly:
> > 
> >> (time (total 10))
> >  cpu time: 4555 real time: 5036 gc time: 63
> > 
> > If I also disable the "Preserve stack trace option", the time goes down a 
> > bit more:
> > 
> >> (time (total 10))
> >  cpu time: 3619 real time: 3662 gc time: 31
> > 
> > I did not expect the debugging option to have such a huge impact.  I don't 
> > even know what it does, as there is a separate "Debug" button in the 
> > toolbar, which seems to work even when the option is disabled.
> > 
> > BTW, I ran raco setup as suggested in a previous message and after that, 
> > DrRacket would not start up anymore [1] and I had to do a fresh install. 
> > Racket was installed from the default installer package from 
> > racket-lang.org. No additional packages added and the installation was not 
> > modified in any way. 
> > 
> > Best Regards,
> > Alex.
> > 
> > [1] 
> > https://drive.google.com/file/d/0B5h4XOdkim72bm04eDBGNWpFZ0E/view?usp=sharing
> > 
> > -- 
> > You received this message because you are subscribed to the Google Groups 
> > "Racket Users" group.
> > To unsubscribe from this group and stop receiving emails from it, send an 
> > email to racket-users+unsubscr...@googlegroups.com.
> > For more options, visit https://groups.google.com/d/optout.

-- 
You received this message because you are subscribed to the Google Groups 
"Racket Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to racket-users+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Re: [racket-users] Racket 6.4 very slow

[Matthias Felleisen]

1. Don’t ever measure anything in drracket (version 6.4.1 or earlier). 

2. Are you comparing two different installations of Racket like the OP does? Or 
are you just saying something is slow sometimes? 




> On Mar 2, 2016, at 8:21 PM, Alex Harsanyi  wrote:
> 
> I have the same problem with Racket 6.4 64Bit Windows (running on Windows 7):
> 
>> (time (total 10))
>  cpu time: 85520 real time: 88070 gc time: 204
> 
> BTW, my "Choose Language" dialog box (Ctrl+L) has the following selected:
> 
>  * Debugging
>  * Populate "compiled directories"
>  * Preserve stack trace
>  * Enforce constant definitions
> 
> When disabling the "Debugging", the time goes down significantly:
> 
>> (time (total 10))
>  cpu time: 4555 real time: 5036 gc time: 63
> 
> If I also disable the "Preserve stack trace option", the time goes down a bit 
> more:
> 
>> (time (total 10))
>  cpu time: 3619 real time: 3662 gc time: 31
> 
> I did not expect the debugging option to have such a huge impact.  I don't 
> even know what it does, as there is a separate "Debug" button in the toolbar, 
> which seems to work even when the option is disabled.
> 
> BTW, I ran raco setup as suggested in a previous message and after that, 
> DrRacket would not start up anymore [1] and I had to do a fresh install. 
> Racket was installed from the default installer package from racket-lang.org. 
> No additional packages added and the installation was not modified in any 
> way. 
> 
> Best Regards,
> Alex.
> 
> [1] 
> https://drive.google.com/file/d/0B5h4XOdkim72bm04eDBGNWpFZ0E/view?usp=sharing
> 
> -- 
> You received this message because you are subscribed to the Google Groups 
> "Racket Users" group.
> To unsubscribe from this group and stop receiving emails from it, send an 
> email to racket-users+unsubscr...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

-- 
You received this message because you are subscribed to the Google Groups 
"Racket Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to racket-users+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Re: [racket-users] Racket 6.4 very slow

[Matthias Felleisen]

Something is wrong with your installation. 6.4 is in the same ball park as 6.1 
Can you share the code that runs 20x slower? Thanks — Matthias





> On Mar 1, 2016, at 10:51 PM, vkelmenson via Racket Users 
>  wrote:
> 
> I recently downloaded Racket version 6.4. I have previously been using 
> version 6.1. It is much slower than version 6.1. Several short functions run 
> approx 20 times slower on 6.4 than 6.1. These were run at the same time on 
> the sam matine in succession.
> I am using Mac osX version 10.8.5
> Has anyone else noticed this? 
> 
> -- 
> You received this message because you are subscribed to the Google Groups 
> "Racket Users" group.
> To unsubscribe from this group and stop receiving emails from it, send an 
> email to racket-users+unsubscr...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

-- 
You received this message because you are subscribed to the Google Groups 
"Racket Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to racket-users+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[racket-users] Racket 6.4 very slow

[vkelmenson via Racket Users]

I recently downloaded Racket version 6.4. I have previously been using version 
6.1. It is much slower than version 6.1. Several short functions run approx 20 
times slower on 6.4 than 6.1. These were run at the same time on the sam matine 
in succession.
I am using Mac osX version 10.8.5
 Has anyone else noticed this? 

-- 
You received this message because you are subscribed to the Google Groups 
"Racket Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to racket-users+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.