Is there a null ostream (like /dev/null) in cpp?

[Bo Peng]

Dear list,

I need to add "disable output" feature to a bunch of objects that output
to ostream (or ofstream). The least intrusive way to do this is pass
them ofstream("/dev/null") but this makes my program less portable. Is
there a null ostream built in cpp? Is it easy to implement one?

Thanks.
Bo

[Jonathan Turkanis]

"Bo Peng" <bp***@rice.edu> wrote in message
news:cd**********@joe.rice.edu...

Dear list,

I need to add "disable output" feature to a bunch of objects that output to ostream (or ofstream). The least intrusive way to do this is pass
them ofstream("/dev/null") but this makes my program less portable. Is there a null ostream built in cpp?
No.
Is it easy to implement one?

Yes. Derive a class from std::streambuf and override the protected
virtual function overflow as follows

int overflow(int c) { return c; }

Then you can use a standard istream and set its stream buffer to an
instance of your streambuf class using rdbuf. Or you can define your
own derived ostream class which automatically uses an an instance of
your streambuf class.

BTW, Daryle Walker has submitted a null_stream class for inclusion in
Boost, which will be reviewed soon. (See
http://groups.yahoo.com/group/boost/files/more_io_4.zip. (You have to
sign up for the boost developers list to access this:
http://www.boost.org/more/mailing_lists.htm#main.) Also, I have
written an Iostreams Library
(http://home.comcast.net/~jturkanis/iostreams/) which allows a null
ostream to be defined as follows:

struct null_sink : boost::io::sink {
void write(const char*, std::streamsize) { }
};

typedef boost::io::stream_facade<null_sink> null_ostream;

Jonathan

[Siemel Naran]

"Bo Peng" <bp***@rice.edu> wrote in message
news:cd**********@joe.rice.edu...

I need to add "disable output" feature to a bunch of objects that output
to ostream (or ofstream). The least intrusive way to do this is pass
them ofstream("/dev/null") but this makes my program less portable. Is
there a null ostream built in cpp? Is it easy to implement one?

You can try o.clear(ios::failbit), and then any statement o << 3 will not do
anything.

You can also set the rdbuf() to NULL in case someone calls o.rdbuf() and
writes to the buffer directly. I'm not sure if the standard allows you to
call o.rdbuf(NULL) though.

Disadvantage:

o.clear(ios::failbit);
o << f(3); // program evaluates f(3) even though it doesn't have to

[JKop]

Bo Peng posted:

Dear list,

I need to add "disable output" feature to a bunch of objects that output to ostream (or ofstream). The least intrusive way to do this is pass them ofstream("/dev/null") but this makes my program less portable. Is there a null ostream built in cpp? Is it easy to implement one?
Thanks.
Bo

int cout;
-JKop

[Richard Herring]

In message <SI*****************@news.indigo.ie>, JKop <NU**@NULL.NULL>
writes

Bo Peng posted:

Dear list,

I need to add "disable output" feature to a bunch of

objects that output

to ostream (or ofstream). The least intrusive way to do

this is pass

them ofstream("/dev/null") but this makes my program less

portable. Is

there a null ostream built in cpp? Is it easy to

implement one?

int cout;

Yeah, right, that's really helpful.

He never mentions cout; he wants to pass an ostream to an object.

class some_class {
public:
void output_me(std::ostream &);
};

some_class x;
int cout;
x.output_me(cout); // and this is going to work how, exactly?
--
Richard Herring

[JKop]

It was a joke!

Consider code like this:

using std::cout;

int main()
{
cout << "blah";

cout << "jhskfh";
}
If you stick "int cout" at the start of main(), then it'll
hide the now global scope "cout" object.

Maybe I'm immature, but I find that amusing.
-JKop

[Rolf Magnus]

JKop wrote:

Bo Peng posted:

Dear list,

I need to add "disable output" feature to a bunch of

objects that output

to ostream (or ofstream). The least intrusive way to do

this is pass

them ofstream("/dev/null") but this makes my program less

portable. Is

there a null ostream built in cpp? Is it easy to

implement one?

