Hi rabbit,


> What does an -override category name extension do for us? If the method name is overridden, unload removes the method even if it did switch categories I believe. Or is it possible to unload a specific category and leave the other category present. Yet I still don’t see how that protects an overridden method.

And this is exactly what the -override suffix prevents. :-)

Let's try it out by example:

Create a new package, let's call it MyPackage.
For this package, we want to change two methods from the base system.
Just as an example, let's say we want to change the methods #testBitmapByteAt and #testWriteOn in the class BitmapBugz.
We'll compare both types of extension methods - regular extension methods and override methods - side by side.
For that, shift-yellow-click #testBitmapByteAt, choose "change category...", and enter a new category named "*mypackage-extension".
Analogously, we put #testWriteOn into another new category named "*mypackage-override".
Now we can change both methods that have been added to our package, for instance, by adding a comment "EXTENSION" to the first method and a comment "OVERRIDE" to the second method:

[cid:a674489f-5540-4b09-a671-2cc026580d3b]

Next, go to the Monticello browser, find our package MyPackage, and unload it:

[cid:0501c835-ce63-45de-92f7-8e07b3bbe0f3]

Finally, head back to the class that we have changed, BitmapBugz.
What has happened?
The regular extension method, #testBitmapByteAt, has completely disappeared; thus, its containing package, Tests, is marked as dirty.
But the extension method, #testWriteOn, still exists, and it has been reverted to its original version again: it is back in the original category (tests), and it does not contain the comment "OVERRIDE" any longer.
Note that the prior version of the method is still available through the versions button in the system browser.

The same thing happens every time you load or unload any package that has override methods: on loading, the original methods are replaced, and on unloading, the original versions that have been remembered are installed again.

Note that the order of recategorizing and editing a method matters: if you first edit it (while it is still contained in the original package) and then move it to the extension category, Monticello will restore the wrong version of the method (including your edits) when unloading the new package.

> > E4. A postscript/preamble of removal uses the Compiler/ClassOrganizer to touch other packages, e.g., to clean up Autogenerated methods.
>
> Would it help to have a preunload and unpostscript, in the MCZ / SAR?

Monticello/MCZs do have four scripts for each package:

[cid:ddd315fa-b606-4e4a-8f35-5e69ad4de335]

The preamble is executed before the package is loaded.
The postscript is executed after the package was loaded.
The preambleOfRemoval is executed before the package is unloaded.
The postscriptOfRemoval is executed after the package is unloaded.
See the hierarchy of MCScriptDefinition for details.

I think this should meet all needs of a package creator. It's only about choosing the right scripts. As an example, there are some packages that put stuff in the Autogenerated package, but I bet that not all of them take care to remove that autogenerated stuff again before unloading themselves again.

Note that SAR files (Smalltalk ARchives) don't have package semantics but contain any installation logic themselves. See SARInstaller>>#fileInFrom:. They can have two install scripts for a preamble and a postscript but no uninstall scripts. However, they could load any package which contains a preambleOfRemoval or postscriptOfRemoval.

Another way next to Monticello scripts and SAR install scripts is to organize any loading/unloading logic in the class-side #initialize and #unload methods of any class. #initialize is sent after a package with a new definition of this method is loaded, and #unload is sent before the class is removed. See MCMethodDefinition>>#postload and Class>>#removeFromSystem:.

HTH,
Christoph


________________________________
Von: Squeak-dev <squeak-dev-bounces at lists.squeakfoundation.org> im Auftrag von rabbit <rabbit at callistohouse.org>
Gesendet: Freitag, 21. Oktober 2022 14:32:53
An: The general-purpose Squeak developers list
Betreff: Re: [squeak-dev] Packaging conventions for Objectland - The Worlds of Squeak (was: Objectland - The Worlds of Squeak)


Hi Christoph, an interesting area and I have a couple of questions and an solution suggestion.

On Oct 20, 2022, at 15:27, Thiede, Christoph <Christoph.Thiede at student.hpi.uni-potsdam.de> wrote:



Hi Dave,


>   "Reloadable means that I can completely remove a package from the

>   image, then add it back in, and everything still works."
>
> But I would also add that I expect that the process of unloading
> a package does not leave me with dirty Monticello packages.

+1. In terms of Monticello semantics, this will (or should) usually not happen. To my knowledge, there is only a small set of exceptions to this rule:

E1. You have defined extension methods in your package that should be override methods instead (i.e., you forgot the -override suffix at the end of the method category name). Once such a package is unloaded, the original methods will not be restored. Unfortunately, this seems to be a pretty unknown fact. I know of a lot of students (including my past self) that do not know this and thus accidentally make their package effectively un-unloadable because they have extended-not-overridden any important method in a system package.

What does an -override category name extension do for us? If the method name is overridden, unload removes the method even if it did switch categories I believe. Or is it possible to unload a specific category and leave the other category present. Yet I still don’t see how that protects an overridden method.

My idea here is to only unload the method version in the unloading package and make sure we leave/revert the previous version that came from no previous version of the package being unloaded that may have loaded over a previous self.

E2. You have defined overlapping packages (see below).
E3. The unloading process fails or is manually interrupted.
E4. A postscript/preamble of removal uses the Compiler/ClassOrganizer to touch other packages, e.g., to clean up Autogenerated methods.

