James Dennett wrote:
> bearophile wrote:
>> Bill Baxter:
>>> I'm no C# guy but that looks like a nice way to fix C's switch.
>> I think the C switch is broken in 2 or more ways, so I'd like to see D improve it. This is some code I have shown time ago:
>>
>> import std.stdio: writefln;
>> void main(string[] args) {
>>   switch (args[1]) {
>>     int x = 1; // NOT initialized?
>>     case "1": writefln("1! x=", x); break;
>>     case "2": writefln("2! x=", x); break;
>>   }
>> }
> 
> C++ fixed this some time in the last century, making the code
> above ill-formed (diagnostic required) as it skips the initialization
> of x.
> 
> (It's still UB in C99 so far as I know.)

That's actually in the spec?  Cool.  I thought it was just a QOI thing.

--bb

Bruce Adams Wrote:

> On Fri, 04 Jan 2008 00:15:25 -0000, S. <S@s.com> wrote:
> 
> > Carlos Santander Wrote:
> >>
> >> How about using existing D syntax?
> >>
> >> switch (s[i]) {
> >>       case ' ', '\t', '\f', '\r', '\n', '\v':
> >>           if (inword) {
> >>               r ~= capitalize(s[istart .. i]);
> >>               inword = false;
> >>           }
> >>           break;
> >>
> >>       default:
> >>           if (!inword) {
> >>               if (r.length)
> >>               r ~= ' ';
> >>               istart = i;
> >>               inword = true;
> >>           }
> >>           break;
> >> }
> >
> > This whole conversation about switch was kind of lost on me, but I have to contribute this.
> >
> > I, at various times, have written code that depends on case statements falling through, while not being identical!
> >
> > For example:
> >
> > switch(foo)
> > {
> > case 'bob':
> >     //Do bob stuff
> >     //Fall through and doo bar stuff too.
> > case 'bar':
> >     //Do bar stuff and exit
> >     break;
> >
> > case 'baz':
> >    //Do some baz stuff
> >    break;
> > }
> >
> > Whatever is changed shouldn't break this.
> 
> This whole conversation boggles me. cases should *never* fall through its dangerous.

In other news, it has been discovered that the risk of choking is greatly increased by eating anything that is not pureed.  As a result, lawmakers have outlawed solid food.

Programming requires a brain.

You should go read DailyWTF for awhile.  It is possible to do the most ridiculous things no matter how carefully the language is crafted.  We should just stay away from hand-holding and focus on making things easier.

I agree with measures that make debugging easier when dealing with naming and hijacking.  However, a simple step-through will reveal a case falling through when it shouldn't.  The aforementioned issues are not as easily caught.

-S.

bearophile Wrote:

> S. Wrote:
> > I, at various times, have written code that depends on case statements falling through, while not being identical!
> 
> If you take a look at my original post, I have suggested to use a "fall" or "continue" statement when you want to go down, and nothing if you want the default of going at the end of the switch...
> 
> Bye,
> bearophile

The same problems you initially mentioned still exist in your proposal.  You just moved them around:

caseof(Foo)
{
  case 'foo':
     writeflin("bar");
     fall;
     x = 3;
  case 'bar':
    writefln("OMGWTF X != 3");
}

Whatever complicated syntax you invent there will be unreachable code.  The compiler should produce ERRORS on any unreachable code.

The error you linked to is not switch-induced, but stupidity induced.  However, I will say this:  D is starting to become so complex that it is becoming difficult to understand the entire specification.   Once it is no longer feasible to have the spec memorized then it becomes easy to make very silly mistakes.

-S.

On Mon, 07 Jan 2008 22:51:31 -0000, S. <S@s.com> wrote:

