On 23-07-2015, at 8:03 PM, Chris Muller asqueaker@gmail.com wrote:

...

Excellent - one tiny change needs to be made to let it work on a Pi2 though; the shell script test for ARMv6l, which is correct for PiA+, B * B+ but the Pi2 is ARMv7. With that changed it fired up on my pi2 ok.

Is it case-sensitive? Mine reported armv6l, not ARMv6l. If it is, I'll bet on consistency, if it isn't, lowercase won't matter. I added a check for armv7l.

`uname -m` seems to return all lower case ‘armv7l’ on my particular Pi2 so I imagine that is typical.

...

...

• Look at the image size, its DOUBLE the size of the 4.6 image. Why?

Space analysis on my Pi2 shows a lot of MC* objects, which as we’ve seen before seem to hold onto a lot of the 4Mb of String and quite a few arrays, symbols etc listed.

I can't believe MCFileBasedRepository flushAllCaches is not part of #prepareNewRelease (which makes me wonder whether 4.6 may have a bunch of cached versions.. please no!). I'm adding it.

What I’m a bit puzzled about is that I have now tried out an assortment of MC reduction stuff - MCFileBasedRepository flushAllCaches. MCDefinition clearInstances. MCWorkingCopy stubAllAncestryForScratchRelease. MCWorkingCopy allManagers do: [:wc | wc unregister]. Project allSubInstancesDo:[:pr| pr resourceManager reset]. for example and it isn’t making any change to the image size at all so far as I can tell. I’ve even whacked the stubAncestry to to cut back to a single stub instead of 10 and no effect can be seen. Sumfing not right here...

...

Oh, and I noticed in the Contents/Resources directory that there seems to be a spare _test.[image|changes] pair. Presumably not actually wanted?

Oops..

Many, many oops. Around 300,000 in that image :-)

tim -- tim Rowledge; tim@rowledge.org; http://www.rowledge.org/tim Strange OpCodes: WAF: Warn After the Fact

StackMorph initializion causes a DNU in Objects tool Presentation category.

:-(

I don't really understand why this does not work now, what can have changed with this tool lately ??

On Thu, Jul 23, 2015 at 7:41 PM, karl ramberg karlramberg@gmail.com wrote:

Great work :-)

It seems control and alt keys are not swapped so out of the box you have to use alt+c to copy text on Windows.

Karl

On Thu, Jul 23, 2015 at 5:55 PM, Chris Muller asqueaker@gmail.com wrote:

...

I managed to successfully incorporate the ARM platform into the 5.0 All-In-One -- so Squeak 5 will now start up with one-click on FOUR platforms, OOTB..!!

I'm thankful to this community for collaborating to produce and shake out the 4.6 release. As this is a special dual release, your help is needed one more time to help shake out this 5.0 release. Would you please download this latest Squeak-5.0-All-in-One and check it out? We still have a chance to avoid obvious bugs, but we want to keep it as close to the 4.6 release as possible.

http://ftp.squeak.org/5.0/Squeak-5.0-All-in-One.zip

Here are some things I've noticed already, but I need your help to figure out how to deal with them:

• someone said the VM is crashing in the Jenkins build due to

ImageSegments tests. Do we want to disable those tests?

• Actually, when I ran the tests in the image, there was no VM

crash, but the dialog, "Running all tests might take a some minutes" (<--- hey that's some nice grammar there Marcel! ;-) ) appeared TWICE. When I clicked "Yes" the first time they ran, "Yes" a second time, no effect. Just a strange UI bug..?

• Look at the image size, its DOUBLE the size of the 4.6 image. Why?

I'm getting excited, thank you!! Chris

On Thu, Jul 23, 2015 at 1:01 PM, karl ramberg karlramberg@gmail.com wrote:

StackMorph initializion causes a DNU in Objects tool Presentation category.

:-(

I don't really understand why this does not work now, what can have changed with this tool lately ??

Its after running the tests I think... Before running tests I think its fine.

Hi,

speedup depends on what is exercised. See my previous post. I get about 20% in my app which I'll gladly use for a higher framerate but I don't consider that dramatic. My first priority is stability. Then I'll check which parts of the program account for the improvements.

