Re: Replacing GOTO. [was Re: Conceptual problem solved by GOTO]
Chris Bensler wrote:
>
> Peter Robinson wrote:
> >
> > Shawn Pringle wrote:
> > >
> > > Fernando Bauer wrote:
> > > >
> > > > Suppose we have GOTO and this kind of code (which I think is common):
> > > > }}}
<eucode>
> > > > procedure proc1()
> > > > if cond1 then
> > > > .. some code here ..
> > > > if cond2 then
> > > > .. some code here ..
> > > > if cond3 then
> > > > .. some code here ..
> > > > goto OUT
> > > > end if
> > > > .. some code here ..
> > > > end if
> > > > .. some code here ..
> > > > end if
> > > > OUT:
> > > > .. some code here ..
> > > > end procedure
> > > > </eucode>
{{{
> > > >
> > > > In Eu 3.1, a possible implementation would be:
> > > > }}}
<eucode>
> > > > procedure proc1()
> > > > for i=1 to 1 do
> > > > if cond1 then
> > > > .. some code here ..
> > > > if cond2 then
> > > > .. some code here ..
> > > > if cond3 then
> > > > .. some code here ..
> > > > exit
> > > > end if
> > > > .. some code here ..
> > > > end if
> > > > .. some code here ..
> > > > end if
> > > > end for
> > > > .. some code here ..
> > > > end procedure
> > > > </eucode>
{{{
> > > >
> > > > Now, without GOTO, we are forced to (mis)use a loop construct (for-loop,
> while-loop)</font></i>
> > > > where there is NO loop.
> > > > Conceptually, IMHO this makes no sense.
> > > > To avoid the conceptual problem we could use flags, but IMHO this is
> > > > even
> worst.</font></i>
> > > >
> > > > Regards,
> > > > Fernando
> > >
> > > All non-goto code has been about making code flow as through a tree. Up a
> > > trunk,
> > > choosing major branches, to smaller branches (routines) and going into
> > > twigs
> > >
> > > (loops, blocks) and then returning through the points until finally
> > > getting
> > > to
> > > a leaf and exiting. This loop idiom to replace goto is something I have
> > > forgotten but I did this once. What you see as an argument that goto can
> > > make
> > >
> > > clearer code I see as an argument that there are times we want to
> > > implement
> > > this kind of control flow but we lack something in the language so we
> > > either
> > >
> > > implement with goto or with a loop.
> > >
> > > You have avoided spegetti code but that does look like misuse. Goto is
> > > also
> > > useful for porting from other languages, and there are times
> > > you wish to jump to another neighboring twig but do not want to put the
> > > overhead
> > > of a function call to do it. So, we need goto. Yet, the reason why we
> > > have
> > > loops and multi-line if/blocks is because it usually easier if we divide
> > > the
> > > kinds
> > > of jumps we are doing into kinds. Is it a loop? Is it an if block? Is
> > > it a loop with an integer counter (for)?
> > >
> > > I feel there is a "missing" kind of structure for putting things in
> > > non-looping
> > > block and being able to jump out of it as though it were a loop with exit.
> > >
> > > The
> > > one iteration loop works but it isn't a loop semantically speaking.
> > >
> > > The keywords while, for, and multi-line if blocks exist because people
> > > decided
> > > that they would replace some usages with goto with control structures that
> > > could
> > >
> > > be easier to read. These are all replacements of usages of goto. The
> > > control
> > >
> > > structure for replaces a usage of goto or of while more directly. We all
> > > agree
> > >
> > > these were good changes even if goto still has a place. I think Fernando
> > > has
> > > hit
> > > on a usage of for which could further be put into a different kind of
> > > control
> > >
> > > structure.
> > >
> > > So, I want you to think of a new keyword that the developers
> > > might decide to put into 5.0 but we mustn't keep changing the target of
> > > 4.0
> > >
> > > or nobody will ever be able to finish it
> > > (we will see Starcraft II in stores first, the cows will come home first).
> > >
> > >
> > > Shawn Pringle
> >
> > I've been thinking along the same lines, except that I don' believe that all
> > the patterns can be defined with the precision needed to fix on one, or even
> > two, constructs. One of the problems is that the same concept might appear
> > in
> > many different forms. You might test the "condition" of sequence data with
> > an
> > 'if' block, or loop through its elements, or use a while 1 construct, or
> > just
> > a series of top-level code. The most you can say about the pattern is that:-
> >
> > - A group of constructs execute some testing of data
> > - The data either fails early, in which case you want to fall through, but
> > maybe
> > not right to the end - maybe to an error point,
> > - or it triggers a retry, and you want to go back, but quite likely not to
> > pass
> > through all the same hoops again.
> >
> > It struck me that part of the lack of structure of the goto comes from the
> > single
> > label approach, rather than defining a block of code. If you could label a
> > block
> > of code, it would have more meaning. Just say you chose the keyword 'during'
> > with a label. Code like Fernando's would look more structured:-
> >
> > procedure proc1()
> > -- preliminary code
> > during data_filtering
> > if cond1 then
> > -- code here
> > if cond2 then
> > -- code modifying data
> > retry data_filtering
> > if cond3 then
> > exit data_filtering
> > end if
> > -- code here
> > end if
> > -- code here
> > end if
> > end data_filtering
> > -- clean_up code
> > end procedure
> >
> > This really does nothing more than labelled jumps, but the labelling of the
> > code as a meaningful block makes it more structured.
> >
> > Of course, someone will say you could just wrap the innards in another
> > procedure
> > and use return to exit, and something else for retry, but that doesn't
> > detract
> > from Fernando's and your point that the language lacks a construct to
> > represent
> > a common logical pattern.
> >
> > And like you, I won't be holding my breath for something like this to appear
> > in the language, which is why I'm settling for the current proposal.
> >
> > Cheers
> > Peter Robinson
>
> I think goto still has a place, but I would agree to a block construct such
> as suggested. I would prefer block/end block to during though I think.
> Aside from creating a place to break out from without abusing the intent of
> other constructs, it would provide a way to create nested private scopes.
>
> In my opinion all the various types of code blocks should be able to be
> targettable
> break points, not just loops. A block construct would be a universal solution.
>
> Code is Alchemy
I really appreciate all your comments and ideas, thanks!
Maybe it's still possible to extend the use of labels to 'if-then-else'
construct.
I read somewhere this kind of syntax:
for i = 1 to length(s) label "xyz" do
For now, by analogy and for completeness, maybe we could have this:
if ..cond.. label "xyz" then
Regards,
Fernando
|
Not Categorized, Please Help
|
|