> Bruce Adams Wrote:
>
>> On Fri, 04 Jan 2008 00:15:25 -0000, S. <S@s.com> wrote:
>>
>> > Carlos Santander Wrote:
>> >>
>> >> How about using existing D syntax?
>> >>
>> >> switch (s[i]) {
>> >>       case ' ', '\t', '\f', '\r', '\n', '\v':
>> >>           if (inword) {
>> >>               r ~= capitalize(s[istart .. i]);
>> >>               inword = false;
>> >>           }
>> >>           break;
>> >>
>> >>       default:
>> >>           if (!inword) {
>> >>               if (r.length)
>> >>               r ~= ' ';
>> >>               istart = i;
>> >>               inword = true;
>> >>           }
>> >>           break;
>> >> }
>> >
>> > This whole conversation about switch was kind of lost on me, but I  
>> have
>> > to contribute this.
>> >
>> > I, at various times, have written code that depends on case statements
>> > falling through, while not being identical!
>> >
>> > For example:
>> >
>> > switch(foo)
>> > {
>> > case 'bob':
>> >     //Do bob stuff
>> >     //Fall through and doo bar stuff too.
>> > case 'bar':
>> >     //Do bar stuff and exit
>> >     break;
>> >
>> > case 'baz':
>> >    //Do some baz stuff
>> >    break;
>> > }
>> >
>> > Whatever is changed shouldn't break this.
>>
>> This whole conversation boggles me. cases should *never* fall through its
>> dangerous.
>
> In other news, it has been discovered that the risk of choking is greatly increased by eating anything that is not pureed.  As a result, lawmakers have outlawed solid food.
>
Is that an example of Ignoratio elenchi? (http://en.wikipedia.org/wiki/Ignoratio_elenchi)
Lawmakers never actually outlawed goto but you don't see it used very often.
That's because putting glass in your food is not a good idea.

> Programming requires a brain.
>
Working in a team requires respecting other people's Brains and level of development.
Fall through unnecessarily complicates things without adding much if indeed any power.
I'd rather use my brain on the important things.

> You should go read DailyWTF for awhile.  It is possible to do the most ridiculous things no matter how carefully the language is crafted.  We should just stay away from hand-holding and focus on making things easier.
>
Its a matter of perspective. Is it easier to debug or easier to read or easier
to write safe code?

> I agree with measures that make debugging easier when dealing with naming and hijacking.  However, a simple step-through will reveal a case falling through when it shouldn't.  The aforementioned issues are not as easily caught.
>
True but its easier still if you don't need to step through at all. Its easy enough to
find the source and targets of a goto with a find in your code. Okay its nowhere near
in the same league as goto but the point's still valid.

> -S.

B.


-- 
Using Opera's revolutionary e-mail client: http://www.opera.com/mail/

Bruce Adams Wrote:

> On Mon, 07 Jan 2008 22:51:31 -0000, S. <S@s.com> wrote:
> 
> > Bruce Adams Wrote:
> >
> >> On Fri, 04 Jan 2008 00:15:25 -0000, S. <S@s.com> wrote:
> >>
> >> > Carlos Santander Wrote:
> >> >>
> >> >> How about using existing D syntax?
> >> >>
> >> >> switch (s[i]) {
> >> >>       case ' ', '\t', '\f', '\r', '\n', '\v':
> >> >>           if (inword) {
> >> >>               r ~= capitalize(s[istart .. i]);
> >> >>               inword = false;
> >> >>           }
> >> >>           break;
> >> >>
> >> >>       default:
> >> >>           if (!inword) {
> >> >>               if (r.length)
> >> >>               r ~= ' ';
> >> >>               istart = i;
> >> >>               inword = true;
> >> >>           }
> >> >>           break;
> >> >> }
> >> >
> >> > This whole conversation about switch was kind of lost on me, but I
> >> have
> >> > to contribute this.
> >> >
> >> > I, at various times, have written code that depends on case statements falling through, while not being identical!
> >> >
> >> > For example:
> >> >
> >> > switch(foo)
> >> > {
> >> > case 'bob':
> >> >     //Do bob stuff
> >> >     //Fall through and doo bar stuff too.
> >> > case 'bar':
> >> >     //Do bar stuff and exit
> >> >     break;
> >> >
> >> > case 'baz':
> >> >    //Do some baz stuff
> >> >    break;
> >> > }
> >> >
> >> > Whatever is changed shouldn't break this.
> >>
> >> This whole conversation boggles me. cases should *never* fall through
> >> its
> >> dangerous.
> >
> > In other news, it has been discovered that the risk of choking is greatly increased by eating anything that is not pureed.  As a result, lawmakers have outlawed solid food.
> >
> Is that an example of Ignoratio elenchi?
> (http://en.wikipedia.org/wiki/Ignoratio_elenchi)
> Lawmakers never actually outlawed goto but you don't see it used very
> often.
> That's because putting glass in your food is not a good idea.

Well one probably not.  For one, there isn't a formal definition given for that term on your cited page.

