> Kevin, I find this hard to believe! I don't use hlfs, but your > point seems to be more general. I've used *each* release of the > kernel since 2.6.12, plus -rc versions, up to 2.6.16 with the 2.6.12 > headers to build desktop systems on one or more of x86, amd64, and > ppc. I try to keep reasonably up to date with packages, and before > X-6.9 I was trying -rc versions, but I've seen no evidence that > newer headers were ever necessary. Specifically, 6.9 (which should > be similar to 7.0, at least in theory) builds ok with either set of > headers (bar occasional breakage in linux-headers). I well understand your point, but the latest Xorg 7.0 (the one with xgl) would not build without Mesa-6.5. <-- I use the modular build which allows using more recent versions of packages than the monoloithic version.
One of the biggest problems I have repeatedly had was with mesa and linux-libc-headers. In the past I had made dirty hacks and other "fixes" that were rather sloppy as I could not specifically isolate the problem. I could cure the symptoms..but not the disease. When I tried to compile Mesa 6.5 against the linux-libc-headers-2.6.12 I discovered that it was trying to use non-existant code in the kernel header's drm.h file. I then looked into my kernel source and found a completely different looking drm.h file that had the missing code. > > Certainly, people have reported problems with inotify and old > headers, but I don't build the affected package. It is indeed true that I have little proof on this matter. The problem with the kernel headers is that parts of it get included by header files of some basic packages like glibc and uclibc. This then causes all header-files that include the relating to indirectly include those linux headers. One thing I have learned from LFS/HLFS and uclibc/glibc is that small problems that can get away unnoticed in basic portions of the system can cause numerous problems that do not directly point to the source. My goal in this is not to always be experimenting and playing around with develpmental projects. I inted to work towards having a stable working machine. Given the nature of computer science, the only way to get this stability is to guaruntee that no problems happen anywhere in the heart of the system first, and then stabalize the rest. Once I guaruntee that the heart of the system is stable and working, I can then guaruntee stability. If I were to use significantly different kernel headers from the kernel in which the system is to run, I cannot guaruntee stability. The mere fact that a single problem like the Mesa issue exists is proof that sticking with linux-headers that are not closely related enough to the running kernel cannot give a guaruntee. It is also most significant to those using an entirely from scratch system: 1) we donot individually have the resources of the major distro's to hack'n'patch all the problems away. 2) I generally believe that people doing scratch systems, hlfs in particular, are using the most recent versions of software. If the most recent version of software is using kernel-space code (directly or indirectly), the stability of programs created from that code cannot be guaruntee'd to be stable. It may be, but history of computer science proves that mistakes happen. Using similar kernel-headers restricts those mistakes to make it much harder to happen. (which just so happens to be the same about security, cannot prevent it, but you can reduce as much as possible, the risks.) 3) you've got the source and the problems. Why not repeat the 2.6.14 system installation using the 2.6.14 kernel headers as the only difference and see if any of those problems have appeared, for yourself. <-- but if your not using hlfs then you may not know. > > As to differences between linux-libc-headers and linux-headers, > yes, there are lots of them. Mostly, they don't really make a > difference to applications because 99% of applications shouldn't be > using them. unfortunately, If none of the linux-headers were used we would definately have no such problems. > > If people on hlfs (or lfs) want to use linux-headers, that will be > great news - we need more testers to find the remaining apps which > we don't cater for. But I can't in all honesty say that using the > new headers is necessary for most users. And of course, if anybody > wants to try them, the instructions are in the clfs book - they are > *not* the same as the instructions for installing > linux-libc-headers! > > Ken I hope you take no offense, and I took none from you. You made a strong point, but I must explain the reasons behind mine. The beautiful thing here is that everyone on this list still has the freedom to take whichever approach they wish. -- http://linuxfromscratch.org/mailman/listinfo/hlfs-dev FAQ: http://www.linuxfromscratch.org/faq/ Unsubscribe: See the above information page
