using std::string; string("hello") vs std::string("hello") in header file.

F

Fei Liu

In Accellerated C++, the author recommends that in a header file one should
not declare
using std::string, using std::vector etc instead one should directly specify
the namespace specifier in code.

for example, this is bad practice:

header.h
#include <string>
using std::string;
class a{
string x;
};

instead, one should write:
#include <string>
class a{
std::string x;
};

The reason given by the author is that 'using std::string' actually pollutes
the scope where 'using std::string' is declared.
I find this style is rather inconvenient when I want to experiment with
different implementation of certain classes in large projects. For example,
I may want to try myns::string or john::string, it's actually much easier to
replace 'using std::string' to 'using myns::string' and instantly switch to
another string implementation. This is much better than going through all
the instances of 'std::string variable' and replace them to 'myns::string
variable'.

What are your thoughts on this coding style?
 
P

Phlip

Fei said:
In Accellerated C++, the author recommends that in a header file one
should
not declare
using std::string, using std::vector etc instead one should directly
specify
the namespace specifier in code.

Doesn't it say "at top level"?
The reason given by the author is that 'using std::string' actually
pollutes
the scope where 'using std::string' is declared.
I find this style is rather inconvenient when I want to experiment with
different implementation of certain classes in large projects. For
example,
I may want to try myns::string or john::string, it's actually much easier
to
replace 'using std::string' to 'using myns::string' and instantly switch
to
another string implementation. This is much better than going through all
the instances of 'std::string variable' and replace them to 'myns::string
variable'.

What are your thoughts on this coding style?

In general, write a friggin "using std::string;" wherever you like, in fresh
code where nothing is called "string" except the std:: class.

The guideline is to never put anything in a .h file that could cause trouble
in any .cpp file that includes it. This rule implies the area outside a
class{} or namespace{} block should be reserved for things that absolutely
must go there, and convenient things like typedef or using are more trouble
than this minor convenience.

In your case, you could write this in a header:

namespace zone {
// using john::string;
using myns::string;
}

Now use zone::string everywhere you would have used std::string.

(And please don't let us think you are redefining 'string' for your own
nefarious purposes! Leave the std:: classes alone!!)
 
F

Fei Liu

Phlip said:
Doesn't it say "at top level"?

I'll double check that tomorrow when I get a hold of the book again.
In general, write a friggin "using std::string;" wherever you like, in fresh
code where nothing is called "string" except the std:: class.

The guideline is to never put anything in a .h file that could cause trouble
in any .cpp file that includes it. This rule implies the area outside a
class{} or namespace{} block should be reserved for things that absolutely
must go there, and convenient things like typedef or using are more trouble
than this minor convenience.

In your case, you could write this in a header:

namespace zone {
// using john::string;
using myns::string;
}

Now use zone::string everywhere you would have used std::string.

This is a good technique, thanks!
(And please don't let us think you are redefining 'string' for your own
nefarious purposes! Leave the std:: classes alone!!)

Think what you will, but I can assure you that redefining 'string' for a C++
programmer is like taking the first step for a baby. It has to be done
sooner or later. :)
 
P

Phlip

The guideline is to never put anything in a .h file that could cause
trouble in any .cpp file that includes it.

What kind of trouble? Consider Noah Roberts's recent charming post "Gotta
love it":
DComponent(const DComponent & copy) : DGenericNode(copy) {}

c:\src\PIPE-FLO 32\DComponent.h(12) : error C2065: 'copy' : undeclared
identifier
c:\src\PIPE-FLO 32\DComponent.h(12) : fatal error C1903: unable to
recover from previous error(s); stopping compilation

Of course we will never know exactly where that error came from. (And just
saying "get a less crappy compiler" is less a useful engineering tip than a
cheap shot at M$.)

It resembles the kinds of bugs that VC++ has regarding namespaces,
regardless whether Noah abused his directly. Even if your code is
syntactically correct, avoiding namespace abuse will reduce the odds of
these kinds of bugs, whether they are your fault's or the compiler's.
 
