Re: Object serialization

Date view Thread view Subject view Author view

From: Alexandre Oliva (oliva@dcc.unicamp.br)
Date: Wed Jan 27 1999 - 12:23:43 EST


On Jan 27, 1999, Godmar Back <gback@cs.utah.edu> wrote:

>>
>> Create an abstract class (or interface, now that they're fast :-)

> They're fast? Just because they're not as superslow as they used to
> doesn't mean I'd call them fast.

> I thought your suggestion of either computing a perfect hash or at least
> have the jit inline the softcall_lookupmethod was a very good one.
> Although it's of course a trade-off since it'll bloat the jitted code
> somewhat.

After we compute a perfect hash, inlining the softcall_lookupmethod
will probably be shorter than pushing its arguments into the stack and
calling it explicitly.

> Btw, did you look at the heap walk I sent?

Yep, but not very carefully (I've got a paper due next Monday, and I
still couldn't start working on it :-(

Some people just don't stop sending me e-mail :-D

> Whenever you see "walking class" following by a conservative scan,
> this means we're scanning jitcode. We really need to find out
> whether that makes any sense whatsoever.

Based on our previous discussions, I think it is safe to refrain from
walking it. If a method refers to a static field (which is the only
reason why we'd want to walk the JIT code), then it also refers to its
class, and the class that contains the static field must appear in the
constant table of the referring class, so the static field won't be
collected while the method is alife.

-- 
Alexandre Oliva  http://www.dcc.unicamp.br/~oliva  aoliva@{acm.org}
oliva@{dcc.unicamp.br,gnu.org,egcs.cygnus.com,samba.org}
Universidade Estadual de Campinas, SP, Brasil


Date view Thread view Subject view Author view

This archive was generated by hypermail 2b29 : Sat Sep 23 2000 - 19:57:51 EDT