• bearophile <bearophileHUGS lycos.com> May 15 2009
• Georg Wrede <georg.wrede iki.fi> May 15 2009
• bearophile <bearophileHUGS lycos.com> May 15 2009
• Christopher Wright <dhasenan gmail.com> May 16 2009
• Andrei Alexandrescu <SeeWebsiteForEmail erdani.org> May 16 2009
• Georg Wrede <georg.wrede iki.fi> May 16 2009
• Andrei Alexandrescu <SeeWebsiteForEmail erdani.org> May 15 2009

bearophile <bearophileHUGS lycos.com> writes:

Georg Wrede:

 arra[i] = arrb[i++];
 arra[i++] = arrb[i];
 I'm not sure that such dependences are good code.
 By stating a definite order between lvalue and rvalue, you
 would actually encourage this kind of code.


I agree that putting such things in code is bad, but nowadays the solution
adopted by C is Just Wrong. To help avoid bugs from your programs the language
has to avoid undefined behaviors (this is one of the lessons of Java).
So the solutions are to define such behaviors as Andrei suggests, or to forbid
such code. I can accept both ways.
I am used to Python and I think I can appreciate the Python solution better,
but the Java solution too is better than having undefined behavior in a modern
language.
So thank you Andrei for this work.

Bye,
bearophile

Georg Wrede <georg.wrede iki.fi> writes:

bearophile wrote:
 Georg Wrede:
 
 arra[i] = arrb[i++];
 arra[i++] = arrb[i];
 I'm not sure that such dependences are good code.
 By stating a definite order between lvalue and rvalue, you
 would actually encourage this kind of code.


 I agree that putting such things in code is bad, but nowadays the
 solution adopted by C is Just Wrong. To help avoid bugs from your
 programs the language has to avoid undefined behaviors (this is one of
 the lessons of Java).


If Walter had ulimited time, the of course I'd favor either defining the 
precedence, or then enforcing errors on them, and not leaving them 
undefined. Or, probably, with Walter's unlimited time, I'd vote for a 
system that is the Theoretically Best mix of programming praxis and 
compiler enforcement.

 So the solutions are to define such behaviors as Andrei suggests, or
 to forbid such code. I can accept both ways.

 I am used to Python and I think I can appreciate the Python solution
 better, but the Java solution too is better than having undefined
 behavior in a modern language.

bearophile <bearophileHUGS lycos.com> writes:

Georg Wrede:
 If Walter had ulimited time, the of course I'd favor either defining the 
 precedence, or then enforcing errors on them, and not leaving them 
 undefined. Or, probably, with Walter's unlimited time, I'd vote for a 
 system that is the Theoretically Best mix of programming praxis and 
 compiler enforcement.


Java is currently the most used language because it gives a big help in
avoiding bugs or fixing them. And today Java is not a slow language. Java shows
why having undefined things is bad for productivity and doesn't give much
performance or portability.

Bye,
bearophile

Christopher Wright <dhasenan gmail.com> writes:

Georg Wrede wrote:
 bearophile wrote:
 Georg Wrede:

 arra[i] = arrb[i++];
 arra[i++] = arrb[i];
 I'm not sure that such dependences are good code.
 By stating a definite order between lvalue and rvalue, you
 would actually encourage this kind of code.


 I agree that putting such things in code is bad, but nowadays the
 solution adopted by C is Just Wrong. To help avoid bugs from your
 programs the language has to avoid undefined behaviors (this is one of
 the lessons of Java).


 If Walter had ulimited time, the of course I'd favor either defining the 
 precedence, or then enforcing errors on them, and not leaving them 
 undefined. 


$ su walter
$ ulimit -t unlimited

Andrei Alexandrescu <SeeWebsiteForEmail erdani.org> writes:

Christopher Wright wrote:
 Georg Wrede wrote:
 bearophile wrote:
 Georg Wrede:

 arra[i] = arrb[i++];
 arra[i++] = arrb[i];
 I'm not sure that such dependences are good code.
 By stating a definite order between lvalue and rvalue, you
 would actually encourage this kind of code.


 I agree that putting such things in code is bad, but nowadays the
 solution adopted by C is Just Wrong. To help avoid bugs from your
 programs the language has to avoid undefined behaviors (this is one of
 the lessons of Java).


 If Walter had ulimited time, the of course I'd favor either defining 
 the precedence, or then enforcing errors on them, and not leaving them 
 undefined. 


 $ su walter
 $ ulimit -t unlimited


I swear I wanted to post this too! :o)

Andrei

Georg Wrede <georg.wrede iki.fi> writes:

Andrei Alexandrescu wrote:
 Christopher Wright wrote:
 Georg Wrede wrote:
 bearophile wrote:
 Georg Wrede:

 arra[i] = arrb[i++];
 arra[i++] = arrb[i];
 I'm not sure that such dependences are good code.
 By stating a definite order between lvalue and rvalue, you
 would actually encourage this kind of code.


 I agree that putting such things in code is bad, but nowadays the
 solution adopted by C is Just Wrong. To help avoid bugs from your
 programs the language has to avoid undefined behaviors (this is one of
 the lessons of Java).


 If Walter had ulimited time, the of course I'd favor either defining 
 the precedence, or then enforcing errors on them, and not leaving 
 them undefined. 


 $ su walter
 $ ulimit -t unlimited


 I swear I wanted to post this too! :o)


Nice quips, from you both...    :-)

But there's a bug in the algorithm. A more correct version would be:

$ su
$ ulimit -t unlimited
$ su walter
$

Andrei Alexandrescu <SeeWebsiteForEmail erdani.org> writes:

bearophile wrote:
 Georg Wrede:
 
 arra[i] = arrb[i++];
 arra[i++] = arrb[i];
 I'm not sure that such dependences are good code.
 By stating a definite order between lvalue and rvalue, you
 would actually encourage this kind of code.


 I agree that putting such things in code is bad, but nowadays the solution
adopted by C is Just Wrong. To help avoid bugs from your programs the language
has to avoid undefined behaviors (this is one of the lessons of Java).
 So the solutions are to define such behaviors as Andrei suggests, or to forbid
such code. I can accept both ways.
 I am used to Python and I think I can appreciate the Python solution better,
but the Java solution too is better than having undefined behavior in a modern
language.
 So thank you Andrei for this work.


Well put! Java has definitely taught us that a clean, robust language 
definition can go a long way. (And also that marketing helps :o)).

I think Java is a bit like the STL in a certain respect: there's a 
feeling of "before" and "after". Java has shown not only that defined 
behavior is good, but also that it can be incorporated into a practical 
language.

Andrei