Naming convention for accessor methods (get/set)

[Generic Usenet Account]

A lot has been said in this newsgroup regarding the "evil" set/get
accessor methods. Arthur Riel, (of Vanguard Training), in his class,
"Heuristis for O-O Analysis & Design", says that there is almost never
an excuse for accessor methods. Personally, I do not go that far. I
do feel that they serve a useful purpose (albeit in a limited manner).
Personally I prefer dropping the "set" and "get" prefixes from the
method names altogether. For example:

class A
{
public:
...
int xyz() { return _xyz; } // instead of get_xyz()
void xyz(const int val) { _xyz = val; } // instead of get_xyz()
...

protected:
...
int _xyz;
...
};


This naming convention is also consistent with the IDL/C++ language
binding, and I wanted to seek your opinion regarding this.

Regards,
KP Bhat

 

[Kevin Goodsell]

A lot has been said in this newsgroup regarding the "evil" set/get
accessor methods. Arthur Riel, (of Vanguard Training), in his class,
"Heuristis for O-O Analysis & Design", says that there is almost never
an excuse for accessor methods. Personally, I do not go that far. I
do feel that they serve a useful purpose (albeit in a limited manner).
Personally I prefer dropping the "set" and "get" prefixes from the
method names altogether. For example:

class A
{
public:
...
int xyz() { return _xyz; } // instead of get_xyz()
void xyz(const int val) { _xyz = val; } // instead of get_xyz()
...

protected:
...
int _xyz;
...
};


This naming convention is also consistent with the IDL/C++ language
binding, and I wanted to seek your opinion regarding this.

(Cross-post to comp.lang.c++.moderated removed. Cross-posting to
moderated groups slows down responses.)

My opinion:

I don't see a real problem with your naming scheme, but I wouldn't use
it personally. Overloaded functions are generally used when multiple
functions perform the same (or a similar) task using different
arguments. In this case, the functions do rather different things, so I
would prefer different names.

Set/get member functions aren't inherently bad, but some beginners get
the idea that they are a standard part of every class, which is
completely wrong. Most classes shouldn't have a need for typical set and
get functions (those that correspond to the actual data members in the
class). The interface should be separate from the representation
(obviously) so except in rather simple cases, you can't expect a
well-designed interface to correspond to the members used to represent
the class's state.

One last thing, I would recommend not using names that begin with an
underscore. It's OK in the context above, but there's a lot of cases
where it's not. The easiest way to avoid problems is to get in the habit
of never using identifiers (macros, variables, etc.) that begin with an
underscore.

-Kevin

 

[Gianni Mariani]

Generic Usenet Account wrote: ....
(Cross-post to comp.lang.c++.moderated removed. Cross-posting to
moderated groups slows down responses.)

My opinion:

I don't see a real problem with your naming scheme, but I wouldn't use
it personally. Overloaded functions are generally used when multiple
functions perform the same (or a similar) task using different
arguments. In this case, the functions do rather different things, so I
would prefer different names.

Set/get member functions aren't inherently bad, but some beginners get
the idea that they are a standard part of every class, which is
completely wrong. Most classes shouldn't have a need for typical set and
get functions (those that correspond to the actual data members in the
class). The interface should be separate from the representation
(obviously) so except in rather simple cases, you can't expect a
well-designed interface to correspond to the members used to represent
the class's state.

One last thing, I would recommend not using names that begin with an
underscore. It's OK in the context above, but there's a lot of cases
where it's not. The easiest way to avoid problems is to get in the habit
of never using identifiers (macros, variables, etc.) that begin with an
underscore.

Opinion seconded.

 

[Samuel Krempp]

le Friday 19 September 2003 13:08, (e-mail address removed) écrivit :

int xyz() { return _xyz; } // instead of get_xyz()
void xyz(const int val) { _xyz = val; } // instead of get_xyz()

There is one small factual inconvienence I can think of.
While you can do :

vector<A> v;
int x;
for_each(v.begin(), v.end(), bind2nd(mem_fun_ref(&A::set_xyz), x) );

You need specific code to reach the overloaded 'xyz' membre function :

void (A::* pmf)(const int) = &A::xyz;
for_each(v.begin(), v.end(), bind2nd(mem_fun_ref(pmf), x) );


oh, and it's easier to search for 'set_xyz' in a file. (the overloaded one
can be searched using a regexp like 'xyz([^)]*)' , it's not too complicated
either)

except for those details, I think it does'nt make much of a difference.

