Peter Samuelson wrote: > [Joe Orton] >> OK nice. Why does this require run-time detection rather than simply >> a smarter configure test (or apr_hints.m4 blacklist or whatever)? > > Because we (Debian) ship binaries rather than source code. Our users > run these binaries on both kernel 2.4 and kernel 2.6. Without runtime > detection you get two bad choices. Take your pick: > > - Users still on kernel 2.4 can't run Apache at all - or > - Users on kernel 2.6 can't use Apache to serve 2+GB files
Take this one step further... Users on a broken kernel2.6 can't use sendfile64 until updating apr, after they have picked up the patches to correct the kernel sendfile64 behavior. And one step further... Users who patch-update such that sendfile64 is broken. Or who find sendfile64 broken on a particular network stack driver. The list goes on :) The advantage (and why I *especially* like this patch) is that the user who has -at this moment- a buggy sendfile64 implementation will still be able to take advantage of it once everything is fixed; but I like even more the fact that the user who breaks sendfile64 won't noticeably suffer. My 2c
