Re: [Grep-devel] proposed new function for dfa

[arnold]

> Yes, a core dump is better if the philosophy is to fail hard rather than
> sweep software errors under the rug, which is the approach used elsewhere
> in dfa.c.  For example, on my platform dfasyntax crashes reliably if
> its first argument is a null pointer.

OK, I'll remove the check.

> >     $1 ~ $2 { .... }
> >
> > The second field on each line is compiled as a regexp, for every line.
> > There's no reason to recompute all the syntax settings every time since
> > the syntax bits don't change over any given invocation of gawk.
>
> Isn't the cost of recomputing syntax settings dwarfed by the cost of
> compiling the regexp?

I don't remember the case, but I did some profiling and in this case
dfasyntax was using a huge percentage of CPU time. I could try to
reproduce it if I really need to convince you. Or else I will just
keep it as a gawk-only change.

Thanks,

Arnold