Fwd: [ENH] Associations Please review and read the tests
stéphane ducasse
ducasse at iam.unibe.ch
Mon Feb 14 10:15:32 UTC 2005
> Let me give you my side of it.
>
> The reason I started mucking with the association code is that it is
> BROKEN. How? Run the tests on the existing image and you will see.
> There are many missing methods in the various "special" association
> classes, because they do not inherit from Association, but from
> LookUpKey.
>
> I didn't invent ReadOnlyVariable binding. But it is there, and it us
> used. Unfortunately, because the inheritance hierarchy is presently
> wrong, ReadOnlyvariableBindings appear in the debugger like
> LookUpKeys, not like associations. I lost many hours that way.
>
> So I fixed it. But I don't like to patch, I like to do it right. So
> I wrote tests, made the existing code fail, and then wrote new
> classes, which made the tests pass. I thought that this was the
> "good" way to do it.
>
> Then I had to make a changeset so that this stuff would get out into
> the image. This was a big job, because of all of the special case
> code that was trying to stop someone from fixing these bugs. Building
> the change set took weeks. The system code is simple and robust, but
> the change set is not. It was built with bit tweezers (a text editor)
> because Squeak has no facilities for ordering changes or committing
> changes atomically.
>
> Then I hoped that it would get into the image. I have had
> conversations with several different harvesters about the whys and the
> wherefores of these changes. But they never get into the image. We
> are now two versions away from the Squeak where I made this changeset
> work, and so I'm nor surprised that it doesn't work for you.
>
> My suggestion from this are:
>
> (1) If we need special case code to stop things from being recompiled,
> there should be a way to turn it off.
>
> (2) I was told that the right way to document the need for changes is
> to create a failing test set and submit it. I did this in this case,
> but it has apparently been lost. So, we should decide what IS the
> right way to document the need for a change.
>
> (3) There should be an accelerated path for getting "delicate" change
> sets like this into the image.
>
> Andrew
>
> --
>
> Prof. Andrew P. Black Home: +1 503 629 5495
> Department of Computer Science, Cell: +1 503 803 1669
> Portland State University Office: +1 503 725 2411
> Portland, Oregon, USA http://www.cs.pdx.edu/~black
>
More information about the Squeak-dev
mailing list
|