daily blurbs
28 Nov 2005 (Mon)
22:45:58 # Life A quick hack to view diffs with docdiff, when you are using subversion. I had a need to preview what's changed in a file, and the file was under subversion version control. The following is a snippet that I came up with. Creates a HTML docdiff for the current buffer in emacs, and invokes mozilla-firefox to view the change. Nice.
(defun svndocdiff-current ()
"Invoke svn docdiff for the current buffer,
and run mozilla-firefox -remote openfile(filename)
to preview the diff"
(interactive)
(let* ((currentname
(buffer-file-name (current-buffer)))
(svnorigname
(concat default-directory
".svn/text-base/"
(file-name-nondirectory
(buffer-file-name (current-buffer)))
".svn-base"))
(diff-command
"/usr/bin/docdiff --html %s %s > %s")
(mozilla-command-line
"mozilla-firefox -remote 'openfile(%s)'")
(moztmpfile-name
(concat default-directory
".svn/.docdifftmp.html")))
(save-buffer)
(shell-command
(format diff-command
svnorigname
currentname
moztmpfile-name)
"*svndocdiff*" "*svndocdiff-err*")
(shell-command
(format mozilla-command-line
moztmpfile-name)
"*svndocdiff*" "*svndocdiff-err*")))
23 Nov 2005 (Wed)
12:58:35 # Life coreutils major change impacting a lot of packages. I seem to be getting more reports on pbuilder failing weirdly. I also noticed cowdancer no longer seems to be working. cowdancer debugging time, again. First, I noticed that cp has changed behavior and cowdancer testsuite was failing to set up a test lab. cp -al 1/ 2/ used to create directory 2/ from 1/, but it now fails complaining that 2/ no longer exists. cp -al 1/ 2 will do what I wanted to do. The main problem was that with 5.3.0 (which was uploaded on 12 Nov), stat --format changed behavior, and that broke cow-shell. According to the NEWS, in new feature(!) section.
With stat, a specified format is no longer automatically newline terminated. If you want a newline at the end of your output, append `\n' to the format string.
Of course, what is worse, older stat doesn't handle \n. and this is going to be a source of compatibility hacks.
$ stat . --format "%d\n" 2049\n
Hey guys, this is meant to be a set of core utilities that people tend to rely on their behavior...
It took me a bit of time to find out, since the solution was to conflict with apt-move, I almost missed it. 339136.
18 Nov 2005 (Fri)
22:51:16 # Life Hello ptrace, on ppc. I've started playing with ptrace on my ibook. It is a bit different from what I have seen on x86 machines. First, the test sample that I have had hanging around in my source tree didn't build. It told me there was no PTRACE_GETREGS symbol. This means that I won't be able to read contents of a register, which sounds like a bad news when capturing system calls. I made a bit of research. Following the header files, by including linux/user.h, asm/user.h asm-ppc/user.h asm/page.hlinux/ptrace.h asm/ptrace.hasm-ppc/ptrace.h are included. PTRACE_GETREGS is not defined anywhere. Looking at ppc64 for comparison, it is different but also did not define PTRACE_GETREGS (PPC_PTRACE_GETREGS is defined, which is a bit of an added mystery). Kernel side source is at arch/ppc/kernel/ptrace.c function sys_ptrace. Looking at that, PTRACE_GETREGS wasn't handled in the switch statement. In arch/ppc64/kernel/ptrace.c, PPC_PTRACE_GETREGS is defined and it seems to be implementing a PPC_PTRACE_GETREGS which looks like functionally what I'm looking for. Looking at that code, it looks really trivial to add that feature, I wonder why it's not there. I don't think ppc32 is overshadowed enough by ppc64 that development could be halted.
13 Nov 2005 (Sun)
09:40:15 # Life Tokyo Area Debian Meeting was held last night. It was the '10th' meeting (12th in non-geek terms) There was discussion about dpkg-statoverride and DWN translation. Main questions and discussion topic follow.
- Even if you configure with dpkg-statoverride, if postinst calls chmod/chown on that file, dpkg-statoverride values will be overridden. However, the next time dpkg-statoverride runs, the value will be 'fixed' and will result in a hard-to-find bug which fixes itself after a while.
- Running statoverride for a nonexisting user will fail with "dpkg-statoverride: non-existing user XXX".
- dpkg-statoverride does not track which package's postinst ran it.
