hmm. strupr and strlwr should be ansi, i think. in any case my gcc builds had no problem. as for gtk problems, i have to apologize about that, i don't seem to have any way of testing gtk currently. (does anyone know if gtk works under cygwin?) also, i have received no response at all from bill perry concerning my complaints about duplicated gtk code. if we could abstract out the duplication, many of these problems would disappear. the problem with passing union-type bugs on to the gcc developers is that they simply result in crashes in random places in the code. often the precise point of the crash changes from build to build even with no code changes and identical build options, and the builds will sometimes randomly work and sometimes not. it would be a real effort to figure out what's compiling wrong. (at one point, however, circa 19.13, i did track down the problem, which was that structure initializations were not happening exactly in parallel, so if you had something like x = foo (x), where x was a union, some fields of the union got assigned before all were being used. i did write to the gcc developers and got an incredibly defensive response -- basically, they quoted chapter and verse from the ansi standard and said "well, we agree that the behavior you are expecting is reasonable but technically it doesn't *absolutely* say in the standard that it has to be done that way, so you shouldn't really expect this to work." in general all my interactions with gcc developers have been bad and most of them have been real assholes.) ------- Original Message ------- On Fri, 20 Dec 2002 11:45:25 +0100 Michael Sperber [Mr. Preprocessor]?wrote: >>>>> "Stephen" == Stephen J Turnbull <stephen(a)xemacs.org> writes: Stephen> FUT xemacs-beta. >>>>> "Ben" == Ben Wing <wing(a)666.com> writes: Ben> i did test this non-kkcc. in fact i tested it in quite a Ben> number of configurations, so i'm surprised to see that it Ben> broke. how did it break? Stephen> Well, for starters the MS-only (?) functions strupr and strlwr were Stephen> left naked in lisp.h, breaking the link. The crash I'm getting is Stephen> Fatal error: assertion failed, file /coda/Software/XEmacs/21.5/src/alloc.c, line 3323, GCLI_lh != 0 Stephen> [...] Stephen> followed by about 150 frames of mark_foo(), details on Stephen> request. Is this symtomatic of broken union-type? This is a Stephen> --use-union-type build. I see this particular assert a fair Stephen> amount myself, and occasionally in backtraces on xemacs-beta. I've fixed this one, along with a whole bunch of typos in the Gtk code. I expect to commit later today, after I've banged on the code for a while. Stephen> It would be nice to have real bug reports (ie, >= C backtrace) on this Stephen> stuff so that (a) we can guess whether we're seeing what you've seen Stephen> and (b) we can pass it on to the GCC core developers who occasionally Stephen> contribute to XEmacs. I also agree that union type is a Right Thing, and it's worth investing some effort to make it work. -- Cheers =8-} Mike Friede, V?lkerverst?ndigung und ?berhaupt blabla