Which language allows you to change an argument's value?

[winterheat]

I wonder which language allows you to change an argument’s value?
like:

foo(&a) {
a = 3
}

n = 1
print n

foo(n) # passing in n, not &n
print n

and now n will be 3. I think C++ and PHP can let you do that, using
their reference (alias) mechanism. And C, Python, and Ruby probably
won’t let you do that. What about Java and Perl?

is there any way to prevent a function from changing the argument’s
value?

isn’t “what i pass in, the function can modify it” not a desireable
behavior if i am NOT passing in the address of my argument? For one
thing, if we use a module, and call some functions in that module, and
the module’s author made some changes to his code, then we have no way
of knowing what we pass in could get changed. Of course, if it is in
Java, Python, and Ruby, and we pass in a reference to object (not C+
+'s meaning of alias reference), so the object can get changed, but
that can be expected, vs passing in n, when n = 1. Even when it is
Ruby, when everything is an object, passing n in when n = 1 won’t ever
make n become 3. Is there a way to prevent it from happening in the
languages that allows it?

[winterheat]

SpringFlowers AutumnMoon wrote:

we have no way
of knowing what we pass in could get changed.

Sure you do. You look at the function’s signature. In order to use
someone else’s library, you have to know the function’s signature. And
the signature explicitly tells you whether the value you pass in could
be changed.

[winterheat]

On 30 sep, 12:47, Summercool [email protected] wrote:

value?
…
Is there a way to prevent it from happening in the
languages that allows it?

Hi,
Of course in C++, functions that don’t modify argument’s value should
(i’d rather say MUST) wait for a CONST reference or a value :

// const ref
foo(const T& a) {
a = 3; // error

}
// value
foo(T a) {
a = 3; // ok, but modify only the formal parameter : the argument
(aka actual parameter) is not changed

}

Now if you want to prevent a function (from a library you are using)
to modify it… Erm well, you shouldn’t : a good librairy will never
wait for a non-const argument if it doesn’t need to be modified. So in
this case “Is there a way to prevent it from happening” is
unpertinent : you could copy the object, but if the purpose of the
fucntion was to modify it, it’s pointless.

[winterheat]

On Sep 30, 2007, at 12:50 PM, Summercool wrote:

and now n will be 3. I think C++ and PHP can let you do that, using
their reference (alias) mechanism. And C, Python, and Ruby probably
won’t let you do that. What about Java and Perl?

Java can’t do that, Perl only works that way.

languages that allows it?
In general it is bad practice to change the state of the arguments
you receive, as bad practice as changing the arguments themselves. A
function shouldn’t do that unless it documents it, unless it makes
sense and it is expected by the caller.

For example, that’s what allows you to write subroutines like chomp
in Perl:

chomp $str;

You expect $str to be chomped and since this is procedural style you
can implement that with pass-by-reference, and that’s fine.

Otherwise you’d need to pass a reference to $str to be able to change
the string through the reference (what you do in C for integers)
which is cumbersome, and in fact would be a workaround for pass-by-
value.

In Perl there’s no way to prevent a subroutine from changing the
values in the caller except passing readonly stuff like literals.
But, you know, people just don’t do that. The same way people don’t
start changing argument’s state gratuitously in Ruby. Both are
analogous bad practices.

– fxn

[winterheat]

On 2007-09-30 12:47, Summercool wrote:

foo(n) # passing in n, not &n
print n

and now n will be 3. I think C++ and PHP can let you do that, using

Since you know C++ can do it why do you include c.l.c++? Honestly cross-
posting to groups discussing so many different groups is seldom a good
idea since the languages differ too much for a useful discussion.

their reference (alias) mechanism. And C, Python, and Ruby probably
won’t let you do that. What about Java and Perl?

C will let you do it with pointers (it is just a syntactical difference
from references in this case) and Java’s references allows it. I do not
know about Ruby, Python and Perl, but (AFAIK) the all have OO support so
I would be surprised if they used purely value semantics.

is there any way to prevent a function from changing the argument’s
value?

