Re: [HACKERS] Its not my fault. Its SEG's FAULT! - Mailing list pgsql-hackers

From Tom Ivar Helbekkmo
Subject Re: [HACKERS] Its not my fault. Its SEG's FAULT!
Date
Msg-id 86emzggl1h.fsf@barsoom.Hamartun.Priv.NO
Whole thread Raw
In response to Its not my fault. Its SEG's FAULT!  (dg@illustra.com (David Gould))
Responses Re: [HACKERS] Its not my fault. Its SEG's FAULT!
List pgsql-hackers
dg@illustra.com (David Gould) writes:

> I would like to discuss the idea of replacing the current scheme of
> explicit memory allocation and and deallocation from separate
> "MemoryDuration" pools with a conservative garbage collector.

Yes!  This is a great idea!  [scrambles, grinning, to finally get to
work on porting the Boehm-Weiser collector properly to NetBSD/* 1.3]

It seems, from recent discussion, reasonable to assume that this will
kill a number of bugs, reduce the memory footprint of the backend and
quite possibly even (judging by the profiling data you quote) give a
welcome performance boost.  Will you be doing some trial runs with
Boehm-Weiser simply linked in as a malloc/free replacement?  Is it a
big project to actually rip out the MemoryDuration allocator's guts to
get rid of some of that overhead?

> I know which one I would rather write! And it is fairly obvious
> which one is more likely to work.

Of course, this one [he said, grinning]:

(define (do-thing)
  (with-handler my-handler
                (do-the-wild-thing)))

> Unsupported assertions to the effect "GC is too slow ... only works
> with lisp ..." etc are ok too, but will be eligible to win valuable
> prizes.

...like a guide to documents on the net debunking these and other
favorite misconceptions about garbage collection?  You're hardly
likely to get too many of those assertions, though: by now, I would
assume that it's gotten through to most programmers that the handling
of memory in a large system can be done more reliably _and_ more
efficiently by a good garbage collector than by a C programmer.  The
fact that the Java designers got this right (no surprise, of course,
with Steele at the core), should by itself have convinced many.

Off-topic: as for Java, we now just have to wait for the byte-code
engine and entire run-time support system to be rewritten in Java, so
that we can get a stable deployment platform for Java on the web that
won't crash the user's browser every other time she loads an applet!

-tih
--
Popularity is the hallmark of mediocrity.  --Niles Crane, "Frasier"

pgsql-hackers by date:

Previous
From: Bruce Momjian
Date:
Subject: Re: [BUGS] General Bug Report: memory leak in postgres backend processes
Next
From: "Vadim B. Mikheev"
Date:
Subject: Re: [HACKERS] Its not my fault. Its SEG's FAULT!