P

Phlip

Fei said:
Think what you will, but I can assure you that redefining 'string' for a
C++
programmer is like taking the first step for a baby. It has to be done
sooner or later. :)

Just so we remain on the same page here, "redefining 'string'" could mean
any one of these:

1. write your own string class as an exercise
2. rebuild std::basic_string<> from scratch
3. build an alternative app-specific XString class

Two of those items are attrocities that shouldn't ever be done for any
reason. Even if you need a string for an application-specific evil character
set, such as an EBCDIC/UTF-7 hybrid or something, you can still provide a
custom type for basic_string<>.
 
F

Fei Liu

Phlip said:
Just so we remain on the same page here, "redefining 'string'" could mean
any one of these:

1. write your own string class as an exercise
2. rebuild std::basic_string<> from scratch
3. build an alternative app-specific XString class

Two of those items are attrocities that shouldn't ever be done for any
reason. Even if you need a string for an application-specific evil
character

There are reasons why a different string implementation would be required.
It all depends. std::string is a general string implementation that fits
some (general) cases very well. But to meet specific requirement, be it
performance, memory management etc, often than not a different domain
specific string implementation is used.

Thanks for your feedback.
 
A

Axter

Fei said:
In Accellerated C++, the author recommends that in a header file one should
not declare
using std::string, using std::vector etc instead one should directly specify
the namespace specifier in code.

for example, this is bad practice:

header.h
#include <string>
using std::string;
class a{
string x;
};

instead, one should write:
#include <string>
class a{
std::string x;
};

The reason given by the author is that 'using std::string' actually pollutes
the scope where 'using std::string' is declared.
I find this style is rather inconvenient when I want to experiment with
different implementation of certain classes in large projects. For example,
I may want to try myns::string or john::string, it's actually much easier to
replace 'using std::string' to 'using myns::string' and instantly switch to
another string implementation. This is much better than going through all
the instances of 'std::string variable' and replace them to 'myns::string
variable'.

What are your thoughts on this coding style?

If you want to experiment, it's much easier if you just create a
typedef, and then change the typedef when you want to change it
throughout your code.
 
P

Phlip

Fei said:
There are reasons why a different string implementation would be required.
It all depends. std::string is a general string implementation that fits
some (general) cases very well. But to meet specific requirement, be it
performance, memory management etc, often than not a different domain
specific string implementation is used.

I challenge all "build don't buy" decisions.

But in the specific case of a string, I have worked in linguistics, CORBA,
ActiveX, object databases, video games, embedded stuff, and chat servers,
and I have _never_ seen a situation that needs an application-specific
string class.

I have, however, seen plenty of custom string classes. They added no value,
and slowed down development. And their authors often held them for ransom.
 
J

Joe Van Dyk

Phlip said:
Fei Liu wrote:




I challenge all "build don't buy" decisions.

But in the specific case of a string, I have worked in linguistics, CORBA,
ActiveX, object databases, video games, embedded stuff, and chat servers,
and I have _never_ seen a situation that needs an application-specific
string class.

I have, however, seen plenty of custom string classes. They added no value,
and slowed down development. And their authors often held them for ransom.

What about a thread-safe string class?

Joe
 
R

richard

Joe said:
What about a thread-safe string class?

If your std::string class does COW then I would hope it does atomic,
lock-free reference-counting which will correctly hide the fact
internal data sharing is occuring; if it does not, then it already is
thread safe. Of course, that doesn't mean you can't do non-thread-safe
things with a string class -- for example, as with any object that is
shared between threads, a global std::string object that might get
written from multiple threads should have a mutex (or similar) to make
it safe. But you wouldn't want to add that overhead to all std::string
objects, would you?
 

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments. After that, you can post your question and our members will help you out.

Ask a Question

Members online

Forum statistics

Threads
473,731
Messages
2,569,432
Members
44,832
Latest member
GlennSmall

Latest Threads

Top