Beefy Boxes and Bandwidth Generously Provided by pair Networks
go ahead... be a heretic

Re: SV creation inside XS

by patcat88 (Deacon)
on Nov 07, 2011 at 01:45 UTC ( #936380=note: print w/replies, xml ) Need Help??

in reply to SV creation inside XS

HV* hv = newHV(); RETVAL = newSVrv((SV*)hv, NULL);
As said before, newSVrv is wrong. Make sure you use newRV_noinc since newHV has a ref count of 1 already. If you bring HV to 2 and return the RV to Perl land, the HV is now memory leaked because when the RV is garbage collected, the HV will stay forever. HV and AVs must be returned in RVs to Perl land. If you try to return a HV or AV without it being in an RV, you will get the "bizarre copy" die the next time a perl op sees it. Remember, RETVALed or OUTPUTed SV* get mortalized behind the scenes (look at your .c file), so they should be left by you with refcount 1, and NOT mortalized. As I recall, perl gets very noisy (can't remember if it is a warning or fatal) if your double mortaled SV * or derivative appears. perl will try to delete the SV twice. perl will catch the double mortaled SV because perl will be freeing something already freed and returned by perl to the SV arena/free SV struct cache. Freed SVs have special patterns in them, like refcount in the 2 billions.

Log In?

What's my password?
Create A New User
Node Status?
node history
Node Type: note [id://936380]
[Corion]: Yesterday I encountered an interesting data structure problem. I have a remote program that emits events, and my client listens for these events with one-shot callbacks, that is, I register the callback and if the event gets generated that callback ...
[Lady_Aleena]: robby_dobby, every day. Chaos is my life with few controls.
[Corion]: ... gets called once. The data structure for that is just a hash of arrays, mapping the event type to a queue of registered one-shots, and the first one-shot from the queue gets removed and called.
[Corion]: But now I want to register a one-shot for two events, of which only one will arrive, so my data structure doesn't work anymore...
[Lady_Aleena]: Corion, ouchy.
[Corion]: (maybe I should write this up as a SoPW) - currently, the "most efficient" data structure I come up with is a single array which I scan for the first fitting one-shot. Not efficient but I don't expect more than five outstanding one-shots anyway
[choroba]: can't you create a meta-key corresponding to the disjunction of the events?
[robby_dobby]: Corion: Heh. This whole thing smells of Strategy Pattern or MVC pattern.
[Corion]: And performance linear to the number of registered one-shots doesn't feel that bad. Maybe I should collect statistics on how many callbacks are outstanding ;)
[Corion]: choroba: Yes, but the longer I thought about efficient hashes mapping the event type back to their callbacks, and how to keep them in sync, the more I thought that all that optimization might just not be worth it, even if it's horribly inelegant

How do I use this? | Other CB clients
Other Users?
Others perusing the Monastery: (9)
As of 2017-05-29 07:54 GMT
Find Nodes?
    Voting Booth?