Windows CE and minimal image

John.Maloney at disney.com John.Maloney at disney.com
Mon May 22 18:33:50 UTC 2000


Lars,

If I understand you, you started with the same 970KB image that
you got by doing "majorShrink" and "abandonSources". You ran it
under both WinCE and Windows 98/NT. When running under Win98/NT,
you can add methods to TestClass and see the right code and
correct variable names when you browse those methods. But if
you add the same methods to TestClass under WinCE, the code
shown by the browser is messed up in various ways.

Is all this correct?

You might check to see if there is an old .changes file
lying around on the WinCE machine. I'm not sure what happens
if you start up an image in which you've done "abandonSources"
and there happens to be a matching .changes file. I've never
tried it.

This idea is just a shot in the dark. But please do let me know
if I've paraphrased the problem description correctly and we'll
look into it.

	-- John


At 9:45 AM -0400 5/22/00, Lars Nilsson wrote:
>Hi all,
>
>I've been trying to get Squeak to run properly on a Windows CE machine (an
>unexpanded IBM z50, meaning it has only 16MB altogether, split 50/50 between
>storage and program execution).
>
>In order to have room for it at all I've been trying to make a minimal
>image. After updating to the latest changeset I am succesful in issuing the
>commands
>
>	Smalltalk majorShrink.
>	Smalltalk abandonSources.
>
>after changing
>
>	SystemDictionary>>version
>		^ EToySystem version , ' of ' , EtoySystem versionDate
>
>to
>
>	SystemDictionary>>version
>
>		^ 'Squeak2.8alpha of 5 January' "or some similar string"
>
>After this change the first two commands complete successfully. What I have
>at this point is an image of about 970KB, which fits very well into the z50.
>
>After copying the required files (the executable, dlls, image and the now
>very small change file) I find myself with something that I can go through
>in a browser without problem. Everything appears to be correct.
>Unfortunately something is not working correctly as the following steps will
>indicate:
>
>	1. Add a class, TestClass, without any instance or class variables
>
>	2. Add a method, test:and:
>
>		TestClass>>test: aNumber and: anotherNumber
>			| x y z |
>			x _ aNumber.
>			y _ x + anotherNumber.
>			z _ x / y.
>			^ z
>
>[So far everything looks ok]
>
>	3. Add a couple of instance variables to TestClass, say foo, bar and baz
>
>If I now pull up the test:and: message in the Browser, I don't see what I
>originally typed in. My previous tests (which I've done 2-3 times) has
>yielded something that looks similar to my code in that the actual
>functionality looks the same, but variable names are completely messed up
>(basically random sequences of characters). When I tried it again over the
>weekend I got something that looked somewhat like
>
>	TestClass>>test: ajkhd and: oieui
>		| jhgj fddfj sdhgf |
>		^ nil
>
>In other words, not even the code appeared to be the same. But, when
>attempting to run this message as 'TestClass new test: 1 and: 3' I ended up
>with '(1/3') as the result, indicating that the actual code is still there
>somewhere and it is only the decompilation that is not working properly.
>
>My experiments so far has not made it possible for me to reproduce this with
>the same image running under Squeak for Windows NT/95/98, so it could either
>mean that it's a bug in the VM itself, or perhaps that there is an
>initialization problem which happens to to not occur on my Windows machines
>(perhaps due to the difference in the processors Intel/AMD vs MIPS). The
>only thing I have yet to do is to pull the resulting image back to my
>Windows machine and see if the message comes out correctly when Browsing it
>under Windows 98 (unfortunately I don't have the device here at work, so it
>will have to wait until I get home).
>
>If anyone at all has any idea what the cause of this might be I would be
>extremely happy to hear about it. Because of being unable to reproduce the
>problem under Windows 98/NT I am leaning towards the VM as I said. I'm not
>familiar enough with the innards of the VM and I don't even know to what
>extent decompilation directly involves the VM. Heck, I'm not even sure how
>decompilation happens in the image itself.
>
>Thanks in advance,
>Lars Nilsson
>PS. It is not critial to me personally that Squeak works under Windows CE,
>but I would much prefer it to using PocketScheme, which up to this point is
>the only language I can stand using on this device, besides Squeak.






More information about the Squeak-dev mailing list