Re: 'conservative' GC == 'risky' GC

hbaker@netcom.com (Henry G. Baker) writes:
>[Dangers of `conservative' garbage collection.]


There have been discussions here about trouble spots with conservative GC,
check the archives. Hans Boehm and David Chase have lobbied for,
basically, a `-gcsafe' option for C/C++/etc. compilers to ensure that
compilers will avoid the sorts of optimizations that Henry mentioned (they
have an interesting list). Boehm and Chase also describe a set of
workarounds for systems that don't support the `-gc' option.


It's also worth noting that even with GC-safe code generation practices it
is still trivial to produce unsafe C programs -- the application can
obscure a pointer e.g. by XORing it, or can store it temporarily on disk
(outside of the default root set) and so on (but writing ANSI C is a good
start). The implication here is that only a subset of all C programs are
GC-safe and, in particular, you should be careful both about your code and
the code you link with.


Despite the caveats, though, I still don't think that conservative
garbage collection is inherently risky. I would say instead that
current compilers and libraries don't support GC, but this could be
fixed. I do agree with Henry that you shouldn't use conservative GC
until you've either fixed the compiler and libraries or you have
acknowledged that using it with existing systems is a source of
possibly obscure bugs.


This is all described in more detail in:


%A Hans-J. Boehm
%A David Chase
%T A Proposal for Garbage-Collector-Safe C Compilation
%J Journal of C Language Translation
%D December 1992
%P 126-141


It's well worth a read.


;-D on ( Automatic sanitation engineering ) Pardo
--

Return to the comp.compilers page.
Search the comp.compilers archives again.