Would it help to have a preunload and unpostscript, in the MCZ / SAR?


> If I have (for example) package MorphicExtras-Examples

Wait a minute ... Why do you have a *package* named MorphicExtras-Examples? In a current Trunk image, there is a MorphicExtras package only. MorphicExtras-Examples is just a system category in that package. And deliberately defining overlapping packages is something that I would consider unidiomatic, Monticello does not expect this. (Or was your argument that is should do?)

If I unload MorphicExtras, my image hangs indeed because unfortunately, the system (Morphic) still depends on it. In my case, because the class Command is used in BorderedMorph and many others. Improving the decoupling between this and many other pairs of packages is still an open to-do that seems to be connected with a lot of design decisions. Maybe we should increase the granularity of our PackageDependencyTest to make any progress on this issue more visible and to avoid further regressions in the form of additional coupling with two already dependent packages?

Best,
Christoph

________________________________
Von: Squeak-dev <squeak-dev-bounces at lists.squeakfoundation.org> im Auftrag von David T. Lewis <lewis at mail.msen.com>
Gesendet: Donnerstag, 20. Oktober 2022 18:33:09
An: The general-purpose Squeak developers list
Betreff: Re: [squeak-dev] Packaging conventions for Objectland - The Worlds of Squeak (was: Objectland - The Worlds of Squeak)

Objectland may have been a poor example for me to pick. It was
fresh in my mind because it is a recent addition.

In my initial mail, I said this:

  "Reloadable means that I can completely remove a package from the
  image, then add it back in, and everything still works."

But I would also add that I expect that the process of unloading
a package does not leave me with dirty Monticello packages. If I
have (for example) package MorphicExtras-Examples and I unload that
package, then the MorphicExtras package is going to be inconsistent.

I would like my hypothetical reloadable package to leave the image
in a consistent state after unloading the package, and I would
like the image to be restored to a consistent state after I reload
the package.

I understand now that making Objectland into a reloadable package
is probably not a good thing to do, for the reasons given below.
But just to work with that as an example, if my goal was to make
ObjectLand be a reloadable package, then putting in a package
called Objectland-Morphic would make this possible without affecting
the overall MorphicExtras package. But then as Marcel says, that
would clutter up the package namespace.

Maybe we do not really have the tools to support this well. Or
maybe my expectations are not realistic. Or maybe both ;-)

Dave



On Thu, Oct 20, 2022 at 09:32:28AM +0200, Marcel Taeumel wrote:
> Hi all --
>
> I like it that we now have a new "handle" for all the Morphic examples that have been living in the image for a very long time, mostly in the "MorphicExtras" package, sometimes "Etoys".
>
> Personally, I think that "MorphicExtras" and "Etoys" are the packages here that need cleaning up with the potential to unload and reload. There will always be some new examples around Morphic that need a place to live. Attaching the "Objectland" label to every tiny example thing does not feel right. "MorphicExtras-Examples" might be a more fitting label or category.
>
> "Objectland" is just one possible entry point to a selected set of examples. There can be others. The "Parts Bin" is already there, providing access to almost the same set of things. The dominant decomposition seems to be along "MorphicExtras-Examples" ... or "-Demo" ... I think ... "Objectland" and "Partsbin" are cross-cutting.
>
> Best,
> Marcel
> Am 20.10.2022 02:32:04 schrieb David T. Lewis <lewis at mail.msen.com>:
> I am a huge fan of Objectland, and I am also a big proponent of reloadable
> packages. Reloadable means that I can completely remove a package from the
> image, then add it back in, and everything still works.
>
> With the recent addition of Objectand to trunk (yay!) I want to also note
> that this seems like a great candidate for a reloadable package. After all,
> we just loaded it, so we know that part works. All we need to do is make
> sure we can unload it and then put it back in.
>
> So this leads to a question - if we want this to be reloadable, then
> what should be the package name? I am thinking that 'Objectland-Morphic'
> would work well, and would be consistent with existing package names
> such as 'ToolBuilder-Morphic'.
>
> If this makes sense, then can we open a new package 'Objectland-Morphic'
> and start moving these recent changes into that new package? The goal
> would be to be able to remove 'Objectland-Morphic' completely from
> any image, than load it again from the trunk repository with everything
> still 100% working.
>
> Dave
>

>




—
Have a good one; keep it, light.
Kindly, langohr . .. … ‘…^,^ 🐇🐇🐇

Sent from Callisto House Mobile - Europa
:: decentralized mobile homeless solutions ::
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.squeakfoundation.org/pipermail/squeak-dev/attachments/20221021/e361f385/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: pastedImage.png
Type: image/png
Size: 60743 bytes
Desc: pastedImage.png
URL: <http://lists.squeakfoundation.org/pipermail/squeak-dev/attachments/20221021/e361f385/attachment-0003.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: pastedImage.png
Type: image/png
Size: 20377 bytes
Desc: pastedImage.png
URL: <http://lists.squeakfoundation.org/pipermail/squeak-dev/attachments/20221021/e361f385/attachment-0004.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: pastedImage.png
Type: image/png
Size: 13801 bytes
Desc: pastedImage.png
URL: <http://lists.squeakfoundation.org/pipermail/squeak-dev/attachments/20221021/e361f385/attachment-0005.png>