Re: [Gcl-devel] Re: ACL2 2.8 Debian packages

[Camm Maguire]

Greetings!

1) The SGC looks very close as you've reported it below.  You just
   need to copy the following into config.h, and perhaps cast the
   argument to memprotect_handler

#ifdef IN_GBC
#undef MPROTECT_ACTION_FLAGS
#define MPROTECT_ACTION_FLAGS SA_RESTART|SA_SIGINFO
#define GET_FAULT_ADDR(sig,code,sv,a) \
 ((siginfo_t *)code)->si_addr
/*  #define GET_FAULT_ADDR(sig,code,sv,a) \ */
/*      ((void *)(*((char ***)(&code)))[44]) */
#endif

2) Had a spare moment and thought I'd try my hand at compiling with
   the mingw32 cross compiler package on debian and running under
   wine.  Problem is that configure builds gclincl.h for my system,
   not for windows.  Could you send me your gclincl.h?

3) If you get a moment and can let me know a) what are the minimum opt
   flags that trigger the ansi universe load bug and b) whether you
   agree that the default should have an 8Mb stack with opt level just
   below this, and (maybe) c) if at some point you could
   confirm/refute that the ignore-errors issue goes away when the
   stack is made large enough to be placed above the text segment,
   that would be great.  In any and all cases, your help is much
   appreciated! 


Take care,

"Mike Thomas" <address@hidden> writes:

> Hi Camm.
> 
> | > I haven't looked into SGC for about 18 months; the state of
> | play at the time
> | > was that I found no way ahead. As there are worse problems on
> | Windows I have
> | > pushed this one down the list.
> | >
> |
> | Agreed as to the priority, but AFAICT, all that is needed is a working
> | mprotect, which apparently is in the mingw libc,
> 
> Yes, but it isn't available in the headers so must be some kind of
> undocumented gcc internal function.  I put a preliminary emulation into CVS
> a long time ago but more work is need to get it to compile:
> 
> gcc -c -g -Wall -DVOL=volatile -fsigned-char -pipe -fno-zero-initialized-in-
> bss -mms-bitfields -g  -mcpu=i386 -march=i386  -I/c/cvs/stable/gcl-2.6.2-c1/
> o -I../h -I../gcl-tk gbc.c
> In file included from gbc.c:1506: sgbc.c: In function `gclmprotect':
> sgbc.c:38: warning: passing arg 4 of `VirtualProtect' from incompatible
> pointer type sgbc.c:40: warning: unsigned int format, pointer arg (arg 3) In
> file included from gbc.c:1506: sgbc.c: In function
> `memprotect_handler_test': sgbc.c:1098: warning: implicit declaration of
> function `GET_FAULT_ADDR' sgbc.c:1098: warning: assignment makes pointer
> from integer without a cast sgbc.c: In function `memprotect_test':
> sgbc.c:1117: error: storage size of `sa' isn't known sgbc.c:1117: error:
> storage size of `sao' isn't known sgbc.c:1117: error: storage size of `saob'
> isn't known sgbc.c:1134: error: `MPROTECT_ACTION_FLAGS' undeclared (first
> use in this function) sgbc.c:1134: error: (Each undeclared identifier is
> reported only once sgbc.c:1134: error: for each function it appears in.)
> sgbc.c:1135: warning: implicit declaration of function `sigaction'
> sgbc.c:1117: warning: unused variable `sa' sgbc.c:1117: warning: unused
> variable `sao' sgbc.c:1117: warning: unused variable `saob' make[1]: ***
> [gbc.o] Error 1
> make[1]: Leaving directory `/c/cvs/stable/gcl-2.6.2-c1/o'
> make: *** [unixport/saved_pre_gcl] Error 2
> 
> | and a proper passing
> | of the fault address into the si_faddr field of the siginfo structure
> | passed to the handler.  At some point you might just try to add
> | #define SGC to config.h, fire up gcl (if it compiles), and then
> | (si::sgc-on t), which should run a test of the address recovery
> | mechanism and report failures if any.
> 
> 
> | Take care,
> 
> And you also,
> 
> Mike Thomas.
> 
> 
> 
> _______________________________________________
> Gcl-devel mailing list
> address@hidden
> http://mail.gnu.org/mailman/listinfo/gcl-devel
> 
> 
> 

-- 
Camm Maguire                                            address@hidden
==========================================================================
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah