Magma....

Ramiro Diaz Trepat ramiro at novamens.com
Thu Jul 8 18:10:03 UTC 2004


    Chris wrote to me yesterday.  He told me all this problems were 
caused because of a flaw in the current Squeak VM.  And it will persist 
until the new 3.7 VM is out. 
    He gently attached the previous VM (3.5) for me to try everything 
again.  Although I still didn't find time to do so, I certainly will, 
most probably this weekend.  I encourage others haveing problems (like 
Kamil) to do so as well.  Nevertheless I will report how did it all go 
next week.
    I don't really know why he did not reply this important matters to 
the list.  Is he shy or is he sick of the ghetto ? :)
    Anyway, here is a fragment of what he wrote to me:

> Hi Ramiro!  Unfortunately, the latest version of Magma has been somewhat in
> limbo while we wait for the next "official" 3.7VM to be built and released by
> the Squeak community.  This VM will have support for the method which is
> failing.
> 
>   Array>>elementsForwardIdentityTo:copyHash:
> 
> To make a long story short, a long time ago during development of Magma, I
> found a "bug" in Squeak that can be demonstrated by the following line:
> 
>   String new becomeForward: #size
> 
> This used to lock up the image because the identityHash of the new String was
> being applied to that of the Symbol #size.  This wreaks havoc on the system
> because #size, which is a system-critical method, could no longer be found in
> the MethodDictionary's.
> 
> I appealed to the Squeak community to not copy the identityHash, and someone
> graciously answered the call and built a new primitive for becomeForward that
> doesn't copy the hash.  However, since those changes were made, no new
> "official" VM has been built with the new primitive, so you have to either
> build your own or use the one I've attached.





Kamil Kukura wrote:

>> I have yet not worked *extensively* using Magma - but I *do* have a
>> fully working web app using it with a Magma-local db. The key here is
>> probably that I use 3.6-5424 + Magma 1.0gamma6.
>> And given that I have had no problems at all, it works smoothly.
>
>
> Lucky you :) I just tried to put into session's root one 
> MagmaCollection with just one object and I got into situation where 
> almost any operation with Magma is causing deadlock. Interrupt shows 
> it is something with submitGuard semaphore.
>
> I wish I'd have weeks or maybe months for debugging this miracle but I 
> think I'll give GOODS a try. Hopefully I'll find out how to use BTree.
>




More information about the Squeak-dev mailing list