On 3/23/22 15:17, Dr. Arne Babenhauserheide wrote:
> Marc Nieper-Wißkirchen <xxxxxx@gmail.com> writes:
>> What Scheme does not yet have, though, is an efficient and usable
>> idiom to write generic algorithms à la those in the C++ template
>> library.
> Having waited a lot for C++ compiles, I do not consider these efficient.
> They *are* runtime efficient, but they have a huge compile-time cost.
>

At the risk of talking about C/C++ in a scheme forum, I'm going to
decide this is relevant because it's about the implementation language
for a lot of scheme implementations.  A simple-ish transformation on
your C/C++ code cuts compile times in large projects drastically. 

I finally figured this out just a few months ago.

C++ compile times explode so bad because there's an exponential effect
when headers include other headers. A header for a basic utility can
wind up getting read and parsed literally hundreds, or in pathological
cases, thousands of times.  Even if every time but the first it "doesn't
count" because of #ifdef guards.  You can cut that off by putting #ifdef
guards around the #include statements in all the places the file is
included from, skipping the USELESS task that C++ compilers wind up
spending 99% of their time on in large projects.

In every header file, you'll see something like

#ifndef LIB_H
at the top and

#endif
at the bottom.

But then the file is #include'd from a hundred places, and each of those
places is #included from three others, and so on..... 

In order to prevent the system from having to read and parse the file
again, leave the above just as it is but also go to all the places in
your program where you say

#include "lib.h"

And say instead

#ifndef LIB_H
#include "lib.h"
#endif

Seriously.  Every place.  *Especially* in the other header files.

Bear