Cheers,

Herbert

Am 23.07.2015 um 20:28 schrieb tim Rowledge:

On 23-07-2015, at 11:17 AM, Herbert König herbertkoenig@gmx.net wrote:

...

Hey, great!

downloaded it on the Pi A+ installed my trusty 24/7 cat surveillance app and it worked. CPU usage went down from 50 to 40%. It exercises BitBlt, Collections, file I/O incl. GPIO, OSProcess and CommandShell Memory went up from 52M to 75M which is critical as I use the PiNoir Camera.

Well the base image is about 20M bigger than the base 4.5 one for some reason, so that explains most of it. The VM is a touch bigger since it has the Cog translation system now, so you should notice a dramatic speedup.

tim

tim Rowledge; tim@rowledge.org; http://www.rowledge.org/tim Useful random insult:- He has a tenuous grip on the obvious.

On 23-07-2015, at 2:43 PM, Levente Uzonyi leves@elte.hu wrote:

I think it's related to the segmented memory model of Spur, which uses more memmory than the the compact model of pre-Spur Squeak.

Highly unlikely to make for a 16Mb -> 35Mb jump.

tim -- tim Rowledge; tim@rowledge.org; http://www.rowledge.org/tim If you don't pay the exorcist do you get repossessed?

Hi Tim and all--

...somewhere there is 9Mb ‘wasted’ in some fashion... Run it under sim I guess and find out...

If anyone has done this, please let me know what you found, so I'll know a bit more going in. Seems like a good use for [1].

-C

[1] http://netjam.org/context/viz

-- Craig Latta netjam.org +31 6 2757 7177 (SMS ok) + 1 415 287 3547 (no SMS)

Hi,

I noticed some glitches: When I use list filtering for the instvars in an inspector the list doesn't automatically return to normal. Maybe a preference I missed. When I play with that list filtering at some point the CPU goes up to 90% and does not return until I shut down the image. The right click (personal) World menu has lost the entry to disable the main docking bar. I disabled it via preferences.

Otherwise it survived the first hour :-))

Cheers,

Herbert

Am 23.07.2015 um 20:17 schrieb Herbert König:

Hey, great!

downloaded it on the Pi A+ installed my trusty 24/7 cat surveillance app and it worked. CPU usage went down from 50 to 40%. It exercises BitBlt, Collections, file I/O incl. GPIO, OSProcess and CommandShell Memory went up from 52M to 75M which is critical as I use the PiNoir Camera.

I'll watch what comes out of Tim's (Rowledge) recent post.

And I'll watch if it runs as stable as 4.5.

Thanks,

Herbert

Am 23.07.2015 um 17:55 schrieb Chris Muller:

...

I managed to successfully incorporate the ARM platform into the 5.0 All-In-One -- so Squeak 5 will now start up with one-click on FOUR platforms, OOTB..!!

The FileList services is not initialized to read .prefs from FileList in the 5.0 image.

Do it: FileList initialize

Then you can load any saved preference file with the FileList.

Karl

On Fri, Jul 24, 2015 at 5:46 AM, Chris Muller asqueaker@gmail.com wrote:

It appears I may have had my .prefs file loaded when I ran #prepareNewRelease, which sets many preferences but not all of them. The ones it doesn't set were left to my preference. I was surprised to notice my scratch pad preference was on, it shouldn't be for the release (actually it should! but I won't go there.. :) ).

I'll have to fix this... somehow. Now we know why Bert wants to support Preferences outOfTheBox. The one use-case we're missing is, we need to be able to _generate_ a #defaultValueTableForCurrentRelease from current preference settings..

On Thu, Jul 23, 2015 at 2:09 PM, Herbert König herbertkoenig@gmx.net wrote:

...

Hi,

I noticed some glitches: When I use list filtering for the instvars in an inspector the list

doesn't

...

automatically return to normal. Maybe a preference I missed. When I play with that list filtering at some point the CPU goes up to 90% and does not return until I shut down the image. The right click (personal) World menu has lost the entry to disable the

main

...

docking bar. I disabled it via preferences.