 

[Le Chaud Lapin]

Opinion seconded.

How about ending with underscore?

-Chaud Lapin-

 

[Kevin Goodsell]

How about ending with underscore?

That's fine as long as it's not immediately preceded by another
underscore. Identifiers containing a sequence of two underscores are
also reserved (though I don't know what for off the top of my head). So
this:

identifier_

is OK. These are not:

_identifier (well, it's OK in some places)
_Identifier (pretty much never OK)
_IDENTIFIER (also never OK)
identifier__
__identifier
foo__bar

-Kevin

 

[=?iso-8859-1?Q?Juli=E1n?= Albo]

Generic Usenet Account escribió:

Personally I prefer dropping the "set" and "get" prefixes from the
method names altogether. For example:
(snip)

This naming convention is also consistent with the IDL/C++ language
binding, and I wanted to seek your opinion regarding this.

In Spain we say: "aunque la mona se vista de seda, mona se queda". "A
monkey is a monkey even if dressed in silk" can be a translation. It is
the same thing wheter you use or not set and get prefixes. Personal
preference, as you say.

Regards.

 

[Marcin Vorbrodt]

How about:

class A {
public:
int& EVIL() { return _evil; }
const int& EVIL() const { return _evil; }
};

This really is a horrible design... because you are exposing the underlying
implementation of your class. Which implies that if the implementation
changes, then the interface may need to be changed as well. That is more
evil than Sadam, Bin Laden, Stalin and Hitler put together. If you need to
be able to create setters and getters, why not just make the variable
public. If you need to have some other unrelated class access
private/protected elements of some other class, use friends instead. At
least you will not be breaking encapsulation.

GRRR

Martin

 

[Mike Wahler]

A lot has been said in this newsgroup regarding the "evil" set/get
accessor methods. Arthur Riel, (of Vanguard Training), in his class,
"Heuristis for O-O Analysis & Design", says that there is almost never
an excuse for accessor methods. Personally, I do not go that far. I
do feel that they serve a useful purpose (albeit in a limited manner).
Personally I prefer dropping the "set" and "get" prefixes from the
method names altogether. For example:

class A
{
public:
...
int xyz() { return _xyz; } // instead of get_xyz()
void xyz(const int val) { _xyz = val; } // instead of get_xyz()
...

protected:
...
int _xyz;
...
};


This naming convention is also consistent with the IDL/C++ language
binding, and I wanted to seek your opinion regarding this.

I also prefer to overload a single meaningful name for this.

This is typically (but not always) feasible since one usually
takes an argument and the other does not. This form has worked
well for me so far.


-Mike

 

[apm]

A lot has been said in this newsgroup regarding the "evil" set/get
accessor methods. Arthur Riel, (of Vanguard Training), in his class,
"Heuristis for O-O Analysis & Design", says that there is almost never
an excuse for accessor methods. Personally, I do not go that far. I
do feel that they serve a useful purpose (albeit in a limited manner).
Personally I prefer dropp

So do I. Here's why:

The Uniform Access Principle espoused by Betrand Meyer states that all
services offerd by a module should be available through a uniform
notation which does not betray whether they are implemented through
storage or through computation. Get get/set convention violates this.

-Andrew Marlow

 

[Francis Glassborow]

How about:

class A {
public:
int& EVIL() { return _evil; }
const int& EVIL() const { return _evil; }
};

This really is a horrible design... because you are exposing the underlying
implementation of your class. Which implies that if the implementation
changes, then the interface may need to be changed as well. That is more
evil than Sadam, Bin Laden, Stalin and Hitler put together. If you need to
be able to create setters and getters, why not just make the variable
public. If you need to have some other unrelated class access
private/protected elements of some other class, use friends instead. At
least you will not be breaking encapsulation.

But you snuck in return by reference which is a quite different issue.
Consider the following public interface for a class:

class point2d {
public:
// constructor:
explicit point2d(double xval=0, double yval=0);
// read access functions
double x()const;
double y()const;
double modulus()const;
double argument()const;
// write access functions
point2d & x(double xval);
point2d & y(double yval);
point2d & modulus(double mod);
point2d & argument(double degrees);
private:
// data choice irrelevant

};
Note that the user has no need to know how the data locating a point on
a plane is organised. However there is a great deal of benefit from
being able to select using Cartesian or Polar co-ordinates according to
what you are doing.

