Hi Stev, you wrote:
even if the point of wolfgang was valid but from a C perspective.
and I must frankly admit, that I can't parse that sentence. Ascribe this to my mother tongue not being English.
So in case you meant my article is valid only from a C programmer's perspective, let me stress, that my perspective is from a programmer who aims to convince himself that the program at hand is correct. If you want to attribute this frame of mind to a profession, I think a mathematician trying to prove a theorem is more accurate than a C programmer.
Proving a program being correct while writing it is much easier, if the first natural number is zero. This convention leads to more elegant and "natural" statements about the state of instance variables.
This is not confined to indexes, but file stream positions as well. The position is zero after initialization and thus does not address a byte. It needs to be incremented before it is a valid pointer. This made it much harder to write and prove the AltoFileStream class correct.
Unix and C let zero be the first natural number. As you might know, the Unix V6 kernel does not contain a single bug. Ironically, the only V6 program that I consider buggy is the debugger. Maybe this is because Ken Thompson and Dennis Ritchie didn't need the debugger. And maybe this is because they were trying to prove their programs to be correct while writing. And maybe this is the reason they decided to let zero be the first natural number.
The last sentences are purely speculative and Dennis might shed some light on them. But it is not speculative that Smalltalk is the only programming language that allows indexed variables and forces the programmer to use one as the first index. If you find another one let me know.
Greetings, Wolfgang
-- Weniger, aber besser.
Wolfgang Helbig wrote:
The last sentences are purely speculative and Dennis might shed some light on them. But it is not speculative that Smalltalk is the only programming language that allows indexed variables and forces the programmer to use one as the first index. If you find another one let me know.
http://en.wikipedia.org/wiki/Array lists BASIC and Fortran as having 1-based array indexing, and it even discusses some of the relative merits of and arguments for 0-based and 1-based array indexing. Which one is more natural is a matter of personal taste and opinion. In most everyday programming situations I consider the 1-based approach much more natural, but the low-level stuff where you're addressing hardware is definitely an exception which can be handled pretty well by implementing a specialized arrayed class which does 0-based indexing. Having learned BASIC as a first language in school, I found C's convention of 0-based indexing pretty awkward, until I realized that C is mostly a kind of portable assembly language which lets you get close to the hardware. It's not a language I would consider well suited for expression my thoughts in a program... The question is: do we want to let hardware characteristics dominate our way of thinking? I at least don't. When I've got 5 elements in a collection, I want the first at index 1 and the fifth at index 5. When I'm using Integers, I don't want to think about 16 vs 32 bits, signed vs unsigned etc. The argument about loop bounds is somewhat related. I've seen quite some beginner's C code which did "for (i=0; i<=10; i++)" to address a 10-element array :-) Seasoned C programmers don't do that anymore, but it tells me a bit about what is "natural" :-)
Cheers, Hans-Martin
On 5/1/06, Hans-Martin Mosner hmm@heeg.de wrote:
Wolfgang Helbig wrote:
The last sentences are purely speculative and Dennis might shed some
light on
them. But it is not speculative that Smalltalk is the only programming
language
that allows indexed variables and forces the programmer to use one as the
first
index. If you find another one let me know.
http://en.wikipedia.org/wiki/Array lists BASIC and Fortran as having 1-based array indexing, and it even discusses some of the relative merits of and arguments for 0-based and 1-based array indexing. Which one is more natural is a matter of personal taste and opinion. In most everyday programming situations I consider the 1-based approach much more natural, but the low-level stuff where you're addressing hardware is definitely an exception which can be handled pretty well by implementing a specialized arrayed class which does 0-based indexing. Having learned BASIC as a first language in school, I found C's convention of 0-based indexing pretty awkward, until I realized that C is mostly a kind of portable assembly language which lets you get close to the hardware. It's not a language I would consider well suited for expression my thoughts in a program... The question is: do we want to let hardware characteristics dominate our way of thinking? I at least don't. When I've got 5 elements in a collection, I want the first at index 1 and the fifth at index 5. When I'm using Integers, I don't want to think about 16 vs 32 bits, signed vs unsigned etc. The argument about loop bounds is somewhat related. I've seen quite some beginner's C code which did "for (i=0; i<=10; i++)" to address a 10-element array :-) Seasoned C programmers don't do that anymore, but it tells me a bit about what is "natural" :-)
So true. The admonition "remember arrays start at zero" is repeated countless times in most literature. You see this in just about every article or book on most modern languages. If it were so "natural" to count groups of things from 0 people wouldn't need to be reminded constantly.
