code completion

[John Zedlewski]

  I started working on this issue a little while back and stopped for a
variety of reasons (mostly including too much other work).  If nobody
else gets too it, I certainly will this summer, and possibly this Spring
(the current semester is a KILLER for me).  If you want to see how this
feature would work, download the free edition of JBuilder from
www.inprise.com/jbuilder.  Obviously, it's a Java product, but it's the
same idea.  Once you use an IDE with code completion, it's hard to
imagine ever going back.
   Basically, the task requires some pretty smart parsing in order to
work well, but it's quite doable.  There was, at some point, a plan to
put all of the classes in the current project + all of the basic KDE/Qt
classes into a local mini-database, so I stopped working on the
completion until that task got completed (has that since been scrapped
as unnecessary/too difficult?).  WsStudio (wxstudio.linuxbox.com) is the
only free IDE I know of that has this feature, though that project seems
pretty much halted right now, it might have some good tips.
   Conceptually, it's not too hard, but it is a lot more complicated
than shell completion.  To take an example, look at:
"
KMyDialog dlg;
... lots of intervening code ...
int doSomething() {
	(*((someOtherClass*)dlg.pointerToMyClass)->funcReturningPointer(x)).
"
Now, when the user enters that last '.', the code completion has to
parse that whole complicated expression, find out what class (or struct,
union, whatever) someOtherClass::funcReturningPointer() actually
returns, and then come up with a list of all the members of that return
type on the fly.  Of course, you can find much simpler and much more
complex examples.
  From what I've heard and seen, other systems do manage to parse the
current file and (if necessary) its includes on the fly to find the type
of the variable to the left of the '.'. '->' or '::' in question.  It
means that a pretty fast parser is necessary if you want to handle large
source files without an unacceptable delay.  More frustratingly, this is
also different from (but closely related to) what the current class
parser does, because it focuses a lot on scope and finding random
variable declarations.  Jonas obviously knows a lot more about these
issues than I do, though, so I could be totally off base.
  So, basically, if you don't find anyone who can/wants to do it, I can
promise to get to it in a few months (because I would REALLY love this
feature, but otherwise I wish good luck to whoever steps up to do it.    
  Keep up the amazing work, guys!
--JRZ