Some languages, like C++ have the ability to make an argument const.

isn’t “what i pass in, the function can modify it” not a desireable
behavior if i am NOT passing in the address of my argument? For one

Being able to pass the actual object instead of a copy is highly
desirable for two reasons. In most languages only one return value is
allowed for a function so the ability to change parameters allows you to
artificially return more without having to wrap them in constructs. The
second reason is that for large objects the performance hit of having to
create a copy each time you call a function can be forbidding.

thing, if we use a module, and call some functions in that module, and
the module’s author made some changes to his code, then we have no way
of knowing what we pass in could get changed. Of course, if it is in

As for knowing when a library function modifies an argument or not, well
that is why we have documentation.

Java, Python, and Ruby, and we pass in a reference to object (not C+
+'s meaning of alias reference)

In what way does the C++ reference differ from those in Java, Python,
and Ruby in this situation?

[winterheat]

And C, Python, and Ruby probably won’t let you do that…

From my understanding, in Ruby all argument passing is by value.
However, the value passed may not be what you expect… The value that
is passed is that of the reference, not the value that the reference
refers to. So, for example:

def foo(arg)
arg = 3
end

a = ‘Mike’

foo(a)

This will not change the value of a in the top level. Why? Because foo
reassigns the argument reference arg to 3, a different object than
‘Mike.’ However:

def bar(arg)
arg[0] = ‘b’
end

a = ‘Mike’

bar(a)

This will change a from ‘Mike’ to ‘bike.’ Why? Because instead of
reassigning arg to a new object, the object itself (‘Mike’) has been
modified through the method call []. So, if the method simply reassigns
the reference, there will be no change in the original (it still refers
to the same object). But, if the method modifies that object, the
original reference will reflect that modification, since it is still
referencing the now modified object.

It is customary in Ruby to append an exclamation mark (!) to indicate
methods that actually modify an object in place. So, bar! would have
been a more appropriate name for the second method.

Hope that helps.

David B. Williams
http://www.cybersprocket.com

[winterheat]

On Sep 30, 3:47 am, Summercool [email protected] wrote:

isn’t “what i pass in, the function can modify it” not a desireable
behavior if i am NOT passing in the address of my argument? For one
thing, if we use a module, and call some functions in that module, and
the module’s author made some changes to his code, then we have no way
of knowing what we pass in could get changed. Of course, if it is in
Java, Python, and Ruby, and we pass in a reference to object (not C+
+'s meaning of alias reference), so the object can get changed, but
that can be expected, vs passing in n, when n = 1. Even when it is
Ruby, when everything is an object, passing n in when n = 1 won’t ever
make n become 3. Is there a way to prevent it from happening in the
languages that allows it?

Some would say that in truely good OO design, the state should be
protected by the object (not the constness of the reference to the
object)

[winterheat]

On Sep 30, 4:18 am, 7stud – [email protected] wrote:

SpringFlowers AutumnMoon wrote:

we have no way
of knowing what we pass in could get changed.

Sure you do. You look at the function’s signature. In order to use
someone else’s library, you have to know the function’s signature. And
the signature explicitly tells you whether the value you pass in could
be changed.

do you mean in C++? I tried to find signature in two C++ books and it
is not there. Google has a few results but it looks something like
prototype. Is signature the same as the function prototype in the .h
file? If so, don’t we usually just include <___.h> and forget about
the rest. Documentation is fine although in some situation, the
descriptions is 2 lines, and notes and warnings are 4, 5 times that,
and the users’ discussing it, holding different opinion is again 2, 3
times of that length. I think in Pascal and C, we can never have an
argument modified unless we explicitly allow it, by passing in the
pointer (address) of the argument.

also i think for string, it is a bit different because by default,
string is a pointer to char or the address of the first char in C and C
++. So it is like passing in the address already. it is when the
argument n is something like 1 that makes me wonder.

[winterheat]

On 2007-09-30 18:49, Summercool wrote:

do you mean in C++? I tried to find signature in two C++ books and it
is not there. Google has a few results but it looks something like
prototype. Is signature the same as the function prototype in the .h
file?

A signature is what is required to identify a function and includes
return type, name of function and the types of the parameters, while it
looks just like a prototype it is not. A prototype is something you
write to satisfy your compiler while a signature identifies a function.
Below are some examples of signatures, only the last can modify the
values of its parameter.

void foo(int, float)
std::string bar(const std::string&, int, int)
void baz(std::string&)

If so, don’t we usually just include <___.h> and forget about
the rest. Documentation is fine although in some situation, the
descriptions is 2 lines, and notes and warnings are 4, 5 times that,
and the users’ discussing it, holding different opinion is again 2, 3
times of that length.

