On Wed, Apr 27, 2005 at 08:09:48AM +0200, Leopold Toetsch wrote: > Robin Redeker <[EMAIL PROTECTED]> wrote: [...] > > 1. i wonder how to load bytecode from the memory to parrot when > > embedding it. i've read embed.pod and couldn't find a function that let me > > create a packfile or something i can run, from a memory buffer that holds > > the > > bytecode. > > from src/embed.c:Parrot_readbc() > > pf = PackFile_new(interpreter, is_mapped); > PackFile_unpack(interpreter, pf, (opcode_t *)program_code, program_size); > > albeit it isn't part of the official embedding API (yet).
Ok, thanks, i'll have a look at that. Any idea when there will be a more complete official embedding API? Any hints when looking for functions that 'maybe' go into the API? > > I've read that parrot wants to do a gc run over special marked objects, that > > 'need' timely destruction. (correct me there..) > > You can flag a PMC for timely destruction: > > needs_destroy $P0 > > On subroutine return (or scope exit) you can trigger a lazy DOD run, > which does nothing if there aren't any objects that need timely > destruction: > > sweep 1 # lazy DOD > > This will eventually be replaced by: > > enter_scope 0 | 1 # NEEDS_TIMELY_DESTRUCTION > exit_scope Does this mean, i have to know whether this subroutine handles objects that need timely destruction? > > or some such. (Currently the register frame, where the "sweep 1" is in, > keeps these objects alive, so the lazy DOD run can only be run after the > register frame is discarded) > > > So, what would happen if i have many marked objects and call subroutines > > often? > > We'll try to make the lazy DOD run as fast as possible. Objects that > need timely destruction are counted. If all such objects have been seen > during DOD, the DOD run is aborted. Well, what happens if some objects are not seen at the beginning, but very late in the run? Does that mean, i maybe have to wait 100ms (or longer, don't have any idea how long it may take) or such if i have around 1*10^3 - 1*10^5 objects that need timely destruction? And does that mean, the time spend in GC is very indeterministic? > > > Or are there any other plans? What do other languages do, which > > have refcounting and want to port to parrot? > > You might have a look at ponie (perl5 on parrot). And, while Python > doesn't guarantee timely destruction, people nethertheless rely on it, > because CPython provides timely destruction due to refcounting. Thanks, i'll have a look. > > That means: most of our major target HLLs need timely destruction and > we'll provide it. cu, robin -- [EMAIL PROTECTED] / [EMAIL PROTECTED] Robin Redeker
