Date: Tue, 13 Jun 95 17:34 MDT
From: mat AT ardi DOT com (Mat Hostetter)
To: Raphael DOT Quinet AT eed DOT ericsson DOT se (Raphael Quinet)
Cc: djgpp AT sun DOT soe DOT clarkson DOT edu
Subject: Re: malloc without malice
Newsgroups: comp.os.msdos.djgpp
References: <DA2DKz DOT EM0 AT jade DOT mv DOT net> <2fdc5b86 DOT sandmann AT clio DOT rice DOT edu> <3ri2i8$5s AT erinews DOT ericsson DOT se>
Cc: cottons AT ardi DOT com, ctm AT ardi DOT com

>>>>> "Raphael" == Raphael Quinet <Raphael DOT Quinet AT eed DOT ericsson DOT se> writes:

    Raphael> It's a pity that good software such as Purify is not
    Raphael> freeware, because this provides an invaluable help when
    Raphael> debugging that kind of errors.  I had exactly the same
    Raphael> problem today in a large program and I found the error in
    Raphael> two minutes with Pufiry 3.0.  I know I would have spent
    Raphael> several hours if I didn't have that tool.  Does anybody
    Raphael> know if there is a freeware program (or malloc library)
    Raphael> offering the same features as Purify?

A similar free package exists for Linux; it's called "Checker".
The basic idea is this:

Each byte in the program's address space has associated with it a read
permission bit and a write permission bit.  Checker replaces gas (the
assembler) with a modified version that wraps each memory-referencing
asm instruction with a call to a small stub that verifies that the
memory has the appropriate read, write, or read/write permissions, and
modifies those permissions.  Checker also provides a special
"Checker-aware" malloc library.

We (ARDI) started playing with Checker a few weeks ago to help us
track down bugs in Executor, our commercial Macintosh emulator for
DOS, Linux, and NEXTSTEP.  Checker was a good idea, but it doesn't
seem to have been extensively tested or optimized.  We found many bugs
in Checker and it was so slow as to be unusable (for our particular
needs).

cottons AT ardi DOT com (with a little help from me) has fixed the bugs and
sped up Checker enormously.  We've already found some really sneaky
bugs in Executor, and we expect to find more once we make Executor's
internal memory-management system use Checker facilities aggressively.
Once we are confident our version is stable, we'll send the new
sources to Checker's original author and hopefully he will make them
the latest official release.

Of course, Checker only works for Linux (right now), but many djgpp
developers also have access to Linux systems.  It could in principle
be ported to djgpp, but there are some tricky issues involving segment
overrides and such which don't crop up under Linux.

-Mat