Unless you read the documentation how do you know which files to
include? And what documentation are you reading which does not clearly
specify the functionality of the functions described?

I think in Pascal and C, we can never have an
argument modified unless we explicitly allow it, by passing in the
pointer (address) of the argument.

In C++ the arguments cannot be modified unless you either pass a pointer
to a non-const object or a non-const reference, so it is just as
explicit. (Notice that it is possible to cast the constness away, but
doing is extremely dangerous and should not be done.)

also i think for string, it is a bit different because by default,
string is a pointer to char or the address of the first char in C and C
++. So it is like passing in the address already.

No. A string in C++ is a string, a char array or a pointer to a char is
something different.

it is when the
argument n is something like 1 that makes me wonder.

Get a good book on whatever language you are interested in (I do not
know which it is since you are all over the place) and read up on that
languages references, if you still do not understand after that ask your
questions in the group discussing that language.

[winterheat]

.oO(Summercool)

I think in Pascal and C, we can never have an
argument modified unless we explicitly allow it, by passing in the
pointer (address) of the argument.

Pascal also allows passing by reference, which is done with the keyword
‘var’ when declaring the function parameters. Object Pascal also allows
const parameters.

Micha

[winterheat]

Summercool wrote:

is not there. Google has a few results but it looks something like
prototype. Is signature the same as the function prototype in the .h
file? If so, don’t we usually just include <___.h> and forget about
the rest. Documentation is fine although in some situation, the
descriptions is 2 lines, and notes and warnings are 4, 5 times that,
and the users’ discussing it, holding different opinion is again 2, 3
times of that length. I think in Pascal and C, we can never have an
argument modified unless we explicitly allow it, by passing in the
pointer (address) of the argument.

You need to get more C++ books :-). You generally won’t find them in
basic books, but some of the more advanced ones talk about function
signatures.

A C++ function’s signature is dependent on the function name, number of
parameters being passed, and the type of each parameter. This is passed
onto the linker. In any C++ program, every function signature must be
unique.

But in this case he’s a little incorrect. You could look at the
function’s signature, but it’s much easier just to look at the
function’s declaration.

[winterheat]

Erik Wikström wrote:

is not there. Google has a few results but it looks something like
prototype. Is signature the same as the function prototype in the .h
file?

A signature is what is required to identify a function and includes
return type, name of function and the types of the parameters, while it
looks just like a prototype it is not. A prototype is something you
write to satisfy your compiler while a signature identifies a function.
Below are some examples of signatures, only the last can modify the
values of its parameter.

No, the function’s signature does not include the return type.

void foo(int, float)
std::string bar(const std::string&, int, int)
void baz(std::string&)

Actually, these are prototypes. The signature is what’s passed onto the
linker - perhaps something like “?foo@@YAXHM@Z”. It’s also generally
compiler dependent for C++ mangled names.

Also, I should clarify - the C++ function signature includes parameter
information. The C function signature is only dependent on the function
name and is almost always the name preceded by an underscore).

You can see the function signatures in a map file if you leave the names
mangled.
.

If so, don’t we usually just include <___.h> and forget about
the rest. Documentation is fine although in some situation, the
descriptions is 2 lines, and notes and warnings are 4, 5 times that,
and the users’ discussing it, holding different opinion is again 2, 3
times of that length.

Unless you read the documentation how do you know which files to
include? And what documentation are you reading which does not clearly
specify the functionality of the functions described?

Exactly.

[winterheat]

Summercool wrote:

foo(n) # passing in n, not &n
print n

and now n will be 3. I think C++ and PHP can let you do that, using
their reference (alias) mechanism. And C, Python, and Ruby probably
won’t let you do that. What about Java and Perl?

I think you’ve missed how Python works, and probably others.
A Python function receives a reference to the argument, and
can modify the object if the object is mutable.

Nevertheless, assigning to the parameter’s name will not
change the passed object. This function does nothing:

 def clear_list_wrong(lst):
     lst = []  # useless re-binding of local name

This version empties the passed list:

 def clear_list(lst):
     del lst[:]

is there any way to prevent a function from changing the argument’s
value?

Sure. First choice: Don’t change the value in the function.
Alternatives include making a copy to use as the argument.

isn’t “what i pass in, the function can modify it” not a desireable
behavior if i am NOT passing in the address of my argument? For one
thing, if we use a module, and call some functions in that module, and
the module’s author made some changes to his code, then we have no way
of knowing what we pass in could get changed.

