The text certainly helped, especially in understanding the premise of
the algorithm. I admit though that I did skim some of the later parts
of the documentation, diving into the code instead. I ultimately just
let the code and algorithm sink in by staring at and working with the
code for an extended period of time.
I was hoping to resolve this issue soon, but it continues to haunt me.
Right now I'm playing with:
Coalesce(a.B.C, d.E.F)
In the current code you'd end up with essentially (a.B.C, N) and
(d.E.F, N), but the Coalesce of those two might be a non-null value
since when we emptily outer join a.B.C, we don't know what d.E.F will
be. So Coalesce(N,N) may result in a true expression. I'm
considering introducing a "V" to represent a non-null non-boolean
value. The result of that Coalesce would be (a.B.C, NV) and (d.E.F,
NV).
Certainly not everything has crystalized in my brain yet.
Patrick Earl
On Fri, Apr 22, 2011 at 1:33 PM, Harald Mueller <[email protected]> wrote:
> Hi Patrick -
>
>> Tonight I had some clarity on how the mappings worked. This allowed
>> me to make some simplifications and fixes. Would be happy to hear
>> what you think.
>
> Interesting. What I think, in short (before taking that time to write more
> tests):
>
> a. The new code is IMO correct, and it is more concise, and it behaves
> "well-behaved" in more cases (I think ... tests ...) - so this makes it
> better!
>
> b. However, "a wrong comment is worse than no comment": Please correct (no,
> don't delete!) the comments in the code I originally added that are now wrong
> - because ...
>
> c. ... what you did is, IMO, a fundamental change in the semantic. This shows
> up in my "design text" (which is sort of a long comment, isn't it?), which is
> now misleading due to a fundamental assumption that runs through the whole
> text, which now does not hold anymore.
> I'd like you to change that comment and point out that modification, because
> then I know that we have the same view on that semantical issue - and I'll
> rewrite the text (and add explanations for the features you added in those
> two modifications).
>
> That ok with you?
>
> BTW: I could not have survived that code without having written that text -
> including examples. Still, would it be better to split the text into snippets
> that go into the source code? ("programmers don't read anything except the
> sourcecode" etc.etc.) Did the text help you?
>
> Regards
> Harald
>
> --
> NEU: FreePhone - kostenlos mobil telefonieren und surfen!
> Jetzt informieren: http://www.gmx.net/de/go/freephone
>
