Subject: Re: Problems using Vortex on x86/linux/gcc-2.96
From: Craig Chambers (chambers@cs.washington.edu)
Date: Wed Sep 26 2001 - 16:40:48 PDT
Erg. I hate how C++ is a moving/underspecified language definition.
In scope-codegen.cecil, line 320 (in my latest version), there's the following
Cecil code:
app.local_header_stream.add_code(
["DBG_E ", debug_struct,
"={\"", cg.debug_file_name(t), "\",",
cg.debug_routine_name(t), ",",
t.bindings_in_order.length.print_string, ",",
dbg_symbols, ",",
t.scope_kind_string, "};\n"]);
You would just need to add "(void*)" in the list before
cg.debug_routine_name(t), i.e.:
app.local_header_stream.add_code(
["DBG_E ", debug_struct,
"={\"", cg.debug_file_name(t), "\",",
"(void*)", cg.debug_routine_name(t), ",",
t.bindings_in_order.length.print_string, ",",
dbg_symbols, ",",
t.scope_kind_string, "};\n"]);
Then you have to compile the Vortex compiler with this change, and then use the
resulting executable. But to compile the Vortex compiler, you first have to
avoid the problem to C-compile the new compiler. One way would be to compile
Vortex with debug_support turned off. Then none of the DBG_E structures (and
lots of other things) would be generated, avoiding the bug. (Also, programs
built w/o debug_support are a lot (~2x?) faster than those built with.) You'll
want to build your new Vortex compiler using profile-guided optimizations, so
make sure that you do "Vortex> load_profile <some path>/vortex.nCCP" before
compiling. See the Performance documentation for information on producing a
fast Vortex-comppiled program.
-- Craig Chambers
Christopher League wrote:
>
> Dear Cecil group,
>
> I'm interested in toying with your Vortex compiler, on an x86-linux
> machine which has gcc-2.96. I'm especially, for now, interested in
> the Java front end.
>
> I've had to tweak the sources here and there, and I've made
> substantial progress getting it to build. But now I'm stuck on a
> particular g++ error message when compiling Java library code
> converted from RTL:
>
> In file included from java.lang.Character.class.rtl.c:2:
> java.lang.Character.class.rtl.local.h:7:
> cannot convert `PairTypeP<int4> (*) (int,
> CecilTaggedPointerClass *)' to `void *' in initialization
>
> The code in question is:
>
> DBG_E E_java_flang_fCharacter_sisLowerCase_aC_rZ =
> { "java.lang.Character.class.rtl",
> &java_flang_fCharacter_sisLowerCase_aC_rZ, /* ERROR IS HERE */
> 1, dbg1, CecilMethodEnv
> };
>
> It seems that g++-2.96 requires an explicit (void*) cast, or else it
> flags an error. I looked, in vain, for a g++ option which would force
> it to ignore the error and move on. This is not an error in C, just
> in C++:
>
> % cat junk.c
> extern int f (int);
> struct s {void* p;} x = { &f };
> % gcc -c junk.c
> % g++ -c junk.c
> junk.c:2: cannot convert `int (*) (int)' to `void *' in initialization
>
> Anyway, perhaps you could point me to the code I would need to tweak
> to convince Vortex to output the (void*) cast in these instances.
> Or perhaps you have a newer, unreleased version which works with later
> version of gcc.
>
> Once I get this working, I'd be glad to send a patch of all the things
> I needed to change. Other than some prototypes, I had to fix the
> _STATIC_SEND and similar macros, removing one occurrence of the ##
> operator after STRIP_PARENS.
>
> Thanks for your time,
> --Chris
> _______________________________________________
> Cecil mailing list
> Cecil@cs.washington.edu
> http://majordomo.cs.washington.edu/mailman/listinfo/cecil
_______________________________________________
Cecil mailing list
Cecil@cs.washington.edu
http://majordomo.cs.washington.edu/mailman/listinfo/cecil
This archive was generated by hypermail 2b25 : Wed Sep 26 2001 - 16:41:06 PDT