Read Only Attribute?

[Brian Genisio]

Hello all,

I am developing a class where speed is important, and memory size is
also important. I need to give the user of the class read-only access
to one of the private members (size as an int). I have two options:

1. Make a method : int getSize() {return m_size;}
2. Use a reference: const int &size;, and link it to the private size value

Option #1 is nice, because it does not take an extra member, but it is
slower.
Option #2 is nice, because it is faster than a method, but it takes 4
bytes per instance of the class

In a non-optimized mode, option 1 is 2 times slower than option 2 (from
some simple tests I did). On the other hand, in optimized mode, I did
not notice a difference, as far as I can tell... but I think my
optimizer was optimizing a lot more, so I am not sure that my test was
accurate.

What do you think? What is the most EFFICIENT way to do things?
Usability and maintenance is not much of a concirn for me, since
efficiency is much more important.

Thanks,
Brian

I think I gave enough info, but in case you want to see the code, here
is the test application.
#include <iostream.h>
#include "timer.h"

class Tester
{
public:
int getSize() { return m_size; }
const int &size;

Tester() : m_size(0), size(m_size) {}
Tester(int setter) : m_size(setter), size(m_size) {}
private:
int m_size;
};
#define LOOP_SIZE 10000000

int main(void)
{
int i;
Tester my_test(20);
int temp_size;
BGTimer timer;

timer.start();
for(i=0; i<LOOP_SIZE; i++)
temp_size = my_test.getSize();
timer.end_and_print();

timer.start();
for(i=0; i<LOOP_SIZE; i++)
temp_size = my_test.size;
timer.end_and_print();

return 0;
}

[Martijn Lievaart]

On Thu, 08 Jan 2004 09:33:21 -0500, Brian Genisio wrote:

Hello all,

I am developing a class where speed is important, and memory size is
also important. I need to give the user of the class read-only access
to one of the private members (size as an int). I have two options:

1. Make a method : int getSize() {return m_size;}
2. Use a reference: const int &size;, and link it to the private size value

Option #1 is nice, because it does not take an extra member, but it is
slower.
Option #2 is nice, because it is faster than a method, but it takes 4
bytes per instance of the class

In a non-optimized mode, option 1 is 2 times slower than option 2 (from
some simple tests I did). On the other hand, in optimized mode, I did
not notice a difference, as far as I can tell... but I think my
optimizer was optimizing a lot more, so I am not sure that my test was
accurate.

Any decent compiler will generate exactly the same code for the two
methods with even modest optimization turned on. #1 is way more idiomatic
and maintenance friendly. #2 has the added disadvantage it grows the
object. Go with #1 and endure the slower speeds on debug builds.

HTH,
M4

[Victor Bazarov]

"Brian Genisio" <Br**********@yahoo.com> wrote...

I am developing a class where speed is important, and memory size is
also important. I need to give the user of the class read-only access
to one of the private members (size as an int). I have two options:

1. Make a method : int getSize() {return m_size;}
2. Use a reference: const int &size;, and link it to the private size value
Option #1 is nice, because it does not take an extra member,
What do you mean by "extra member"?
but it is
slower.
And you assert this based on what?
Option #2 is nice, because it is faster than a method, but it takes 4
bytes per instance of the class
Again, what is the base for this assertion? The language Standard
says that it is unspecified whether references require storage.
In a non-optimized mode, option 1 is 2 times slower than option 2 (from
some simple tests I did).
Please explain how those _simple_ tests relate to the real programs
you're to write.
On the other hand, in optimized mode, I did
not notice a difference, as far as I can tell... but I think my
optimizer was optimizing a lot more, so I am not sure that my test was
accurate.
How can you be sure of the accuracy of the non-optimised code, then?
What if your non-optimiser added some running-around a lot more?
What do you think? What is the most EFFICIENT way to do things?
The simpler and more straight-forward it is, the more efficient it
is. I would undoubtedly use the method #1.
Usability and maintenance is not much of a concirn for me, since
efficiency is much more important.
Usability and maintenance is always a concern. If not to you, then
to somebody who's going to maintain your code. Also, efficiency and
maintainability are not mutually exclusive. Code doesn't have to be
obfuscated to perform well.

Thanks,
Brian

I think I gave enough info, but in case you want to see the code, here
is the test application.
#include <iostream.h>
#include "timer.h"

class Tester
{
public:
int getSize() { return m_size; }
const int &size;

Tester() : m_size(0), size(m_size) {}
Tester(int setter) : m_size(setter), size(m_size) {}
private:
int m_size;
};
#define LOOP_SIZE 10000000

int main(void)
{
int i;
Tester my_test(20);
int temp_size;
You should declare temp_size 'volatile'. Otherwise the optimiser
should throw your loops away as having no side effect.
BGTimer timer;

timer.start();
for(i=0; i<LOOP_SIZE; i++)
temp_size = my_test.getSize();
timer.end_and_print();

timer.start();
for(i=0; i<LOOP_SIZE; i++)
temp_size = my_test.size;
timer.end_and_print();

return 0;
}

[Brian Genisio]

Usability and maintenance is not much of a concirn for me, since
efficiency is much more important.

Usability and maintenance is always a concern. If not to you, then
to somebody who's going to maintain your code. Also, efficiency and
maintainability are not mutually exclusive. Code doesn't have to be
obfuscated to perform well.

No, code does not have to be obfuscated to perform well... but sometimes
it does. I also agree that Usability and maintenance is always a
concern, but in this case, for this issue, it is not _much_ of a concern
(as I noted).

B

[Kevin Saff]

"Brian Genisio" <Br**********@yahoo.com> wrote in message
news:3f******@10.10.0.241...

Hello all,

I am developing a class where speed is important, and memory size is
also important. I need to give the user of the class read-only access
to one of the private members (size as an int). I have two options:

1. Make a method : int getSize() {return m_size;}
That should be : int getSize() const {return m_size;}
"const" should be used on all member functions that do not change the state
of an object.

First, you should not worry about efficiency until you have readable,
tested, and correct code for your entire program. Then, use a profiler to
see what exact functions need to be improved. Intuition is often wrong
about these things, and you are better off having readable code than making
possibly unnecessary optimizations.

If you include the function definition in the class definition like you do
here, C++ treats the function as "inline", which is a strong hint to the
compiler to "inline" the function - directly incorporating the function code
into the calling code without a function call. In this case, only a very
bad compiler would generate a function call for this. The compiled code
will probably be identical for:

int x = object.getSize();

as for:

int x = object.m_size;

if m_size were public.
2. Use a reference: const int &size;, and link it to the private size value

This is a bad idea. See http://cpptips.hyperformix.com/cpptips/ref_danger
for some reasons.

In addition, contrary to your opinion, a reference is usually stored as a
pointer, so using a public reference like this introduces an additional
pointer indirection which is not present when the above code is inlined - in
other words #1 in a release build will (probably) access the int directly
whereas #2 requires a pointer lookup. Your intuition on the speeds of these
functions is reverse of the truth.

From a design standpoint, if you couple your design so tightly to storing
size as an int, you may as well make that member public.

Furthermore, "size" is something that might not make sense to store directly
as an int, and not accessing it through a function will restrict your
design. "Optimizing" here can prevent you from doing more important
algorithmic improvements. Short-sighted optimization like this is unlikely
to result in an overall faster program, but is likely to restrict your
options in making algorithmic improvements.
In a non-optimized mode, option 1 is 2 times slower than option 2 (from
some simple tests I did). On the other hand, in optimized mode, I did
not notice a difference, as far as I can tell... but I think my
optimizer was optimizing a lot more, so I am not sure that my test was
accurate.
Your "non-optimized" mode is probably a debug build, which does not inline
the function because then you cannot manually step through the source.

