On Feb 6, 2009, at 11:08 AM, Christiaan Hofman wrote:


On 6 Feb 2009, at 7:40 PM, Maxwell, Adam R wrote:

On 02/06/09 07:18, "Christiaan Hofman" <[email protected]> wrote:

It would be quite a bit of work, but I
think it's doable. And it can be done in 3 steps. I already think
I've
a pretty good view of what should be done to remove OmniAppKit,
including a replacement for the preferences.

I have a few suggestions:

1) create a branch (duh)

Merging in using SVN is a bit of a mess. But sure, especially as it
would probably take a while.

Maybe it would be easier to create a branch for releases and rewrite on the trunk. As long as they're segregated so releases can progress in case of major problems...

What's release status anyway? Goldstein was asking about a release a month or so ago, I thought...


2) drop 10.4 support at the same time
3) avoid incorporating bindings at the same time

Dropping 10.4 lets you use the kit's source list table/outline views
instead
of the OAGradientTableView, and allows cleanup to proceed using for/
in and
properties if desired.  Dumping the Omni frameworks will also likely
introduce a bunch of regressions that are 10.4-specific, and
debugging those
will be a PITA.

I think that's a separate issue. I don't really see sufficient reasons
yet to drop 10.4 support.

I mainly think of debugging. BD can no longer be compiled on 10.4, and who still has a 10.4 system and is willing and able to run gdb? Most of the current code was developed on 10.4 or at least tested fairly extensively on it. It's my belief that the major rewrite involved in removing the Omni frameworks will create various bugs that'll be a bitch to track down, especially since some of them will be specific to 10.4.

Using bindings on a project that multiple people are developing is a
big
problem, IMO, because it's hard to see what's changed in a diff or
figure
out what the binding paths are.  This kept me from working on the
BD2 test,
and also Skim, to some extent, as it became impossible to keep up with
changes and reverse-engineer everything in IB to see how it worked.

--
adam


On the other hand bindings can really clean things up. I like it much
more than the annoying glue code. I would rather move things to
bindings. But I understand your point.

I know there are benefits to bindings (performance, less code), but I'm not convinced that the savings in glue code is worthwhile, since you can at least step through code in the debugger. IMNSHO, bindings don't clean things up; they just hide the cruft so it's harder to find.

As a side note, I ran into a fun problem with TeX Live Utility, using bindings to validate a toolbar item. I bound it to an NSOperationQueue's operatio...@count path, and started seeing weird crashes. Sprinkling NSParameterAsserts around revealed that NSOperationQueue delivers its KVO notifications on arbitrary threads, so its KVO compliance is useless for binding. Implementing validateUserInterfaceItem: was more code, but much saner :).

--
Adam

Attachment: smime.p7s
Description: S/MIME cryptographic signature

------------------------------------------------------------------------------
Create and Deploy Rich Internet Apps outside the browser with Adobe(R)AIR(TM)
software. With Adobe AIR, Ajax developers can use existing skills and code to
build responsive, highly engaging applications that combine the power of local
resources and data with the reach of the web. Download the Adobe AIR SDK and
Ajax docs to start building applications today-http://p.sf.net/sfu/adobe-com
_______________________________________________
Bibdesk-develop mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/bibdesk-develop

Reply via email to