Don’t rely on undocumented behavior. Modules worth using are by
good programmers. Authors of library modules tend to be zealous
about not breaking client code.

[winterheat]

Erik Wikström wrote:

their reference (alias) mechanism. And C, Python, and Ruby probably
won’t let you do that. What about Java and Perl?

C will let you do it with pointers (it is just a syntactical difference
from references in this case) and Java’s references allows it.

Neither C or Java has call by reference.

C pointers and Java references may work similarly in most cases
but it is still call by value.

                                                           I do not

know about Ruby, Python and Perl, but (AFAIK) the all have OO support so
I would be surprised if they used purely value semantics.

I can not see why OO should indicate anything about call by reference
support.

isn’t “what i pass in, the function can modify it” not a desireable
behavior if i am NOT passing in the address of my argument? For one

Being able to pass the actual object instead of a copy is highly
desirable for two reasons. In most languages only one return value is
allowed for a function so the ability to change parameters allows you to
artificially return more without having to wrap them in constructs. The
second reason is that for large objects the performance hit of having to
create a copy each time you call a function can be forbidding.

Usually it is not a good thing, because it makes the code much
more difficult to read.

But sometimes it is handy.

I think C# got it right.

It allows it but require an explicit marking of it in both formal
argument list and actual argument list.

Java, Python, and Ruby, and we pass in a reference to object (not C+
+'s meaning of alias reference)

In what way does the C++ reference differ from those in Java, Python,
and Ruby in this situation?

C++ and Java are very different in this aspect.

Arne

[winterheat]

On Mon, 01 Oct 2007 01:53:47 GMT, Roedy Green
[email protected] declaimed the following in
comp.lang.python:

Seem to me you could in FORTRAN and Pascal, and maybe even Algol. It
has been a while.

Everything in classic FORTRAN is a passed as a reference – even
constant arguments are passed as a reference to the memory location
containing that constant (which is why it was possible in very early
FORTRANs to have “a = 1 + 1” yield something other than “2” if preceded
by, say, “call mutate(1)” where mutate looks like:

subroutine mutate(arg)
arg = arg * 2
end

)

[winterheat]

On Sun, 30 Sep 2007 10:47:13 -0000, Summercool
[email protected] wrote, quoted or indirectly quoted someone
who said :

and now n will be 3. I think C++ and PHP can let you do that, using
their reference (alias) mechanism. And C, Python, and Ruby probably
won’t let you do that. What about Java and Perl?

Seem to me you could in FORTRAN and Pascal, and maybe even Algol. It
has been a while.

I have come to appreciate Java’s strong isolation convention. It makes
it a lot easier to track down WHERE in the code a value could
potentially change and keeps those places to a minimum.

[winterheat]

On 2007-09-30 23:03, Arne Vajhøj wrote:

Erik Wikström wrote:

their reference (alias) mechanism. And C, Python, and Ruby probably
won’t let you do that. What about Java and Perl?

C will let you do it with pointers (it is just a syntactical difference
from references in this case) and Java’s references allows it.

Neither C or Java has call by reference.

I never said that, what I said was that C allows you to simulate it by
passing pointers instead, and since you always use references in Java
(except for primitive types) you always pass a reference.

C pointers and Java references may work similarly in most cases
but it is still call by value.

The difference I am trying to show is that between value semantics
(where a copy of the parameter is used in the function) verses reference
semantics where you pass a reference to the object. It is true that when
using a pointer in C you pass a copy of the pointer, but not of the
object it refers to. Similarly you might pass a copy of the variable
holding a reference in Java, but not a copy of the object it refers to.
Thus, in the function you will be working on the referred object, just
like when passing by reference in C++.

know about Ruby, Python and Perl, but (AFAIK) the all have OO support so
I would be surprised if they used purely value semantics.

I can not see why OO should indicate anything about call by reference
support.

There is no direct connection, except that all OO programming languages
that I know of supports reference semantics. You seldom want to work on
copies of objects, rather you want to work on the original object and
for that you need to be able to refer to it, for that you need some kind
of reference.

Usually it is not a good thing, because it makes the code much
more difficult to read.

But sometimes it is handy.

I think C# got it right.

It allows it but require an explicit marking of it in both formal
argument list and actual argument list.