Secondly, my point is that most things that are generic enough to be useful, are inherently dangerous if misused.    Consider most kitchen utensils.  The ones that are "safe" only do one thing, and if you want a "safe" kitchen then you need hundreds of these gadgets.

goto, and break, are defined in such general terms that they are useful over a wide range of constructs.  If you want to make things "safe" you need to pollute your "kitchen" with gadgets.

I would like to see the specification stay simple enough that know what exactly each "tool" does in a general sense and then apply it in a specific way.  Now sometimes, if something is used enough, then it makes sense to have a special tool for it.  (Like having a coffee pot instead of just boiling grounds and straining them out later.)  This is why I like D instead of just straight up C.


> > Programming requires a brain.
> >
> Working in a team requires respecting other people's Brains and level of
> development.
> Fall through unnecessarily complicates things without adding much if
> indeed any power.
> I'd rather use my brain on the important things.
> 

And calling into another function in order to duplicate functionality requires introduce a new scope....

If the specification is easily understood then there isn't much problem here...

> > You should go read DailyWTF for awhile.  It is possible to do the most ridiculous things no matter how carefully the language is crafted.  We should just stay away from hand-holding and focus on making things easier.
> >
> Its a matter of perspective. Is it easier to debug or easier to read or
> easier
> to write safe code?

It should be easier to write safe code, and easier to debug and easier to read.   Sometimes though, the compiler should just warn you when you're using a construct incorrectly.   For example, a return statement can cause unreachable code.  Should we get rid of the return statement in favor of 15 different keywords to do similar (but specific) things, or make it an error when it produces unreachable code?

I think my take on how I run my kitchen is applicable to D.

> 
> > I agree with measures that make debugging easier when dealing with naming and hijacking.  However, a simple step-through will reveal a case falling through when it shouldn't.  The aforementioned issues are not as easily caught.
> >
> True but its easier still if you don't need to step through at all. Its
> easy enough to
> find the source and targets of a goto with a find in your code. Okay its
> nowhere near
> in the same league as goto but the point's still valid.

By aforementioned issues I meant hijacking and protection levels.  Which are basically the same reason we require our variables to be defined instead of just implying them when they're used.

Is the trade off of having to declare your variables worth not having to figure out what is going on when you mistype a variable name?   Yes, most certainly.     I don't think switch, goto, break, and the like are anything like that.


-S.

S. wrote:

> However, a simple step-through will reveal a case falling through when it shouldn't.

Eliminating the possibility of unintentional fall-through altogether will eliminate the need for even your "simple step-through".

And simple step through may not be so simple when the program is hundreds of thousands of lines long.  Mistaken fall-through in some switch in some function in some module can change a state, and that state can later trigger another different change of state and so on, until finally somewhere your program crashes because of an incorrect state a dozen causal steps away from the original problem.

So "simple step through" may mean "simply" stepping through thousands of lines of code to try to figure out where things first went wrong.  It can be very non-trivial.

Implicit switch fall-through is evil.

--bb

S:

>The compiler should produce ERRORS on any unreachable code.<

I agree (D too has a simple way to spot dead code under a return in not-release mode).
Sometimes Pascal syntax is better than the C one, the switch is one of them, I like the Pascal version better.
In the Python community people collect use cases, do a simple frequency count of them, and then usually look for a simple solution able to cover most of them. So we can collect some use cases of the switch. I presume most use cases are covered by a Pascal-like syntax. The other situations are probably covered by putting commas between alternative cases. What other use cases do you people have?


>The error you linked to is not switch-induced, but stupidity induced.<

Then the language (and language designers) must be twice intelligent to avoid errors done by "stupid" humans. Technology must adapt itself to the limits of the human brain, otherwise it's far more stupid than the humans. In this situation I think such adaptation doesn't require a more complex syntax or the usage or more CPU or more memory. A change has the disadvantage that it makes D syntax looking a bit less like C, and this may be a disadvantage.


>However, I will say this:  D is starting to become so complex that it is becoming difficult to understand the entire specification.   Once it is no longer feasible to have the spec memorized then it becomes easy to make very silly mistakes.<

I too like simpler languages better, that's why I like Python and D (D is simpler than C++ still) :-) D has many parts, so it's more complex than a language with less parts, but usually each D part has few interactions with all the other parts, so the actual complexity isn't too much high :-)

Bye,
bearophile

I just noticed this in the specs.
"The third form, goto case;, transfers to the next CaseStatement of the
innermost enclosing SwitchStatement."