Hi Marcus!
Marcus Denker denker@iam.unibe.ch wrote:
On 30.03.2006, at 14:53, goran@krampe.se wrote:
and SqueakMap is broken for 3 Months.
"broken" is a bit harsh - yes, it flunks in the current 3.9alpha or beta, but not in 3.8 AFAIK.
Now, if/when you fix it btw - make sure that the "SqueakMap" package (the script) picks the correct release of base and also - check that your fix works back in 3.6, 3.7 and 3.8 too.
Is this reasonable? One codebase for all Squeak versions? I mean, this
It has worked so far - and no, not for *all* versions, but from 3.5 I think. :)
would imply to add code that tests if the deprecated method has been removed or not, or it would not to test for the version (if version X, call method Y, else method Z). This is bound to become very messy over time...
Well, it seems that *in practice* it doesn't get *that* hairy - SM mostly uses "vanilla code" in the image.
Wouldn't it make more sense to freeze the external packages with the release (to be developed much more cautiously) while forking a new unstable branch for the new unstabe Squeak release?
I am not sure I parsed what you wrote - but I am guessing that, yes, when it gets too hairy the principal idea is of course to "branch" like you describe BUT... there is one "small" catch. SM is a client server system where we synhronize the model by transferring it in an ImageSegment to the clients and smacking it in. In short - having different client side code running against a server with... the latest code? Well, as you can imaging *that* gets hairy much faster I think :).
Also just trying to maintain different forks for different Squeak versions (automatically upgrading etc when people try to connect to the server with old code blabla) is not easy either.
To be backward compatible in the sense that all new versions of packages in 3.9 need to work even on old Squeak versions seems to be not really practical...
No, of course not - but as I said - SM is different in the sense that it has to support older versions of Squeak (right? :)) and it is based on the ImageSegment approach and also - it seems appropriate that the same domain code runs on the server as on the clients.
Anyway - the "one line 5 minute fix" doesn't include branching SM for 3.9. :)
Marcus
regards, Göran
PS. If you don't get around fixing the issue before say in ... 5 hours then I will fix it - Maya should be asleep then. :) ANd also - there may be other small things I should fix at the same time, we will see.