Update on version 5.0?

OA,

Would it be possible to get a short status report on Dolphin 5.0 now
that it is March?  Don't let it keep you from the important work at
hand, though.

I wonder if one change will make it into this version: a compiler or
parser implementation in Smalltalk instead of the one hidden away in the
DLL.  There was talk of this once.  Since no one posted a Dolphin
solution to :+=, I assume that it isn't possible in Dolphin, possibly
due to lack of access to the compiler or parser.

Thanks.

Keith Alcock

No, the D5 compiler is delivered in DolphinCR005.dll and is not suitable
for modification.


Keith Alcock wrote:

Keith Alcock <[hidden email]> wrote in message
news:[hidden email]...
> No, the D5 compiler is delivered in DolphinCR005.dll and is not suitable
> for modification.
>
> Keith Alcock wrote:
...
> >
> > I wonder if one change will make it into this version: a compiler or
> > parser implementation in Smalltalk instead of the one hidden away in the
> > DLL.  There was talk of this once.  Since no one posted a Dolphin
> > solution to :+=, I assume that it isn't possible in Dolphin, possibly
> > due to lack of access to the compiler or parser.

While we don't have compiler source code in Dolphin Smalltalk (yet) I
believe there is a parser included as part of the refactoring browser.
There is reformatting code in D5, perhaps it could be extended to replace
:+= with the desired code expansion.  If that was not a suitable option
perhaps one could hook into the compiler methods of CompilerLibrary to do
the code swap there.  I think this could be done by renaming the existing
methods, then writing "wrapper" methods to call the original methods with
the modified source.  I think what you want to do could be done, but I am
not sure extending the syntax is a good idea for most people.

Chris

Christopher, etc.,

One of the problems with having multiple Smalltalk parsers in the image is that they
are likely to (and do!) parse different languages.  There is a message somewhere in
the newsgroup about a message being sent to the transcript when the refactoring
browser parser cannot parse valid Dolphin code.  In addition it is duplication of
effort, both in writing and maintaining two versions.

Last time I checked, the Dolphin compiler insists that binary operators have at most
two characters.  Other Smalltalks are more lenient and might allow :+= to be defined
as an operator.  So far my best solution is to define a numeric value holder class
and then provide methods for += and -=.  Normally I wouldn't do that, but it was a
case of

veryDescriptiveVariableName := veryDescriptiveVariableName + first part of lengthy
equation.
several more calculations.
veryDescriptiveVariableName := veryDescriptiveVariableName + second part of lengthy
equation.
more calculations.
etc.

and I thought that += made it obvious that the value wasn't being reset, but being
used to accumulate partial results.  The statements also fit on single lines that
way.

Anyway, it worked for me.

Keith Alcock



Christopher J. Demers wrote:

"Keith Alcock" <[hidden email]> wrote in message
news:[hidden email]...
> ...
> Last time I checked, the Dolphin compiler insists that binary operators
have at most
> two characters.  Other Smalltalks are more lenient ...

As far as I remember it is the other way around: Dolphin's compiler has
always (?) allowed a binary selector to have more than 2 characters, whereas
this is not true in some other Smalltalks, and in fact the RB-based parser
does not like them either. I can't quite remember what the ANSI standard
says on this issue, and I don't have my copy to hand.

>...and might allow :+= to be defined
> as an operator...

The colon character is not really legal in a binary selector (though the
Dolphin Compiler allows it as the first character so it would allow :+=),
there being a clash with its use as the separator in keyword selectors.

Regards

Blair

Blair,

Not that it matters much, but if I try to define some binary methods like :+= or
+=& with D5, then messages like these show up on the transcript:

Error 32 on line 1 of MyTest>>:+= -> 'expecting message pattern'
Error 41 on line 1 of MyTest>>+=& -> 'variable name expected'

The methods do seem to compile nevertheless--and even run!  Maybe the messages come
from some other parser (one for a different language) instead of from Dolphin.

Keith


Blair McGlashan wrote:

Keith Alcock <[hidden email]> wrote in message news:<[hidden email]>...
> Not that it matters much, but if I try to define some binary methods like :+= or
> +=& with D5, then messages like these show up on the transcript:
>
> Error 32 on line 1 of MyTest>>:+= -> 'expecting message pattern'
> Error 41 on line 1 of MyTest>>+=& -> 'variable name expected'
>
> The methods do seem to compile nevertheless--and even run!  Maybe the messages come
> from some other parser (one for a different language) instead of from Dolphin.

I believe they're coming from the parser in the Refactoring Browser.  HTH.

Bob Jarvis