C faults, etc

[Steven Schveighoffer]

On Mon, 20 Jul 2009 20:41:03 -0400, Walter Bright  
<newshound1 at digitalmars.com> wrote:

> bearophile wrote:
>> Then we have to work on the efficiency of its implementation :-) (See  
>> my timings lower in this thread).
>
> I've implemented printf. There is nothing fundamental about its  
> efficiency that would be any better than writefln. If writefln is  
> slower, it's due to a bug, not anything fundamental.

Part of it could be fundamental.

* D deals with UTF, whereas printf does not
* Phobos calls the C runtime, so it's guaranteed never to beat printf,  
since printf is not type-safe, and therefore does less.
* Both Tango and Phobos output functions make use of delegates to output  
data.  In the case of Phobos, it uses a delegate to call the C runtime's  
"delegate" fputc, so it's going to be double the delegate penalty for each  
character.

If you're measuring the performance of millions of writes, then you will  
see even a slight difference.

Bearophile,

You may want to develop a more specialized function to do file output if  
you are writing such large quantities of uniform data, calling the system  
calls directly, it might even beat printf.

-Steve