Raspberry Pi 3 Model B+ (2017)

62 nanoseconds
1.11 microseconds
1.37 microseconds

SqueakJS (Squeak 5.0) on my machine

2 microseconds
24 microseconds
28 microseconds

**

What are better ways to benchmark "Error, Halt, Warning" than via #bench?

Best,
Marcel
Am 30.10.2019 09:25:10 schrieb Marcel Taeumel <marcel.taeumel at hpi.de>:
Hi Levente,

that's correct. Thanks that reminder. So we need a benchmark on a slower machine to justify that class variable. Until then, your speculation is as good as my fast machine. :-D

I do like the point of the extensibility through a class variable though...

Best,
Marcel
Am 29.10.2019 16:20:33 schrieb Levente Uzonyi <leves at caesar.elte.hu>:
On Tue, 29 Oct 2019, Marcel Taeumel wrote:

> Hi Christoph.
> > However, currently, a #deprecated call in a drawing method makes the image unusable ...
>
> Yes, there is a lot of room for improvement in Squeak's strategy to keep the image alive and in a recoverable state. Annoying things include errors in event handlers (such as #mouseOver:) and the thing about Warnings you
> mentioned for #displayWorldSafely.
>
> Since the exception mechanism can be used to implement dynamic scope (see CurrentReadOnlySourceFiles), we should never catch Exception in general to install such handlers.
>
> Warning might be a nice addition to #displayWorldSafely.
>
> [Error] bench '219,000,000 per second. 4.57 nanoseconds per run.'
> [Error, Halt] bench '11,900,000 per second. 84.4 nanoseconds per run.'
> [Error, Halt, Warning] bench '9,910,000 per second. 101 nanoseconds per run.'

#bench is not suitable to measure things that are fairly quick, because []
bench doesn't give 0 ns. It's not too far off, but definitely not
accurate, especially if you compare ratios.

#bench doesn't measure the effects of allocation/gc well. The last two
code puts pressure on the garbage collector.

>
> Since this is called only once to fourteen times per frame (see DamageRecorder) -- but usually only once -- we are good with 100 nanoseconds, given that 60 frames-per-second translate to 16 666 667 nanoseconds per frame.

Not all machines (physical and virtual) running Squeak are as fast as
yours. What's 100ns on your machine, may take up to 3 magnitudes longer
on others. In that case, you only have 16 666 microseconds to render, and
you spend 100 microseconds on something that could be almost 0.

If there were a class variable holding the exception set, it could be
reused with no additional cost. It would also let users to add their own
exceptions to the set without modifying Trunk code.

Levente

>
> Best,
> Marcel
>
> Am 25.10.2019 20:46:52 schrieb Thiede, Christoph :
>
> I see, but isn't it just their existence what we want to test for to prevent from loads of Debuggers appearing?
>
> Or should we go the other way around and signal a "DebuggerRaisedNotification" each time before opening a debugger?
>
>
> However, currently, a #deprecated call in a drawing method makes the image unusable ...
>
> _________________________________________________________________________________________________________________________________________________________________________________________________________________________________
> Von: Squeak-dev im Auftrag von Taeumel, Marcel
> Gesendet: Mittwoch, 23. Oktober 2019 15:48:15
> An: John Pfersich via Squeak-dev
> Betreff: Re: [squeak-dev] The Trunk: System-mt.1093.mcz
> > Why isn't it sufficient to test for UnhandledError instead? Otherwise, we would also need to test for Warning etc. ...
> UnhandledError and UnhandledWarning are (private) implementation details of Squeak's exception handling mechanism. They should never be exposed to (or used by) applications/frameworks.
>
> Best,
> Marcel
>
> Am 22.10.2019 18:50:59 schrieb Thiede, Christoph :
>
> Why isn't it sufficient to test for UnhandledError instead? Otherwise, we would also need to test for Warning etc. ...
>
> _________________________________________________________________________________________________________________________________________________________________________________________________________________________________
> Von: Squeak-dev im Auftrag von Taeumel, Marcel
> Gesendet: Dienstag, 15. Oktober 2019 13:18:26
> An: gettimothy via Squeak-dev
> Betreff: Re: [squeak-dev] The Trunk: System-mt.1093.mcz
> Yeah, I wonder whether we should expand "Error" to "Error, Halt"?
> Best,
> Marcel
>
> Am 15.10.2019 12:45:42 schrieb Balázs Kósi :
>
> Hi Hannes!
>
> This morning I've just run into this exact same situation: putting a halt into a method, called by a morph's #drawOn:
> makes the image unusable.
>
> The problem stems from WorldState >> displayWorldSafely: being safe only for Errors and not for other kind of
> Exceptions, and Halt being only an Exception not an Error.
>
> For a quick fix add Halt to the handled exceptions in #displayWorldSafely:
> [aWorld displayWorld. finished := true] on: Error, Halt do: [:ex |
> Balázs
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.squeakfoundation.org/pipermail/squeak-dev/attachments/20191030/f90b49fa/attachment-0001.html>