[squeak-dev] a small issue with updating the trunk

Hi,

I just updated my "always up-to-date trunk image" and noticed that the
progress bar box appearing would mention it downloads some packages
that I know *for sure* are already installed.

Updating *again*, immediately after that first update, would again
mention some packages that should just have been installed.

It doesn't seem to be a big problem (no errors or anything), but it
just doesn't feel right. Can anyone confirm this behaviour? Even more
important: what can be done about it?

Best,

Michael

Michael Haupt wrote:
> It doesn't seem to be a big problem (no errors or anything), but it
> just doesn't feel right. Can anyone confirm this behaviour? Even more
> important: what can be done about it?

Confirmed. The issue is caused by the
MCConfiguration>>updateFromRepositories which seems to be forcefully
downloading newer packages even if they are loaded already. The only
thing I know to do about it is to update the last configuration map to
include all current packages.

Cheers,
   - Andreas

Hi Andreas,

On Fri, Jul 31, 2009 at 9:08 AM, Andreas Raab<[hidden email]> wrote:
> Confirmed. The issue is caused by the
> MCConfiguration>>updateFromRepositories which seems to be forcefully
> downloading newer packages even if they are loaded already. The only thing I
> know to do about it is to update the last configuration map to include all
> current packages.

could you please briefly fill me in about what a configuration map is?
I have only ever used Monticello with single packages, and agree that
this is a gap in my knowledge. :-) Is it some kind of "milestone" with
a set of concrete versions for all packages?

As for the issue with updateFromRepositories... I haven't looked at
the code yet, but wouldn't it be possible as well to check for the
version (the image knows about installed versions, right?) and avoid
downloading it if it is already there?

Best,

Michael

Michael Haupt wrote:
> could you please briefly fill me in about what a configuration map is?
> I have only ever used Monticello with single packages, and agree that
> this is a gap in my knowledge. :-) Is it some kind of "milestone" with
> a set of concrete versions for all packages?

That is exactly right. For example,
http://source.squeak.org/trunk/update-ar.1.mcm contains this:

repository ('http://source.squeak.org/trunk')
dependency ('39Deprecated' '39Deprecated-md.11'
'861c35c4-aaa0-11da-9ffe-000d933a223c')
dependency ('Balloon' 'Balloon-ar.13'
'bdaaa697-27c2-4049-a7e8-13e44a41f929')
dependency ('Collections' 'Collections-edc.91'
'7ade7e28-a6ba-4a06-97b7-da8f9623686a')
... etc ...
>
> As for the issue with updateFromRepositories... I haven't looked at
> the code yet, but wouldn't it be possible as well to check for the
> version (the image knows about installed versions, right?) and avoid
> downloading it if it is already there?

I was just looking at this and it was a bit more tricky. The method
#updateFromRepositories can be used in situations where the package
isn't even in the image and so the only way to obtain the information
then is to download it. But I've just added a little optimization that
will simply check if the version that we are trying to download is
already present in the image and if so, use that instead. This works
fine for the update process and doesn't affect the other uses.

Cheers,
   - Andreas

Hi Andreas,

On Sat, Aug 1, 2009 at 5:29 AM, Andreas Raab<[hidden email]> wrote:
>> Is it some kind of "milestone" with
>> a set of concrete versions for all packages?
>
> That is exactly right.

wonderful, thanks for clarifying this.

>> As for the issue with updateFromRepositories...
>
> I was just looking at this and it was a bit more tricky. The method
> #updateFromRepositories can be used in situations where the package isn't
> even in the image and so the only way to obtain the information then is to
> download it. But I've just added a little optimization that will simply
> check if the version that we are trying to download is already present in
> the image and if so, use that instead. This works fine for the update
> process and doesn't affect the other uses.

Cool, thanks again.

Best,

Michael