Garrett D'Amore wrote: > It seems like if you really have code that needs to use the specific > implementation artifacts, you could just supply your own version. We're > talking about one, or maybe two, depending on on style, lines of C, > after all.
How would you know? I mean: if Jörg hadn't discovered this problem and narrowed it down to misbehavior in strcpy, how long would such a thing lay dormant? Especially so since it happens only with particular alignments and lengths. How many users would blame their own configuration files, or a bad application patch, or some other alignment of the stars before blaming something as dead-simple and historically bug-free as string copy? I think we've been given a gift and we seem to be doing all we can to reject it. > The point about Linux and glibc is that *many*, even *most* applications > are not impacted. That you have a particular application that has a > problem with this suggests that you ought to fix your application, > regardless. His application does indeed fall outside the bounds of what's documented to work. And it should probably be fixed. But I don't think that's the end of the story here. The change in strcpy behavior is new, both here and in Linux, and it's *really* obscure. How do we know for certain that there are indeed no problems with any applications -- as opposed to merely not having seen and root-caused the problems yet? What research has been done? > At this point the historical relevance is not interesting. What is > interesting is what the *standards* that we conform to today say, and > what the existing code base is. (And that's where the Linux example is > pertinent -- their set of software applications is vast enough to act as > a significant counterexample to the argument that "many applications > will break". So far its just one. Yours. Fix it. :-) It's somewhat relevant in that [Open]Solaris offers binary compatibility as a major feature of the operating system, while Linux often does not. Having an old binary on Solaris stop working would, I think, be bad news. Old binaries on Linux by definition don't exist because Linux isn't old enough yet. ;-} -- James Carlson 42.703N 71.076W <carls...@workingcode.com> _______________________________________________ opensolaris-code mailing list opensolaris-code@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/opensolaris-code