If you don't trust your tests, why do them? Here's another test for you:
dynamically allocate (operator "new") a bunch of these objects. In most
programs dynamic allocation is a much larger bottleneck than function calls
(even if the function were not inlined). Since your 2nd method increases
the size of the object, I bet it will be much slower in dynamic allocation
in both debug and release builds.
What do you think? What is the most EFFICIENT way to do things?
The inline function. Hands down.
Usability and maintenance is not much of a concirn for me,
In that case, please do not write any code which I may ever have to look at.
since efficiency is much more important.

You cannot know if efficiency is important unless you have tests of correct
code which indicate in which areas efficiency is important.

HTH
--
KCS

[Howard]

"Kevin Saff" <go********@kevin.saff.net> wrote in message
news:Hr********@news.boeing.com...

You cannot know if efficiency is important unless you have tests of correct code which indicate in which areas efficiency is important.

HTH
--
KCS

I agreed with everything you wrote, up to this point. I work with real-time
audio software that has portions of it where procesing must take place in
the absolute minimum amount of time possible, with as little access to other
objects or function calls as possible. In these instances, I sometimes have
to intentionally do what would otherwise be poor coding practices, because I
need to get in, compute a large volume of information VERY quickly, and get
out. So there *are* occasions where efficiency is *known* to be more
important. It's rare, admittedly, and I only do this in these rare,
isolated circumstances. But it does happen, and unless you know the details
of why the OP says efficiency is more important, you cannot state with
accuracy that he's wrong.
-Howard

[Kevin Saff]

"Howard" <al*****@hotmail.com> wrote in message
news:bt********@dispatch.concentric.net...

"Kevin Saff" <go********@kevin.saff.net> wrote in message
news:Hr********@news.boeing.com...

You cannot know if efficiency is important unless you have tests of correct

code which indicate in which areas efficiency is important.

HTH
--
KCS

I agreed with everything you wrote, up to this point. I work with

real-time audio software that has portions of it where procesing must take place in
the absolute minimum amount of time possible, with as little access to other objects or function calls as possible. In these instances, I sometimes have to intentionally do what would otherwise be poor coding practices, because I need to get in, compute a large volume of information VERY quickly, and get out. So there *are* occasions where efficiency is *known* to be more
important. It's rare, admittedly, and I only do this in these rare,
isolated circumstances. But it does happen, and unless you know the details of why the OP says efficiency is more important, you cannot state with
accuracy that he's wrong.
-Howard

I'll take you word for it. I phrased it strongly since the OP was so sure
efficiency was important without indicating some experience proving it so;
and fascination with these kinds of "efficiency" issues is so often a
distraction for beginners. I have to trust experts to make the right
decisions, but the question asked was not an expert's question. Do you find
the following amendment more palatable, or still unnecessarily harsh?

You cannot know if efficiency is important unless you have tests of
correct code which indicate in which areas efficiency is important,

OR efficiency interferes with the development of such tests.

--
KCS

[Howard]

"Kevin Saff" <go********@kevin.saff.net> wrote in message
news:Hr********@news.boeing.com...

I'll take you word for it. I phrased it strongly since the OP was so sure
efficiency was important without indicating some experience proving it so;
and fascination with these kinds of "efficiency" issues is so often a
distraction for beginners. I have to trust experts to make the right
decisions, but the question asked was not an expert's question. Do you find the following amendment more palatable, or still unnecessarily harsh?

You cannot know if efficiency is important unless you have tests of
correct code which indicate in which areas efficiency is important,

OR efficiency interferes with the development of such tests.

That's fine.

Or. just add "In general, " before your statement "You cannot know...".
Sorry to be such a nit-picker, but statements of "absolutes" tend to attract
my attention, especially when they contradict my personal experience.
(Similar statements I've noticed have been "Never use global variables",
"Always limit function bodies to one screenful of information", etc... I've
responded to those as well, and sometimes started rather lengthy debates.
:-))

-Howard