 

[WW]

(e-mail address removed) (Generic Usenet Account) wrote in message


So do I. Here's why:

The Uniform Access Principle espoused by Betrand Meyer states that all
services offerd by a module should be available through a uniform
notation which does not betray whether they are implemented through
storage or through computation. Get get/set convention violates this.

Without going to quoting authority opinions and theories I would rather
tackle the problem from a T word point of view. A class (be it any class)
represents a concept in a problem domain. if that concept suport the
existence of things called getXxx setXxx then you should have them. If not,
you should not.

Furthermore a class is either just a representation of some simple storage
facility (in which case getters and setter might be appropriate) or it is
not. IMHO it is very very rare in real life design when you need such a
class. And if you feel you do, you really have to defend the idea, because
unless that class provides an adapter or facade to an exteranl storage
system they seem to have absolutely no function whatsoever.

Just let me give one example of a place where get/set might be appropriate
(as a name):

struct Radio {
xxx setVolume()...
xxx getVolume()...
private:
....
};

 

[Jerry Coffin]

[ ... ]

That's fine as long as it's not immediately preceded by another
underscore. Identifiers containing a sequence of two underscores are
also reserved (though I don't know what for off the top of my head).

Just in case you happen to really care: it's reserved because some
compilers used that as the separator between the name and the name
mangling -- this was (and may still be) particularly widely used among
compilers that produced C as output, since the result had to be a legal
C identifier.

 

[Ben Liddicott]

A lot has been said in this newsgroup regarding the "evil" set/get
accessor methods. Arthur Riel, (of Vanguard Training), in his class,
"Heuristis for O-O Analysis & Design", says that there is almost never
an excuse for accessor methods.

Accessor methods can become neccessary when you have several properties of a class which are really different views/transformations of the same underlying property. An example already given was polar vs. cartesian coordinates.

Here's another:

class P{
public:
// directory
std::string GetDirectory();
void PutDirectory(const std::string& newDir);
// filename = basename.extension
std::string GetFileName();
void PutFileName(const std::string& newFileName);
// full path
std::string GetFullPath();
void PutFullPath(const std::string& newPath);
// now add BaseName and Extension.
};

// example:
// this/is/the/directory/basename.extension


It may perfect sense to Get/Put any of these five properties independently; depending on what the program intends to do with the information. If it wants to create a sibling file, it can use GetDirectory. If it wants to name a known sibling file according to some convention, it might want to set the extension only. If it wants to name a companion file in a different folder, to set the directory, and so forth.

Personally I prefer dropping the "set" and "get" prefixes from the
method names altogether. For example:
(deleted)

This naming convention is also consistent with the IDL/C++ language
binding, and I wanted to seek your opinion regarding this.

You are talking about the CORBA IDL C++ language binding of course... as opposed to the many other IDL's, such as DCE/RPC, MIDL, MACH, Sun RPC and so forth.

It is time to invoke the concept of Least Surprise. If the code is intended or likely to be mixed with CORBA objects and calls, this would be a good convention to use. If it is intended or likely to be used together with with some other convention, you may wish to use that convention.

Personally I prefer Get/Put as prefixes, because one day you will need an indexed property.

class A{
public:
std::string GetValue(const std::string& name);
void PutValue(const std::string& name, const std::string& value);
};

Some times it is appropriate to use more than one convention:


class B{
public:
long get_Value(const std::string& name, std::string& value) throw();
long put_Value(const std::string& name, const std::string& value) throw();
std::string GetValue(const std::string& name) throw(MyException){
std::string tmp; long lErrCode = get_Value(name, tmp);
if(lErrCode)throw MyException(lErrCode);
return tmp;
}
void PutValue(const std::string& name, const std::string& value) throw(MyException){
long lErrCode = put_Value(name, value);
if(lErrCode)throw MyException(lErrCode);
}
};


This gives more flexibility to the user of the class. The verbosity and duplication involved mean that it is only really appropriate for very small or very stable libraries which are frequently used and thoroughly tested, or of course, for machine-generated code, for example from some sort of IDL.

 

[Ed Avis]

The Uniform Access Principle espoused by Betrand Meyer states that
all services offerd by a module should be available through a uniform
notation which does not betray whether they are implemented through
storage or through computation. Get get/set convention violates this.

Could you explain what you mean - it seems that wrapping everything in
get_x() and set_x() would indeed hide what was stored and what was
computed (though it's not clear what it means to call set_x() if x is
a computed value).

