Suggestion for improving C++ parser memory usage

[Robert Dewar]

Gabriel Dos Reis wrote:

> It appeared obvious to me that, if Roger was going to send actual
> patch, he would also provide data points.  So at this point of
> pre-patch, my understanding is that you're merely emitting negative
> feedbacks than actually helping.
> 
> Did you follow gcc-patches, you would have noticed that bunch of
> patches are being applied to GCC (and the C++ front-end in particular)
> in the name of < 1% improvement. 

Indeed, implementing dubious optimizations that decrease stability
without suffiecient demonstrated value is a common hazard of enthusiastic
compiler writers! It's not the 1% above that is worrisome, it is
the less-than sign :-)

I know the "any little bit helps" argument, but mostly I find it
unconvincing. Real improvements in performance are seldom achieved
by small incremental fiddling. What is needed is the much more
difficult work of instrumenting real programs and finding out
what improvements would really help, and that is indeed difficult.

It's always hard to retrofit rapid compilation and small memory
use into a compiler unless it has been designed from the start.

And please don't think I am not in favor of fast compilers (have
a look at Realia COBOL some time :-)