On Mon, Nov 12, 2018 at 08:54:10PM -0500, Theodore Y. Ts'o wrote: > On Tue, Nov 13, 2018 at 12:18:05PM +1100, Dave Chinner wrote: > > I'm not interested in making code fast if distro support engineers > > can't debug problems on user systems easily. Optimising for > > performance over debuggability is a horrible trade off for us to > > make because it means users and distros end up much more reliant on > > single points of expertise for debugging problems. And that means > > the majority of the load of problem triage falls directly on very > > limited resources - the core XFS development team. A little bit of > > thought about how to make code easier to triage and debug goes a > > long, long way.... > > So at least in my experience, if the kernels are compiled with > CONFIG_DEBUG_INFO and/or CONFIG_DEBUG_INFO_REDUCED, > scripts/decode_stracktrace.sh seems to do a very nice job with inlined That doesn't help with kernel profiling and other such things that are based on callgraphs... > functions. Now, ext4 generally only has about 3 or 4 nested inlines, > and so I don't know how it works with 20 or 30 nested inlined > functions, so perhaps this is not applicable for XFS. > > But it perhaps toolchain technology has advanced since the Irix days > such that it's no longer as necessary to force the non-inlining of > functions for easing debugging? Not that I've noticed. Indeed, modern toolchains are moving the opposite direction - have you ever tried to debug a binary with gdb that was compiled with LTO enabled? Or maybe even just tried to profile it with perf? Cheers, Dave. -- Dave Chinner david@xxxxxxxxxxxxx