Otherwise it survived the first hour :-))

Cheers,

Herbert

Am 23.07.2015 um 20:17 schrieb Herbert König:

...

Hey, great!

downloaded it on the Pi A+ installed my trusty 24/7 cat surveillance app and it worked. CPU usage went down from 50 to 40%. It exercises BitBlt, Collections, file I/O incl. GPIO, OSProcess and CommandShell Memory went up from 52M to 75M which is critical as I use the PiNoir Camera.

I'll watch what comes out of Tim's (Rowledge) recent post.

And I'll watch if it runs as stable as 4.5.

Thanks,

Herbert

Am 23.07.2015 um 17:55 schrieb Chris Muller:

...

I managed to successfully incorporate the ARM platform into the 5.0 All-In-One -- so Squeak 5 will now start up with one-click on FOUR platforms, OOTB..!!

Thanks for the pointers.

Am 24.07.2015 um 17:31 schrieb Chris Muller:

...

When I play with that list filtering at some point the CPU goes up to 90% and does not return until I shut down the image.

How to reproduce?

By now I think the high CPU has nothing to do with list filtering but either with the new inspector or with not enough Umph of the Pi.

To reproduce, file in the attached class, from a Browser get a sample instance.

In that inspector click on "all inst vars". Maybe I have too many too big arrays but it hasn't returned from 96% CPU since a quarter of an hour.

With that knowledge I checked the same in a 4.5 Image running on the VM that comes with the Pi. It shows the same problem.

So it's not a 5.0 or 4.6 issue. Sorry for the noise.

OTOH on my Windows laptop the same test is instantaneous. Which means it is over a Factor 1000 slower on the ARM V6. I seem to remember the Pi being a factor of 70 slower than this laptop.

Cheers,

Herbert

Hi Herbert,

first., apologies fror being unresponsive. I've had things to do for the last few days.

On Thu, Jul 23, 2015 at 11:17 AM, Herbert König herbertkoenig@gmx.net wrote:

Hey, great!

downloaded it on the Pi A+ installed my trusty 24/7 cat surveillance app and it worked. CPU usage went down from 50 to 40%. It exercises BitBlt, Collections, file I/O incl. GPIO, OSProcess and CommandShell Memory went up from 52M to 75M which is critical as I use the PiNoir Camera.

So there's two things going on here. One is that the object format is indeed less compact tan the previous one and Spur image files are about 15% larger, so a 57Mb image file should result in something around 65Mb. Another is that right now I've chosen to have the memory manager allocate a chunk of free space at startup that defaults to the growHeadroom, which defaults to 16Mb. You can see what the free old space is via vm parameter 54, e.g.

Smalltalk vmParameterAt: 54 22689208

You can change this by the command line switch -maxoldspace. Really I think there needs to be a way of a) specifying a free headroom on the command line, and b) setting a grow size in the image header, rather than from Smalltalk code.

Basically Spur is very green and we need to evolve its facilities for management of memory management. So please feel free to make suggestions and requests for functionality here, but especially on vm-dev. We could do with careful discussions here as I don't want to hack, but I'm not pretending to have the right answers.

I'll watch what comes out of Tim's (Rowledge) recent post.

And I'll watch if it runs as stable as 4.5.

Thanks,

Herbert

Am 23.07.2015 um 17:55 schrieb Chris Muller:

...

I managed to successfully incorporate the ARM platform into the 5.0 All-In-One -- so Squeak 5 will now start up with one-click on FOUR platforms, OOTB..!!

I believe every reproducible issue in this thread is addressed except the following:

I wrote:

someone said the VM is crashing in the Jenkins build due to ImageSegments tests. Do we want to disable those tests?

No tests were disabled. Are we ok?

• Look at the image size, its DOUBLE the size of the 4.6 image. Why?

The top 10 consumers of space are essentially the same in 4.6 and 5.0. So the extra space consumption must be something related to Spur internals.

Herbert wrote:

When I use list filtering for the instvars in an inspector the list doesn't automatically return to normal. Maybe a preference I missed.

Marcel (un)set this as the new default for this release. "Filterable Lists Clear Automatically".

Karl wrote:

It seems control and alt keys are not swapped so out of the box you have to use alt+c to copy text on Windows.