Thanks.
Bo

int cout;

cout << "Hello world\n";

[JKop]

Rolf Magnus posted:

cout << "Hello world\n";

Allow me to sabotage your prog:
#include <iostream>

using std::cout;
int main()
{
int cout;

cout << "Hello World!!\n";
}

-JKop

[Richard Herring]

In message <43*****************@news.indigo.ie>, JKop <NU**@NULL.NULL>
writes

Rolf Magnus posted:

cout << "Hello world\n";
Allow me to sabotage your prog:
#include <iostream>

using std::cout;
int main()
{
int cout;

cout << "Hello World!!\n";

E2087 Illegal use of pointer.}

--
Richard Herring

[Rolf Magnus]

JKop wrote:

Rolf Magnus posted:

cout << "Hello world\n";

Allow me to sabotage your prog:
#include <iostream>

using std::cout;
int main()
{
int cout;

cout << "Hello World!!\n";
}

g++ says:

fake_cout.cpp: In function `int main()':
fake_cout.cpp:10: error: invalid operands of types `int' and `const
char[15]' to binary `operator<<'

And even if this worked as you might have expected, how would it satisfy
the OP's request for a stream that just throws everything away?

[Rolf Magnus]

JKop wrote:

It was a joke!

Consider code like this:

using std::cout;

int main()
{
cout << "blah";

cout << "jhskfh";
}
If you stick "int cout" at the start of main(), then it'll
hide the now global scope "cout" object.

Maybe I'm immature, but I find that amusing.

Doesn't look immature to me, but I can't find anything funny in that
either.

[Jonathan Turkanis]

"Jonathan Turkanis" <te******@kangaroologic.com> wrote in message
news:2l************@uni-berlin.de...

Yes. Derive a class from std::streambuf and override the protected
virtual function overflow as follows

int overflow(int c) { return c; }

Strictly, this is wrong if c == EOF. For an arbitrary character and
traits type, the implementation would be

int_type overflow(int_type c)
{
return traits_type::not_eof(c);
}

For the case discussed, int overflow(int c) { return 0; }should
suffice.

Jonathan

[Bo Peng]

Siemel Naran wrote:

You can try o.clear(ios::failbit), and then any statement o << 3 will not do
anything.

o.clear(ios::failbit);
o << f(3); // program evaluates f(3) even though it doesn't have to

This seems to be an easy solution but in practice I have to create 'o'
in some way. As far as I know, I can not instantiate ostream like
ostream o;
o.clear(.....)
and if I use
ofstream o(...)
I have to use a valid filename which will create an empty file even if I
do not write anything to it.

Of course it is a bad idea to cripple cout or cerr by
cout.clear(ios::failbit).

I guess I will follow Jonathan's method.

Bo Peng

[Siemel Naran]

"Bo Peng" <bp***@rice.edu> wrote in message
news:cd**********@joe.rice.edu...

Siemel Naran wrote:

You can try o.clear(ios::failbit), and then any statement o << 3 will not do anything.

o.clear(ios::failbit);
o << f(3); // program evaluates f(3) even though it doesn't have to

This seems to be an easy solution but in practice I have to create 'o'
in some way. As far as I know, I can not instantiate ostream like
ostream o;
o.clear(.....)
and if I use
ofstream o(...)
I have to use a valid filename which will create an empty file even if I
do not write anything to it.

If you use an invalid filename, the stream will have the failbit and
possibly the badbit set. Why is it important to instantiate o with the
failbit set, and what's wrong with ostream o followed by
o.clear(ios::failbit)?
Of course it is a bad idea to cripple cout or cerr by
cout.clear(ios::failbit).

I guess I will follow Jonathan's method.

Yes, it is cleaner. But it's also not a good idea to replace the underlying
streambuf of cout, cerr. If you do, remember to restore the original
streambuf, else the program may crash at program termination.