On Wed, 26 Apr 2006, Robert Dewar wrote: > Bernd Trog wrote: > > > I'm chasing a bug that only appeares when Standard.Integer'Size is 16: > > http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26849 > > Trying to make the compiler work with standard integer size of > 16 will be very difficult I fear.
Do you have an example where the compiler (not the runtime) assumes a target Standard.Integer'Size >= 32? Reading the comments in ttypes.ads suggests that there is at least a clear distinction beteen host and target integer types. > It really assumes an integer size of at least 32 I would guess, though > of course type Uint does not depend on the format of Integer. Hmm, it seems that Uint does depend on the target Integer size: package i is subtype I32767 is Integer range -32767 .. 32767; -- Note: -32767 is in the Uint_Direct range! end; compiles OK. While package i is subtype I32768 is Integer range -32768 .. 32767; -- Note: -32768 is NOT in the Uint_Direct range! end; fails here: ada/trans.c:2588: gcc_assert (!TREE_CONSTANT_OVERFLOW (gnu_result)); Both examples compile OK when the target Integer size is 32, of course. > FOr a discussion of the representation of Uint, see the private > part of Uintp. Thanks for the pointer! __________________________________________________ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com
