Matthias Klose wrote:
Frank Heckenbach writes:
Besides, some changes according to recent bug reports and more cleanups have been made. The two makefiles (Make-lang.in and Makefile.in) have been merged (also for gcc-3 integration) -- I don't know if anyone cares, but perhaps those who build GPC in special ways ...
who should build gpc in special ways?
Apparently quite a few, telling from some messages I've seen here ... ;-)
Find attached a patch to build gpc with current gcc-3_2-branch.
--- src/gcc/p/gpc.h~ 2002-11-06 06:51:56.000000000 +0100 +++ src/gcc/p/gpc.h 2002-12-02 19:13:32.000000000 +0100 @@ -749,7 +749,9 @@ extern tree boolean_true_node; extern tree ptrsize_unsigned_type_node; extern tree ptrsize_integer_type_node; +#if (__GNUC__ <= 3) && (__GNUC_MINOR__ <= 2) && (__GNUC_PATCHLEVEL__ <= 1) extern tree size_type_node; +#endif extern tree packed_array_unsigned_long_type_node; extern tree packed_array_unsigned_short_type_node; extern tree const_string_type_node;
Not quite. __GNUC__ etc. checks the version of the build compiler, but the problem seems to be related to the backend (source) used. (Apart from that, the condition would be wrong, since it would fail for, e.g., gcc-2.95.x where __GNUC_MINOR__ is 95.)
Apparently the problem stems from some difference of the current branch an the gcc-3.2.1 release (I've just tested the latter on a Debian GNU/Linux system, IA32, without any problems).
I suppose they've changed something about sizetype. I'd have to see the changes exactly to tell if it's enough to remove the declaration, or if more changes are needed. (Maybe you can send me the relevant changes, or a diff against 3.2.1 by private mail ...)
It might require another version check in config-lang.in.
Frank