[gclist] GC FAQ & uncooperative environments

[Marcin 'Qrczak' Kowalczyk]

Dnia czw 10. lipca 2003 19:54, Fergus Henderson napisał:

> You don't need to resort to "almost legal C" to find pointers;
> if the C code is automatically generated by your compiler, then
> you can use strictly conforming standard C code to find all the
> pointers.

Well, I meant that the runtime in general is almost legal C. Violations of the 
standard are not related to finding pointers.

For example I assume that there exists an integer type with the same alignment 
constraints as a pointer (and thus don't bother to align those integers 
differently than pointers), that structs consisting of pointers and such 
integers can be aligned to pointer size, I don't care whether a pointer is 
read from its representation using the same pointer type every time, I assume 
that structs which contain only pointers have no padding, etc.

I bet this will work on all realistic platforms except those which have 
different pointer representations depending on pointer type (Crays?).

And I will use GMP - perhaps the only *required* external library.

Another non-standard asumption is that many characters are significant
in external names, and that their case is distinguished. I.e. if the C 
implementation has unreasonable restrictions here, programs using long 
identifiers or identifiers which differ only in case may experience name 
clashes.

Other crimes are entirely optional, e.g. I plan to postprocess assembly output 
of a C compiler replacing trampolined tail calls with direct jumps, and use 
two global register variables on gcc.

-- 
   __("<         Marcin Kowalczyk
   \__/       qrczak@knm.org.pl
    ^^     http://qrnik.knm.org.pl/~qrczak/