[squeak-dev] Re: [Question] ToolBuilder in Squeak 4.1 and Pharo 1.0
compatible?
Andreas Raab
andreas.raab at gmx.de
Thu Apr 22 09:10:47 UTC 2010
On 4/22/2010 1:21 AM, Henrik Johansen wrote:
> ToolBuilder is one of the projects I believe deserve to stay fork-agnostic.
>
> Exactly how we should go about coordinating so they stay in synch I'm interested in hearing :)
That's not so difficult. Let's start by just merging the code bases, and
see where this leads. I expect few, if any difficulties here. Let's add
some test to document the changes (like the one Torsten was bitten by)
and we should be pretty much covered.
Further out, I don't expect us to stay closely in sync. The whole issue
of upstream vs. downstream packages is an open one, so my opinion is
that we should merge when it makes sense (i.e., new features being
added) but otherwise let each project have its own choices.
> With ToolBuilder, I also include the protocol of UIManagers.
> In that regard, one of the changes I think would be hard to get "Pharoers" to back down from, is the change of nil instead of '' return for request: dialog cancels.
What can I say ... I actually said in that discussion that the right
choice is to introduce a different protocol, i.e., request:onCancel: or
something similar. Now we'll end up with a new protocol anyways - I
don't expect Pharo to change (too much pride involved) and I don't
expect Squeak to change (too much legacy code involved) so a new
protocol for the people who care about compatibility going forward is
really the only option (which can also be back-ported if necessary). The
end result will be that we'll declare the return value of #request: to
be "undefined" if the dialog is canceled and recommend using the new
protocol.
It's a great lesson about how not to break an existing protocol. If
Pharo had introduced a new protocol we could have just implemented it in
Squeak and be done. Instead, the end result will be that a new protocol
is introduced anyway and that nobody will be able to reliably use the
the old protocol. Consider it a lesson for the next time an issue like
this comes up - if you're interested in compatibility breaking an
existing protocol isn't the smart choice, in particular when it comes to
cross-platform / cross-dialect protocols.
Cheers,
- Andreas
More information about the Squeak-dev
mailing list
|