Date: Mon, 19 Dec 2011 16:11:59 -0800 From: Steve Kargl <sgk@troutmask.apl.washington.edu> To: Brooks Davis <brooks@FreeBSD.org> Cc: Doug Barton <dougb@FreeBSD.org>, Garrett Cooper <yanegomi@gmail.com>, Max Khon <fjoe@FreeBSD.org>, svn-src-all@FreeBSD.org, David Chisnall <theraven@FreeBSD.org>, src-committers@FreeBSD.org, G?bor K?vesd?n <gabor@t-hosting.hu>, svn-src-head@FreeBSD.org, Warner Losh <imp@bsdimp.com> Subject: Re: svn commit: r228143 - in head: . share/mk tools/build/options Message-ID: <20111220001159.GA50555@troutmask.apl.washington.edu> In-Reply-To: <20111219230443.GB66858@lor.one-eyed-alien.net> References: <201111291946.pATJkHMs064094@svn.freebsd.org> <4ED544E1.3050307@t-hosting.hu> <4ED545A9.8000304@FreeBSD.org> <CAGH67wSn-kGhtW-qVwwEVvmmu076ncAqMXx=42PnWwb3a4cCbg@mail.gmail.com> <0CAA5754-4FAC-4B87-92B7-439B109473C0@bsdimp.com> <C0239F5F-E290-4AA9-88A4-474E43A1E3D5@FreeBSD.org> <20111219204129.GA34783@troutmask.apl.washington.edu> <20111219230443.GB66858@lor.one-eyed-alien.net>
next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, Dec 19, 2011 at 05:04:43PM -0600, Brooks Davis wrote: > On Mon, Dec 19, 2011 at 12:41:29PM -0800, Steve Kargl wrote: > > On Mon, Dec 19, 2011 at 08:09:32PM +0000, David Chisnall wrote: > > > On 19 Dec 2011, at 19:52, Warner Losh wrote: > > > > > > > -1. The needs of the many? Please. Let's break a useful feature because some people don't understand it and are impatient? That's lame. > > > > > > How useful is gprof-based profiling these days? Now that we > > > have the DTrace pid provider, don't we have access to much more > > > fine-grained profiling information without the need for shipping > > > two versions of every library? > > > > It is quite uesful given that for the last 20 or so years, > > I can do > > > > cc -o z -pg a.c -lm_p > > ./z > > gprof -b -l ./z z.gmon | more > > > > % cumulative self self total > > time seconds seconds calls ms/call ms/call name > > 72.1 0.91 0.91 0 100.00% _mcount [1] > > 11.1 1.05 0.14 8388608 0.00 0.00 sinf [4] > > 8.2 1.16 0.10 8388608 0.00 0.00 nextafterf [5] > > 4.6 1.21 0.06 0 100.00% .mcount (9) > > > > to ge the information I want. > > I'd tend to agree that we should leave it on at least in HEAD. I think it should be the default on all branches. It adds about 28 MB to /usr/lib and 10 minutes to buildworld on my x86_64 systems, ie., it is in the noise. > > dtrace(1M) does not seem to contain an example that gives the > > equivalent information. In fact, the manpage contains no examples, > > only the statement: > > > > See the Solaris Dynamic Tracing Guide for detailed examples > > of how to use the dtrace utility to perform these tasks. > > > > which, of course, is not very useful given that I do not have a > > Solaris Dynamic Tracing Guide. > > http://docs.oracle.com/cd/E19253-01/817-6223/ is it and the first hit in > google... > Which isn't very useful if the system that I'm working on has no or very limit internet access. A quick scan of Ch. 19, 'profile Provider' does not give anything that looks remotely similar to the above 3 command lines. Ch. 33 'User Process Tracing' isn't any better. Telling a users that getting an execution profile for her code can be done by first learning the D programming language and then reading a 41 Chapter document available on the web isn't too user friendly. And, if I read Chapter 26 of the FreeBSD Handbook correctly: 1) dtrace is experimental 2) one needs to build a kernel with the KDTRACE_HOOKS, DDB_CTF, and possibly KDTRACE_FRAME. 3) one needs to 'kldload dtraceall' module, which is a showstopper for anyone who builds his kernel with 'makeoptions NO_MODULES' -- Steve
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20111220001159.GA50555>