Re: RFC: GDL2 differences to EOF / WO4.5.1

[David Ayers]

Manuel Guesdon wrote:

> Hi David,
>
> On Wed, 12 Mar 2003 18:23:18 +0100 David Ayers <address@hidden> wrote:
>
> >| Hello Manuel,
> >| 
> >| I'm coming up with oddities of EOF 3.0 of WO4.5.1 and subtle differences 
> >| to GDL2 here and there:
> >| - EONull compareAscending: nil returns NSOrderedAscending instead of 
> >| NSOrderedSame
>
> I have no specific preference. 
>  
I think GDL2 behaviour is correct so I would like keep it.

> >| - EOModel does not implement isEqual:
>
> And WO4.5 does ? So we'll  to implement it...
No it doesn't, but I think GDL2 should anyway.  (It's rather trivial, 
but it changes behaviour.)

> >| And I haven't gotten far enough into .gswd/.wod-parsing to analyse the 
> >| GDL2 feature of NSDictionary valueForKeyPath and quoted key names.  I 
> >| would like to propose the following default: 
> >| GSUseStrictWO451Compatiblity or GSUseStrictEOF451Compatiblity or 
> >| GSUseStrictEOF30?Compatiblity or what ever better name someone comes up 
> >| with.
>
> I don't like too much this kind of things because if you use 2 frameworks or 
> libraries which set a different behaviour, the result is unknown and may 
> depend on some calling order.
> I still don't see real problem with quoted key names but if you want to make have 
> a parameter to control this, I'm OK :-)
>  
Well the point is that as soon as I continue to fine tune the tests, 
more and more of these issues will arise, and I wanted to put a general 
mechanism in place, which I could consistently use.  But I do recognize 
your concerns...

> For this one, we could act as WO4.5 anyway.
We could, I guess... (eventhough I still think current GDL2 behavior is 
better.)

> >| - have [EOModel isEqual:] call super instead of a sane isEqual: 
> >| implementation, that I would like to have a look at later.
>
> For this one, WO4.5 implementation should be OK, shouldn't it ?
>
>  
Actually it was during writing testcases that I noticed that it wasn't 
implemented.  And it's not implemented in WO4.5 either.  I just think we 
should.

Testing OS 4.2 vs WO 4.5 show some substantual differences in KVC of 
NSDictionary when was looking into GDL2 quoted key feature.  I know that 
many of these are edge cases and subtle differences that cause no 
immediate harm in 98% of the usages.  But in my experience, it's these 
2% edge cases which makeup an unproportional amount of the debugging 
effort, especially when porting.  This was my original concern.

On the other hand it is doubtful that all edge cases can be identitfied, 
no matter how estensive the tests are.  And I may end up adding a lot of 
extensive edge case handling and still miss the ones that are causing 
problems.  I also mean to keep and add comprehensive features, that are 
not part of EOF, and these might also trigger such problems.

In retrospect, I'm thinking of holding the default back for now and 
leaving GDL2 behaviour as it is in these edge conditions and comment 
code and test cases.  I'll sleep over it again.  Comments appreciated.

Cheers,
Dave