Re: Re[4]: How's your Perl?

[BrowserUk]

I will finally admit :) that whilst I think I know the When and the Where, I haven't a clue as to the Why!

P:\test>perl -MO=Concise  -e"sub f{}; goto +f"
b  <@> leave[t1] vKP/REFC ->(end)
3     <0> enter ->4
4     <;> nextstate(main 2 -e:1) v ->5
a     <1> goto vKS/1 ->b
9        <1> refgen sK/1 ->a  ## <<<HERE.
-           <1> ex-list lKRM ->9
5              <0> pushmark sRM ->6
8              <1> entersub[t2] lKRMS/NO(),TARG,INARGS,1 ->9
-                 <1> ex-list lK ->8
6                    <0> pushmark s ->7
-                    <1> ex-rv2cv sK/129 ->-
7                       <#> gv s ->8
-e syntax OK

Even wading through pp_goto, I couldn't quite work out why it only happens if the function is called in a list context, or indeed why it takes a reference at all. Gut feel tells me its something to do with the fact that the code tries for the goto SUB; variation prior to the goto -LABEL and goto -EXPR variations, and that something is being stacked and not cleaned up, but that's about as far as I got before my brain went into overload.

One good thing came out of it. I now have a much better appreciation of why goto \&func is so gaddamn slow. All that chundering around unwinding stack frames, validating the target isn't a prohibited one etc. I never imagined. I'd always invisioned a simple longjump().

---

Examine what is said, not who speaks.

"Efficiency is intelligent laziness." -David Dunham
"Think for yourself!" - Abigail
Hooray!

[xmath]

ok, here's why:

initially at parse time a goto-expression is free of special cases, so goto &foo is just a goto-element with the ENTERSUB expression &foo as argument. Now, the "goto SUB" variation needs a subref ofcourse, and to accomplish this perl effectively puts a \ in front of the argument, turning goto &foo into goto \&foo. Now here comes the problem: it does this whenever the argument is an ENTERSUB, and forgets to check for the ampersand (!(o->op_flags & OPf_STACKED)), so goto +foo becomes goto \foo, which results in the strange error message (the +-prefix is already dropped at parse time, and is just to prevent the special-case handling for goto LABEL to be triggered)

Questions? :-)