kevin@kosman.UUCP (Kevin O'Gorman) (05/06/89)
First: I'm told I should send this to some mailing list or other rather than posting. Sorry, but I get no ACK from the mailing list, and it never appears in the newsgroups, so somebody please make sure this gets to wherever else it should go, and let me know what I need to do to do the same in the future. The substance: I just got a fresh gcc 1.35 from oli-stl (which maintains anonymous uucp in California for gnu-philes). Unfortunately, gnulib won't make because it has to go through the stock cpp, which can't handle that many defines. I'm gonna hack it by running it through the gcpp. Could that be made a standard feature of the Makefile? Probably by changing the 'cp' to the proper gcpp invocation... ------ While I'm at it department: an odd thing I have noticed about the stock cc on this machine creates a compatibility problem not covered by the "struct-return" flag. When compiling a function returning a pointer, this cc returns the result in BOTH d0 and a0. When calling a function returning a pointer and dereferencing in the same expression, the compiler uses the result in a0. In all other cases I've tried, the compiler only uses the result in d0. Thus if a routine compiled with the stock cc calls one compiled with gcc there can be a problem. This is a pretty rare case, and I think the stock cc is pretty far off the wall and wastes a lot of bytes over this, but as long as we have compatibility stuff, could the struct-return thing force results into a0 also???