This is how it was in 4.6 and even 4.5. Won't change this for 5.0 at this point.

=======

I just uploaded a new 5.0 All-in-One with all the other issues addressed.

On Fri, Jul 24, 2015 at 10:33 AM, karl ramberg karlramberg@gmail.com wrote:

Disable project saving from menus until we can make it work again.

You probably have to do a "Rebuild menus" from Extras menu in the Docking bar

On Thu, Jul 23, 2015 at 5:55 PM, Chris Muller asqueaker@gmail.com wrote:

...

I managed to successfully incorporate the ARM platform into the 5.0 All-In-One -- so Squeak 5 will now start up with one-click on FOUR platforms, OOTB..!!

I'm thankful to this community for collaborating to produce and shake out the 4.6 release. As this is a special dual release, your help is needed one more time to help shake out this 5.0 release. Would you please download this latest Squeak-5.0-All-in-One and check it out? We still have a chance to avoid obvious bugs, but we want to keep it as close to the 4.6 release as possible.

http://ftp.squeak.org/5.0/Squeak-5.0-All-in-One.zip

Here are some things I've noticed already, but I need your help to figure out how to deal with them:

• someone said the VM is crashing in the Jenkins build due to

ImageSegments tests. Do we want to disable those tests?

• Actually, when I ran the tests in the image, there was no VM

crash, but the dialog, "Running all tests might take a some minutes" (<--- hey that's some nice grammar there Marcel! ;-) ) appeared TWICE. When I clicked "Yes" the first time they ran, "Yes" a second time, no effect. Just a strange UI bug..?

• Look at the image size, its DOUBLE the size of the 4.6 image. Why?

I'm getting excited, thank you!! Chris

It is probably a VM configuration.

In the Squeak.ini file set 3ButtonMouse to 1 or 0 depending on your what it is currently.

Karl

On Fri, Jul 24, 2015 at 6:50 PM, Stéphane Rollandin lecteur@zogotounga.net wrote:

Just a question:

On Windows both all-in-one VMs have the effect of preference swapMouseButtons inverted with regard to what I'm used to: when turned off the yellow button is the middle one while I would expect it to be the right one, as it is in my current VM. Both VMs say they are a Cog VM 4.5 from july 6, 2015.

Is it a new convention, or am I missing something ?

Stef

Hi Stéphane, no, 3ButtonMouse should always be 1, else right-click won't work. The 5.0-All-in-One I just uploaded today has the 4.6 preferences restored, so your usual right click should work as you expect. The one you tested which had swapMouseButtons unset was from yesterday, its now fixed in today's.

On Fri, Jul 24, 2015 at 12:33 PM, Stéphane Rollandin lecteur@zogotounga.net wrote:

...

In the Squeak.ini file set 3ButtonMouse to 1 or 0 depending on your what it is currently.

Ah, that's it. Thanks !

Stef

That's correct, because we still don't know what the final verdict is w.r.t. the SqueakSSL plugin. Eliot said he was going to try to get the plugin maintainer to merge some code that would make it know about default depths to look for Fowarders when there is a primitive failure -- something needed for Spur.

I don't know what his status is on that yet, and so we haven't been able to finish the All-in-One to the point of it being ready to ask someone to put in the Mac VM and sign it.

On Tue, Jul 28, 2015 at 8:35 AM, Bert Freudenberg bert@freudenbergs.de wrote:

On 24.07.2015, at 18:22, Chris Muller asqueaker@gmail.com wrote:

I just uploaded a new 5.0 All-in-One with all the other issues addressed.

I have not tried previous 5.0 All-in-Ones. The one downloaded today gives me this on OS X:

Not even the usual right-click-to-open works. I have to disable Gatekeeper system-wide. Not sure why that is.

With disabled gatekeeper it still does not open the default image. That’s because the Info.plist still refers to 4.6:

<key>SqueakImageName</key> <string>Squeak4.6-15102.image</string>

After fixing the Info.plist, the VM exits immediately, with this error:

This interpreter (vers. 6505) cannot read image file (vers. 6521).

Apparently the 5.0 All-in-One does not include a Spur VM for Mac?

• Bert -