win32 cleanup

[bulk88]

I decided to look inside NYTProf on Win32, since Ive used it a couple times, but it has some dire warnings about it using Time::HiRes added in bc9145a , but looking at my binary shows this to be false, perl has had HAS_GETTIMEOFDAY since 2002 Perl/perl5@57ab3df in Win32 which a wrapper around Win32's GetSystemTimeAsFileTime C function. No compensation or adjustments are done to the time other than fixed unit conversions. So NYTProf doesn't use QueryPerformanceCounter directly (I'd like to implement that) or indirectly through Time::HiRes. Can you give some explanation about that comment about NYTProf on Win32?

I also have a PERL_NO_GET_CONTEXT patch almost done.

Also why does https://github.com/timbunce/devel-nytprof/blob/master/NYTProf.xs#L4305

        (void)hv_fetch_ent((HV *)SvRV(sv), caller_subname_sv, 1, 0);

exist?

Also load_sub_callers_callback and pretty much all of NYTProf uses av_fetch a ton. Why not use AvARRAY everywhere? I assume the private av can't be tied.

[timbunce]

I don't remember the details off-hand, but this:

        /* XXX temp hack way to store calling subname */
        sv = *av_fetch(av, NYTP_SCi_CALLING_SUB, 1);
        if (!SvROK(sv))               /* autoviv */
            sv_setsv(sv, newRV_noinc((SV*)newHV()));
        (void)hv_fetch_ent((HV *)SvRV(sv), caller_subname_sv, 1, 0);

looks like it's using hv_fetch_ent to ensure that a key for caller_subname_sv at least exists.
I've added a comment.

I agree that most av_fetch's in hot code paths could be replaced with something faster. Many have lval true though. So perhaps something like this:

#define av_fetch_fast(a, i, l) ((i > AvFILLp(a) || !AvARRAY(a)[k]) ? av_fetch(a, i, l) : &AvARRAY(av)[key])

I'd be delighted if you'd work on that.