Regards, Laurence
Cheers,
Hans-Martin
Hi Wolfgang,
Your point of view is correct, but sorry, it is a C point of view. That means a point of view of how can i better insure the low level implementation.
And in fact, i trust you, it is more convenient to have 0-based if your are dealing with pointers and offset.
But the aim of Smalltalk was not to ease such implementation. Smalltalk ignore the notion of pointer, and assume all this low level goes into the VM.
Smalltalk goal is to ease the upper level programming. And at upper level, sorry, if i want the object of rank 100 in a collection, i naturally write (collec at: 100), and would find it very irritating to write (collec at: 99), having to think of how the processor will translate this into an offset.
Looking at the history of numbers, you will discover that the zero is a relatively new invention. Initially, it just meant empty, none or nihil (our Smalltalk nil).
And looking closer at the english words, you know very well that fourth must be associated to number four, fifth to number five, etc... If you associate first to 0, then you associate fourth to 3, something that does not sound natural to me.
That is why we index starting at 1 in Smalltalk, and why we are reluctant to change that (though it is not at all in the syntax, just a convention choice in the library).
Beside, it does not prevent us to use 0-based algorithm like with Stream position, when what we need is an offset. So i do not really see that as an obstacle. It is just a programming habit, and i can tell you i also made a lot of index-errors when switching between FORTRAN and C code, so the lack of errors is not an objective criterium...
Nicolas
Hello Wolfgang,
WH> index. If you find another one let me know.
Matlab! And there's a big user group using it for dsp (digital signal processing) which are deeply frustrated about it.
So if you want to look into this discussion search the comp.dsp archives.
For this discussion:
As a LISPer I'm annoyed by (nth 1 ofMyList) being the second element, as a DSPer I'm annoyed whenever I read some 1-based matlab indexing while zero-based indexing feels so natural in DSP.
As a child of two worlds I bloody my nose with off by one errors more frequent than others but by now I have a rugged nose (TM), so I just laugh and change the indexing.
Cheers
Herbert mailto:herbertkoenig@gmx.net
Wolfgang,
Unix and C let zero be the first natural number. As you might know, the Unix V6 kernel does not contain a single bug. Ironically, the only V6 program that I consider buggy is the debugger. Maybe this is because Ken Thompson and Dennis Ritchie didn't need the debugger. And maybe this is because they were trying to prove their programs to be correct while writing. And maybe this is the reason they decided to let zero be the first natural number.
The people's favorite language Perl has a special variable "$[" that let you specify the index of the first element in an array.
As everyone wrote: it is about either to think about the n-th index , or the offset or distance and I'd say that you can't simply say which is more natural.
Also, in Smalltalk, Streams are 0-based as nicolas and others mentioned.
a _ ReadStream on: 'ABC'. a next. " => $A" a position. "=> 1" a position: 0. a next. "=> $A" a position: 3. a next. "=> nil"
The AltoFileStream should've follow this convension, instead of allowing [1..size] indices.
(The coordinates in a Form are 0-based, by the way...)
Just my $0.02,
-- Yoshiki
Dear Wolfgang,
But it is not speculative that Smalltalk is the only programming language that allows indexed variables and forces the programmer to use one as the first index. If you find another one let me know.
To add to the list, there's Icon.
Icon indexes in an interesting way. It indexes not the elements, but before, after, and between elements. It also indexes from both ends of a list or array. E. g.,
(1,-3) A (2,-2) B (3,-1) C (4,0)
There are three elements, A, B, and C, and 8 indices, -3 to 4. 1 and -3 refer to the point before A, 2 and -2 to the point before B, and 3 and -1 to the point before C. 4 and 0 refer to the end. To access an element you use one of the indices that comes before it.
Best regards,
Bill
But it is not speculative that Smalltalk is the only programming language that allows indexed variables and forces the programmer to use one as the first index. If you find another one let me know.
Even if it were true that no other language used 1 origin array indexing, so what? Popularity is no indication of correctness. If we cared about being like all the other languages we wouldn't have Smalltalk. If we cared about being like all the other kids we wouldn't have geeks like us to create stuff like Smalltalk. And then where would the world be?
Ultimately, Smalltalk, like all other languages, systems, cars, planes, boats, foods, political parties, comics, games etc is what it is. If you like it - Enjoy! If you don't - find one you do like and Enjoy!
tim -- tim Rowledge; tim@rowledge.org; http://www.rowledge.org/tim Asking whether machines can think is like asking whether submarines can swim.
squeak-dev@lists.squeakfoundation.org
-
Bill Spight -
Hans-Martin Mosner -
Herbert König -
Laurence Rozier -
nicolas cellier -
tim Rowledge -
Wolfgang Helbig -
Yoshiki Ohshima