- dpkg-statoverride information is not tracked by the dpkg command, dpkg -c or dpkg -L will not show any trace of dpkg-statoverride information. The behavior is unknown when alternatives and diversions are used simultaneously. The problem isn't known possibly because dpkg-statoverride isn't widely used.
- For logging and user creation/deletion, it has to be handled in postinst with a relatively complex code. Traceability is lacking. However, as a sysadmin, it is a useful feature.
- Isn't it better to have /usr/lib/dpkg/statoverride.d/packagename?
- /var/lib/dpkg/info/diversion and /var/lib/dpkg/info/statoverride file formats are pretty non-extensible. It would be hard to add ACL/SElinux support. The command interfaces look like they should be extended to support such.
- Translating DWN: DWN is a reading material, and won't trash users' system even if there was mistranslation. It's less of a pressure. Reviewers give good comments, and they are helpful.
- For producing text from WML, Imai's doing manual work with emacs auto-fill and making sure [1] etc don't come at beginning of line. It should be possible to automate that.
- A dictionary is required. There are many words which are unclear, and that needs to be fixed. dpkg/apt/aptitude/dselect are using different terms. curses based tools have word length limitations, for example 'dounyu' (2 letters) is sometimes used instead of 'install'(6 letters) for translating 'install'. The words should really be synchronized with synaptic and rpm, but haven't yet gotten around to getting in touch with them. It is strange to say 'unstable, stable, testing' are 'distribution', especially when they are translated to Japanese.
- Since DWN is translated weekly, if dictionary is maintained, it would be good.
- For DWN, RSS generation and having a HTML anchor for each topic would be nice, is it possible?
- Finally had a Japanese-style Enkai (drink party) in 'nono-ji'.
20:56:08 # Life ibookg4 does suspend and resume fine. I didn't really configure anything, but I noticed that when I close the lid on my ibookg4, it suspends; and when I open it up, it resumes. I am using 2.6.14. Amazing. Nice laptop, heh. Update: Phil Kern and Wouter Verhelst commented that suspend/resume doesn't really function when USB devices are plugged in. I didn't have anything plugged in today, that will need to be investigated later...
eth0: suspending, WakeOnLan disabled radeonfb (0000:00:10.0): suspending to state: 3... uninorth-agp: disabling AGP on device 0000:00:10.0 uninorth-agp: disabling AGP on bridge 0000:00:0b.0 radeonfb (0000:00:10.0): resuming from state: 3... PCI: Enabling device 0000:00:10.0 (0000 -> 0003) radeon: PAD_CTLR_STRENGTH doesn't stabilize ! agpgart: Putting AGP V2 device at 0000:00:0b.0 into 4x mode agpgart: Putting AGP V2 device at 0000:00:10.0 into 4x mode PCI: Enabling device 0001:10:1b.0 (0000 -> 0002) PCI: Enabling device 0001:10:1b.1 (0000 -> 0002) PCI: Enabling device 0001:10:1b.2 (0000 -> 0002) ehci_hcd 0001:10:1b.2: park 0 ehci_hcd 0001:10:1b.2: USB 2.0 restarted, EHCI 1.00, driver 10 Dec 2004 eth0: resuming PHY ID: 4061e4, addr: 0 hda: Enabling Ultra DMA 5 hdc: Enabling MultiWord DMA 2 adb: starting probe task... adb devices: [2]: 2 c9 [3]: 3 1 [7]: 7 1f ADB keyboard at 2, handler 1 ADB mouse at 3, handler set to 4 (trackpad) adb: finished probe task... agpgart: Putting AGP V2 device at 0000:00:0b.0 into 4x mode agpgart: Putting AGP V2 device at 0000:00:10.0 into 4x mode [drm] Loading R200 Microcode
9 Nov 2005 (Wed)
18:50:58 # Life oprofiling debuild. referring to Red Hat Magazine article for rpm, I tried doing similar with Debian debuild. Rough information is available, but since Debian does not come with debug infos per default, really meaningful result is not easy to get at. To obtain a vmlinux file, I can invoke make vmlinux from the kernel build directory that I invoked make-kpkg. I am not sure how I can use make-kpkg to obtain a debug kernel. On my amd64 box, csum_partial_copy_generic is taking a large chunk of time. arch/x86_64/lib/csum-copy.S. I tried getting the call graph using -c option, but it didn't give me a meaningful answer, maybe I'm missing something.
