[BUG][FIX] TestRunner UI interaction

Doug Way dway at mailcan.com
Thu Mar 4 21:40:07 UTC 2004


Avi Bryant wrote:

>
> On Mar 3, 2004, at 11:28 AM, Marcus Denker wrote:
>
>>> The reason was I presume that SUnit was in the image - but not as an SM
>>> package.
>>> Now with Marcus update - it is installed as a package, and thus will be
>>> independently upgradeable.
>>>
>> Shouldn't  a "update code from server" then do both a update wrt. to 
>> the update
>> stream and tell SM to install all updated packages?
>
>
> No, I think we should explicitly include instructions in the update 
> stream to pull in specific new versions from SM.  That way you'll 
> always get the right version for the update level you are at, and 
> it'll always come at the right point in the stream.  If you just 
> always pull in the latest version you're bound to run into trouble 
> (particularly if you're updating an image that hasn't been updated for 
> a while).


In the short term at least, I agree with Avi.  I don't think there's a 
good way to have a mixture of an update stream and updating from 
packages, and have consistent results.  So for now, we have updates 
which specifically install certain versions of SM packages.  These are 
one-liner changesets, so they're not too complex.  (And for now, we want 
all of these packages to be on the SM server or at least backed up on 
the SM server... we don't want the update stream to depend on loading 
some package on Joe's geocities website which has a 70% uptime.)

Farther in the future, we could have a way to build up an image (even an 
alpha image, I suppose) just from SM packages without an update stream.  
Or, we might still use an update stream.  We'll see which is most 
practical once the image is broken down into packages and we have the 
dependency scheme in place...

> These updates should be "fast tracked" into the stream, of course - if 
> the maintainer of a base package says that a new version needs to go 
> in, we probably don't need a lengthy review process to approve this.


Yes... if it's the official maintainer of a package posting an new 
version for the update stream, that could probably be a self-approval, 
since we're implicitly trusting the maintainer.

- Doug





More information about the Squeak-dev mailing list