Performance of CString::Replace

[dududuil]

hi all,
I have a huge application that does a lot of strings manipulations.
Analyzing its performance, I have noticed that a lot of time is spend in
CString::Replace.

calls like :
str.Replace("\n","\n\t"); or
str.Replace("\r", "\0x00");
take lots of time. (25% of overall performance !!)

How can I improve the performance?

[David Lowndes]

Post by dududuil
I have a huge application that does a lot of strings manipulations.
Analyzing its performance, I have noticed that a lot of time is spend in
CString::Replace.
str.Replace("\n","\n\t"); or
str.Replace("\r", "\0x00");
take lots of time. (25% of overall performance !!)

If you change the second example to:

str.Replace( '\r', '\0x00');

It might result in some improvement.

However the first example needs to insert an extra character, so
there's no easy way to alter the fact that the string will require
reallocation - which is likely to be the main hog.

Dave

[Ulrich Eckhardt]

Post by dududuil
I have a huge application that does a lot of strings manipulations.
Analyzing its performance, I have noticed that a lot of time is spend in
CString::Replace.
str.Replace("\n","\n\t"); or
str.Replace("\r", "\0x00");
take lots of time. (25% of overall performance !!)
How can I improve the performance?

The call to Replace() has to allocate a second buffer, transform the content
accordingly into the new buffer and then release the old one. It's the
allocating and releasing that probably takes the performance, or at least
it's the part that can be avoided mostly.

If you actually have more than one call to Replace() on the same string,
consider rolling your own function that does the replacements. If you have
the two above, I'd do it like this:

CString tmp;
tmp.Reserve(str.GetLength()); // preallocate, not sure if MFC allow this
for(int i=0, len=str.GetLength(); i!=len; ++i)
{
switch(str[i]) {
case '\n': tmp += "\n\t"; break;
case '\r': tmp += "\0x00"; break;
default: tmp += str[i]; break;
}
}
str = tmp;

The preallocation makes the target string overallocate its internal buffer,
so that not every call that adds something causes an allocation.

Uli

[dududuil]

I have failed to find how to preallocate a CString to a certin size - does
anyone know how can I set the size of a CString ?

Post by Ulrich Eckhardt

Post by dududuil
I have a huge application that does a lot of strings manipulations.
Analyzing its performance, I have noticed that a lot of time is spend in
CString::Replace.
str.Replace("\n","\n\t"); or
str.Replace("\r", "\0x00");
take lots of time. (25% of overall performance !!)
How can I improve the performance?

The call to Replace() has to allocate a second buffer, transform the content
accordingly into the new buffer and then release the old one. It's the
allocating and releasing that probably takes the performance, or at least
it's the part that can be avoided mostly.
If you actually have more than one call to Replace() on the same string,
consider rolling your own function that does the replacements. If you have
CString tmp;
tmp.Reserve(str.GetLength()); // preallocate, not sure if MFC allow this
for(int i=0, len=str.GetLength(); i!=len; ++i)
{
switch(str[i]) {
case '\n': tmp += "\n\t"; break;
case '\r': tmp += "\0x00"; break;
default: tmp += str[i]; break;
}
}
str = tmp;
The preallocation makes the target string overallocate its internal buffer,
so that not every call that adds something causes an allocation.
Uli
--
C++ FAQ: http://parashift.com/c++-faq-lite
Sator Laser GmbH
Geschäftsführer: Thorsten Föcking, Amtsgericht Hamburg HR B62 932
.

[Doug Harrison [MVP]]

On Mon, 26 Apr 2010 04:51:01 -0700, dududuil

Post by dududuil
I have failed to find how to preallocate a CString to a certin size - does
anyone know how can I set the size of a CString ?

CSimpleStringT::Preallocate. Microsoft made the CString documentation
pretty inscrutable a couple of versions ago, and you have to look at
CStringT and its base class CSimpleStringT at a minimum.

[Tom Serface]

You can also create a CString using the constructor:

CString(TCHARch,intnRepeat=1);

Where the repeat character is the number of characters you want ultimately.

You might also want to try a regex engine like boost to see if that is
faster:

http://onlamp.com/pub/a/onlamp/2006/04/06/boostregex.html

Tom

Post by dududuil
I have failed to find how to preallocate a CString to a certin size - does
anyone know how can I set the size of a CString ?

Post by Ulrich Eckhardt

Post by dududuil
I have a huge application that does a lot of strings manipulations.
Analyzing its performance, I have noticed that a lot of time is spend in
CString::Replace.
str.Replace("\n","\n\t"); or
str.Replace("\r", "\0x00");
take lots of time. (25% of overall performance !!)
How can I improve the performance?

The call to Replace() has to allocate a second buffer, transform the content
accordingly into the new buffer and then release the old one. It's the
allocating and releasing that probably takes the performance, or at least
it's the part that can be avoided mostly.
If you actually have more than one call to Replace() on the same string,
consider rolling your own function that does the replacements. If you have
CString tmp;
tmp.Reserve(str.GetLength()); // preallocate, not sure if MFC allow this
for(int i=0, len=str.GetLength(); i!=len; ++i)
{
switch(str[i]) {
case '\n': tmp += "\n\t"; break;
case '\r': tmp += "\0x00"; break;
default: tmp += str[i]; break;
}
}
str = tmp;
The preallocation makes the target string overallocate its internal buffer,
so that not every call that adds something causes an allocation.
Uli
--
C++ FAQ: http://parashift.com/c++-faq-lite
Sator Laser GmbH
Geschäftsführer: Thorsten Föcking, Amtsgericht Hamburg HR B62 932
.