CPU: AMD64 processors, speed 2193.7 MHz (estimated)
Counted CPU_CLK_UNHALTED events (Cycles outside of halt state) with a unit mask
of 0x00 (No unit mask) count 860000
CPU_CLK_UNHALT...|
samples| %|
------------------
11587 35.6940 /home/dancer/shared/git/linux-2.6/vmlinux
3971 12.2328 /usr/bin/perl
CPU_CLK_UNHALT...|
samples| %|
------------------
2981 75.0693 /usr/lib/libperl.so.5.8.7
649 16.3435 /lib/libc-2.3.5.so
292 7.3533 /lib/ld-2.3.5.so
43 1.0829 /lib/libpthread-2.3.5.so
2 0.0504 /lib/libdl-2.3.5.so
1 0.0252 /usr/bin/perl
1 0.0252 /lib/libm-2.3.5.so
1 0.0252 /lib/libnss_compat-2.3.5.so
1 0.0252 /usr/lib/perl/5.8.7/auto/POSIX/POSIX.so
3926 12.0941 /usr/lib/gcc/x86_64-linux-gnu/4.0.2/cc1
CPU_CLK_UNHALT...|
samples| %|
(snip)
$ opreport -t 1 -l ~/shared/git/linux-2.6/vmlinux
CPU: AMD64 processors, speed 2193.7 MHz (estimated)
Counted CPU_CLK_UNHALTED events (Cycles outside of halt state) with a unit mask of 0x00 (No unit mask) count 860000
samples % symbol name
2575 22.2232 csum_partial_copy_generic
833 7.1891 __down_interruptible
410 3.5384 timer_interrupt
309 2.6668 do_wp_page
274 2.3647 free_pages_bulk
244 2.1058 get_user_pages
242 2.0885 vsnprintf
234 2.0195 __handle_mm_fault
198 1.7088 vma_adjust
196 1.6916 shrink_zone
182 1.5707 do_mlock
170 1.4672 generic_file_aio_write_nolock
162 1.3981 ioread16_rep
157 1.3550 mlock_fixup
153 1.3204 sort
151 1.3032 sys_mincore
148 1.2773 error_exit
140 1.2083 invalidate_inodes
18:53:35 # Life assembly output for x86_64. I compiled a C program which just gives out 'hello world', and analyzed the gcc assembly output. I suppose subq/addq for rsp is for int parameter; or it could be an ABI requirement. It is surprising that pointer is movl'd to edi (32-bit) for the puts parameter. To set eax=0 (for the return value), xorl (32-bit) is used. I wonder what happens to the higher 32-bits. Maybe it's time to seek for the CPU reference.
.LC0:
.string "Hello world"
.text
.p2align 4,,15
main:
.LFB12:
subq $8, %rsp
.LCFI0:
movl $.LC0, %edi
call puts
xorl %eax, %eax
addq $8, %rsp
ret
21:43:34 # Life Christian Aichinger commented on the previous blog entries about amd64 assembly output, and vmlinux. Apparently, on amd64, int is 32-bits. I've tested with sizeof, and apparently so. And as for vmlinux, it was a hot topic that I've been missing some discussion in debian-devel. Time to delve in.
$ realcsh.c
REAL csh: printf("%i %i %i %i\n",sizeof (short), sizeof (int), sizeof (long), sizeof (long long));
realcshMJmQzq: In function 'main':
realcshMJmQzq:8: warning: format '%i' expects type 'int', but argument 2 has type 'long unsigned int'
realcshMJmQzq:8: warning: format '%i' expects type 'int', but argument 3 has type 'long unsigned int'
realcshMJmQzq:8: warning: format '%i' expects type 'int', but argument 4 has type 'long unsigned int'
realcshMJmQzq:8: warning: format '%i' expects type 'int', but argument 5 has type 'long unsigned int'
2 4 8 8
3 Nov 2005 (Thu)
13:41:11 # Life cvs-commitinfo for cvs on alioth. After searching and googling and testing, I finally figured out how to do CVS log message mails. In commitinfo file, add a line with DEFAULT /usr/bin/cvs-mailcommit --mailto mail-address --from mail-address --diff --full --root %r --dir %p %{sVv}. There are many different ways to write it, but most examples on the web were wrong, since apparently --cvs option no longer works with newer cvs releases.
1 Nov 2005 (Tue)
09:05:07 # Life pbuilder hacking. 336081 is weird, in that something that should be output on the stderr is mixed in the diff.gz output. Weird.
Junichi Uekawa
$Id: 200511.html.en,v 1.13 2005/11/28 14:00:47 dancer Exp $