 

[Ed Avis]

I believe Stroustrup favours the style of accessor where a single
overloaded function name is used for both getting and setting.
At least, ISTR seeing such a thing in the C++ book, and didn't see
any get_x() or set_x() methods.

 

[Ed Avis]

class A {
public:
int& EVIL() { return _evil; }
const int& EVIL() const { return _evil; }
};

This really is a horrible design... because you are exposing the underlying
implementation of your class.

It's not necessarily evil - for some concrete types it may be simplest
to expose the implementation. If a complex number has 'real' and
'imag', and you don't plan to change to polar coordinates later, you
might as well expose those rather than clutter the interface with
accessors. Especially so if you are writing this class for your own
use and can simply change all the callers in the unlikely event of the
implementation changing.

However if you are writing a class for 'complex number with magnitude
less than 1', then you wouldn't make the members public because it
would let a caller too easily break the class invariant.

If you need to be able to create setters and getters, why not just
make the variable public.

Indeed. The above code, however, may be preferred in shops where
coding standards mandate using accessors for all members and never
making them public. It's a way to simulate public visibility, in
cases where that would be the best design, while staying within the
letter of the religious law. (Of course, in such cases the accessors
would probably be called getEvil() or getMyEvil() or
pIGetMyEvilAsReference().)

 

[James Kanze]

(e-mail address removed) (apm) writes:

|> (e-mail address removed) (Generic Usenet Account) wrote in message
|> > A lot has been said in this newsgroup regarding the "evil" set/get
|> > accessor methods. Arthur Riel, (of Vanguard Training), in his
|> > class, "Heuristis for O-O Analysis & Design", says that there is
|> > almost never an excuse for accessor methods. Personally, I do not
|> > go that far. I do feel that they serve a useful purpose (albeit in
|> > a limited manner). Personally I prefer dropp

|> So do I. Here's why:

|> The Uniform Access Principle espoused by Betrand Meyer states that
|> all services offerd by a module should be available through a
|> uniform notation which does not betray whether they are implemented
|> through storage or through computation. Get get/set convention
|> violates this.

So do any of the alternatives in C++.

Or rather, none of them do, if used consistently. How does get and set
betray the underlying implementation?

 

[Dave Harris]

class A {
public:
int& EVIL() { return _evil; }
const int& EVIL() const { return _evil; }
};

This really is a horrible design... [...] If you need to be able
to create setters and getters, why not just make the variable
public.

There are degrees of encapsulation. Your interface at least allows
some indirection. For example:

int &EVIL() {
if (_pImpl == NULL)
_pImpl = new Impl;
return _pImpl->EVIL();
}

There is quite a lot of flexibility available. Eg if A has more than
one variable, we can use more than one Impl class and allocate them
at different times.

Reference-returning functions restrict the implementation less than
exposing an actual instance variable. Personally I avoid them (even
for things like vector:perator[]), but they are not evil.

-- Dave Harris, Nottingham, UK

 

[Torsten Robitzki]

A lot has been said in this newsgroup regarding the "evil" set/get
accessor methods. Arthur Riel, (of Vanguard Training), in his class,
"Heuristis for O-O Analysis & Design", says that there is almost never
an excuse for accessor methods. Personally, I do not go that far. I
do feel that they serve a useful purpose (albeit in a limited manner).
Personally I prefer dropping the "set" and "get" prefixes from the
method names altogether. For example:

class A
{
public:
...
int xyz() { return _xyz; } // instead of get_xyz()
void xyz(const int val) { _xyz = val; } // instead of get_xyz()
...

protected:
...
int _xyz;
...
};


This naming convention is also consistent with the IDL/C++ language
binding, and I wanted to seek your opinion regarding this.

I'm unsure if you are looking for opinions of if accessor functions are
a design mistake or how to name them in the rare cases where they are
useful.

For the latter I would call them getX() and setX() (or get_x...) just
for the case I have to address one of the functions as an parameter to
an ScopeGuard to rollback a prior set value back to its original value.

But if you are looking for some excuse to use accessors (especial
setters) I would tend to say that it's a design flaw in most cases. I
experienced two situations in my career by now, where setters where
useful. The first was a class that cached configuration data from a data
base by it's getters but also had setters to change the configuration.
But the setters didn't only updated the configuration into the DB but
informed other processes about the configuration changes, so the same
class linked into an other process read the configuration from the DB.
The second case where I used setters where I had to create Object as
copy from other object and to make little changes on them.

regards Torsten