I think you are confusing two different things here, one is the ability
to pass a reference to an object, instead of a copy of the object. This
is what C++ references allows you to do, just like Java references, C
pointers, and any other kind of references that I know about.

The other thing is to allow the function to change what a reference
refers to, to this not possible with C++ references, C pointers, Java
references, or most other reference types that I know about. You can do
this by using a reference to a pointer in C++, a pointer to a pointer in
C, or using the ref and out keywords in C#. Doing that can be confusing,
but often the semantics for doing so are pretty obvious.

Java, Python, and Ruby, and we pass in a reference to object (not C+
+'s meaning of alias reference)

In what way does the C++ reference differ from those in Java, Python,
and Ruby in this situation?

C++ and Java are very different in this aspect.

Might be, my question was about what aspect we are talking about.

[winterheat]

Neither C or Java has call by reference.
C pointers and Java references may work similarly in most cases
but it is still call by value.
so what? the references in c++ are passed by value too, its just a nice
interface to pointers.
At the end those parameters are pushed on the stack … thats it.

[winterheat]

Summercool [email protected] writes:

I wonder which language allows you to change an argument’s value?

[…]

What about Java and Perl?

Perl will let you change the value of a passed-in object directly.
Others have already answered about Java.

is there any way to prevent a function from changing the argument’s
value?

Make a copy of the object, and pass in the copy.

Without making a copy, of the languages I know, C++ comes closest to
supporting an unmodifiable argument passed by reference. Using a
const reference or const pointer indicates that the reference won’t be
changed, but even that can be subverted by the function’s author with
casting.

-----Scott.

[winterheat]

On Sep 30, 6:49 pm, Summercool [email protected] wrote:

On Sep 30, 4:18 am, 7stud – [email protected] wrote:

SpringFlowers AutumnMoon wrote:

we have no way
of knowing what we pass in could get changed.

Sure you do. You look at the function’s signature. In order to use
someone else’s library, you have to know the function’s signature. And
the signature explicitly tells you whether the value you pass in could
be changed.

That’s not really correct. It’s more a case that if you don’t
know what a function does, you don’t use it. Regardless of the
language, calling a function without knowing what it does is a
sure recepe for disaster.

do you mean in C++? I tried to find signature in two C++ books and it
is not there. Google has a few results but it looks something like
prototype. Is signature the same as the function prototype in the .h
file? If so, don’t we usually just include <___.h> and forget about
the rest. Documentation is fine although in some situation, the
descriptions is 2 lines, and notes and warnings are 4, 5 times that,
and the users’ discussing it, holding different opinion is again 2, 3
times of that length. I think in Pascal and C, we can never have an
argument modified unless we explicitly allow it, by passing in the
pointer (address) of the argument.

In C, that’s partially true; no part of the expressions involved
in the function invocation will be modified through the function
argument unless the address is specifically taken (or the
argument has array type). The situation in Pascal, C++, Ada and
most other languages is different; Pascal and C++ have reference
parameters (called VAR in Pascal), as well as value paramters,
Ada has in, inout and out parameters.

The situation in Java is somewhat more complicated; formally,
Java only has pass by value, and there’s no way a function can
modify any of the arguments. In practice, however, if the
expression has an object type, the “value” is a pointer (called
reference in Java) to the object; if the object is a well
defined “value” type (e.g. java.lang.String), it will be
immutable, but entity objects don’t follow this rule, and not
all value objects (e.g. java.awt.Dimension) are well designed.

also i think for string, it is a bit different because by default,
string is a pointer to char or the address of the first char in C and C
++.

Strings in C++ are class types, with full value semantics.

So it is like passing in the address already. it is when the
argument n is something like 1 that makes me wonder.

C++ requires an lvalue unless the reference is to a const, so
you cannot pass 1 to a C++ function and expect/worry about it
being changed. The same holds for most modern languages, I
think, although the mechanisms involved may differ. (I’d be
very surprised, for example, if Ada allowed anything but what in
C++ would be an lvalue to be passed to an inout or an out
parameter.) In most modern languages, however, you can
circumvent the controls with enough indirections.

–
James K. (GABI Software) email:[email protected]
Conseils en informatique orientée objet/
Beratung in objektorientierter Datenverarbeitung
9 place Sémard, 78210 St.-Cyr-l’École, France, +33 (0)1 30 23 00 34