[Cuis-dev] Saving a CodePackage and determining its class when installing it

[Nicolás Papagna Maldonado]

Thanks, Juan!

It's working like a charm now.

On Mon, Oct 26, 2020 at 11:14 AM Juan Vuletich <juan at jvuletich.org> wrote:

> Yes, same fix there. I just pushed it to GitHub. I hope that's all, but if
> you find further trouble, please tell, and let's fix them.
>
> Thanks,
> On 10/26/2020 1:44 AM, Nicolás Papagna Maldonado via Cuis-dev wrote:
>
> Hi Juan,
>
> No worries! FakePackageClass is a hack until I have time to write some
> tests and be able to break the dependencies and inject Smalltalk and
> SystemOrganization into CodePackage instances, so these can be controlled
> from tests at will.
>
> I gave your fix a try but my package is still installed as an instance of
> FakeCodePackage.
>
> After some debugging, it looks like CodePackageFile>>install is checking
> whether any of the classes to be installed is a direct (just checks one
> level above) subclass of CodePackage, and uses it to create the code
> package.
> Here's an extract of that method that (I believe) exposes that behavior:
>
> *pckClass _ CodePackage*.
>   classes do: [ :ee |
>     (ee hasDefinition and: [*ee superclassName = 'CodePackage'*]) ifTrue:
> [
>       ee fileInDefinitionAndMetaclass.
>       *pckClass _ Smalltalk at: ee name* ]].
>
> From what I understand, the method I mentioned in my previous email is
> used in the postPackageInstall phase, but this one happens before that.
>
> Should it be fixed too?
>
> Thanks for the help.
>
> Cheers!
> Nico PM
>
> On Sun, Oct 25, 2020 at 9:24 PM Juan Vuletich <juan at jvuletich.org> wrote:
>
>> On 10/25/2020 12:04 PM, Nicolás Papagna Maldonado via Cuis-dev wrote:
>>
>> Hi Folks,
>>
>> For some experiments I'm doing, I've created a subclass of CodePackage
>> (let's call it FakeCodePackage) to make my tests run fast that avoids
>> accessing Smalltalk or SystemOrganization.
>> When I filed-out the package and installed it in a fresh image, to my
>> surprise, the newly installed package was an instance of FakeCodePackage.
>>
>> After poking around, I found this method in CodePackage that is used
>> when saving an instance:
>>
>> codePackageClass
>>   "Answer the specific CodePackage subclass to use."
>>
>>   self class == CodePackage ifFalse: [
>>     ^ self class ].
>>   self classesDo: [ :cls |
>>     (cls inheritsFrom: CodePackage)
>>       fTrue: [ ^ cls ]].
>>   ^ nil
>>
>> This means that whenever there is a subclass of CodePackage in the
>> package being saved, It is assumed that is the class that should be used to
>> re-creating the CodePackage instance when installing it.
>>
>> In my case, FakeCodePackage is not fully functional as I use it only in
>> my tests, so I was unable to either delete the package or re-install it
>> because of all sorts of errors.
>> Luckily this was a fresh image, so I had nothing to lose :)
>>
>> I was wondering what is the use case for that, and if it still
>> relevant/in use.
>>
>> Cheers!
>> Nico PM
>>
>>
>> Too bad you found this bug. I hadn't thought of people writing subclasses
>> of CodePackage, unless for instantiation of the package itself. This is
>> used by ColorExtrasPackage and MorphicMisc1Package. It is only used to run
>> #prePackageInstall and #postPackageInstall, and I don't think it is a clean
>> solution. Anyway, I just posted a fix to that method, to also check for the
>> package name.
>>
>> Thanks,
>>
>> --
>> Juan Vuletichwww.cuis-smalltalk.orghttps://github.com/Cuis-Smalltalk/Cuis-Smalltalk-Devhttps://github.com/jvuletichhttps://www.linkedin.com/in/juan-vuletich-75611b3
>> @JuanVuletich
>>
>>
>
> --
>
> Nicolás Papagna
>
>
>
> --
> Juan Vuletichwww.cuis-smalltalk.orghttps://github.com/Cuis-Smalltalk/Cuis-Smalltalk-Devhttps://github.com/jvuletichhttps://www.linkedin.com/in/juan-vuletich-75611b3
> @JuanVuletich
>
>

-- 

Nicolás Papagna
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cuis.st/mailman/archives/cuis-dev/attachments/20201026/e7eeddcb/attachment.htm>