Nat!'s Web Journal

Just after the last update the internet router went dead, which just kind of fit into the general mood of how things were going.

Well the linker did "wing" it so I could link and build the packages. Next day on site installing the release I got hit by the problem, that some of our plists are not what the new NSPropertyListSerialization likes. Instead of giving just a warning and crunching through, Apple's programmer opted to raise an exception there. Not reading those plists would've meant a no show - not good!. So I needed to write a plist parser at the customers site to read them in and use NSPropertyListSerialization to write them out "properly" again. Then I found out that writing OpenStep files is not supported by Foundation... only just not quite as fast as I would like as I wrote something like this:

   dictionary = [NSDictionary dictionaryWithObject:@"Value"
                                            forKey:@"Key"];
   error = nil;
   data = [NSPropertyListSerialization dataFromPropertyList:dictionary
                                                     format:NSPropertyListOpenStepFormat
                                           errorDescription:&error];
   NSLog( @"what gives? %@ %@", data, error);

   dictionary = [NSDictionary dictionaryWithObject:@"Value"
                                            forKey:@"Key"];
   error = @"xxx";
   data = [NSPropertyListSerialization dataFromPropertyList:dictionary
                                                     format:NSPropertyListOpenStepFormat
                                           errorDescription:&error];
   NSLog( @"what gives? %@ %@", data, error);

On the client side I chose Kung Log a nice desktop client, to edit, load and upload your entries. It has a nice option of making automatic links to the tune iTunes is currently playing, which happens to be Bastards of a lying breed from the album "The Crusher" by Amon Amarth. All that HTML from "be" to ".All" was generated by Kung Log! Neat :)

Saw X-Men-2, liked it.

There is a program out now which even in the demo version, can do today what ImagePrinter could do then. The name of the program is OmniGraffle 3.0. With version 3.0 you can now print on the full page. The download version is limited to 20 objects, but you only need one to drag an image into a pre-specified area and to get the same functionality as ImagePrinter. I was considering to release some CD case templates with it, but someone already beat me to it, which is good, because it means less work for me. And less work is "was ich schätze".

So please, please, pretty please with sugar on it, ignore ImagePrinter and use Omnigraffle.

Started to write a little demo application for the intro chapter. found out that the gcc option -finline-functions seems to have the same or even less effect than a placebo pill. I could get my static inline function to inline when I turned up -O3 from -Os, but specifying or not specifying -finline-functions is a moot anyway with -O3. Curse Stallmann and the FSF and curse Apple. (Though they have little to do with the compiler, there is no guarantee that they didn't muck with it. And if they are innocent, it still hits the right target, he he) No I am too lazy to verify this with Linux (see picture of the previous entry :))

Unfortunately -O3 turned out to be too aggressive and cleverly figured out that my code was a glorified NOP and threw it all away. Took some time to massage the code into being useful enough for the optimizer.

Looked with interest at this PPC 970 article. I don't know how Altivec centered these testings are, as the programs I usually run don't utilize it. So I don't care all that much about AltiVec. But assuming Bryce is not yet AltiVec optimized, that would mean that a single 1.8 Ghz PPC 970 would be on par with a 3 Ghz Pentium IV. Hmm I certainly would hope so... but the relatively low Ghz count on the PPC 970 makes me wonder, if a better architecture can really bridge that big a gap, We will see.

Currently in iTunes Criminals from the album "Viva Emptiness" by Katatonia. I prefer the previous albums at the moment.

Dann kam das hier:

819 MB erschienen mir zwar etwas übertrieben, aber man hat ja keine Chance zu diskutieren. Aber mal mit Apple-Ctrl-Shift-4 einen Screenshot davon gemacht. Diesen in Preview rein kopiert und versucht mit JPG zu exportieren. Das geht aber nicht, da das Clipboard ein PICT rüberschickt, Preview PICTs aber anscheinend nicht in JPG umwandeln kann (???) merkwürdig. Also erst mal als PNG exportiert, wieder mit Preview geladen und dann als JPG exportiert. Zielgrösse 8K,...16K sinds geworden. Auf mich hört ja keiner...

Marked it again. Did the same thing, leakage again. Hmm... now try it with MallocDebug. MallocDebug starts up and gives me after the same operations 48 bytes of definite leakage.

48 bytes is nothing. The "possible" leakage is 5.4K and remains stable, no matter what I do in the menu. That would indicate that everything is fine. But on the other hand the idle memory footprint is steadily increasing...

Hmm... need to do more reading.

Wrote some more stuff for the article. Happily found some backup of code, that I thought I would never see again, as it was on my stolen iPod.

As it turned out I already had made a much better improvement there and had forgotten all about it. It reduced some code to constant, which today I thought I could optimize from linear to logarithmic. Now I need to redo quite a few drawing again. This also proves that I am getting dumber...

Which leads me to the next bitch and moan, Omnigraffle a feature rich but unfortunately not very accurate drawing program makes my drawing look like this:

The minor problem is the strangely skewed box in the lower left corner. I selected all left boxes and said move to front I took all right boxes and said move to background. Then I said "align to grid" nevertheless something's wrong. The geometry inspector doesn't find any difference between the blue 0x50 and the red 0x70 box, apart from the vertical offset, but I sure can see the difference. I can live with that though, although it's a little ugly.

What I can not live with is the way the text hugs the top barrier. Now that is damn ugly.

To say something positive about Omnigraffle, a real nice feature is the possibility to export parts of your drawings as TIFF via copy/paste. That's how the article is done. I draw in OmniGraffle and copy/paste TIFF into TextEdit. In the end I will do a manual conversion into HTML.

The DP-101-P+ offers a choice of Samba, Appletalk and LPR, it can even do DHCP. And you need the really neat HPIJS for Mac OS X packages. After installing HPIJS, and adding a route to the default IP address 128.192.0.10 with sudo route add -net 192.168.0.0 -interface eth0 I got my old Laserjet 4 as a selectable printer model. I chose it and I could print!

Well except that no paper came out!

The DP-101-P+ has a http interface where you can configure things like IP address and the like. You can look at the printer status and some factory settings, but everything seemed to be alright. On mere accident I telnetted into the printer server. This was an accident, because the extremely lacking manual, doesn't tell you that this option exist and I was just trying to telnet to the lpr port really :). To my surprise I was greeted with an ASCII interface and a whole lot more options. I could f.e. look at the printer queue. There were jobs but nothing was printed. Being a software guy I toggled a few options and tried and tried but to no avail.

Another accident brought the solution. When I touched the printer server, the lights on my printer suddenly lit up. As it turns out the ground of the parallel connector did not fit too snugly with the printers socket. I bend the ground inwards a bit, until the connector sat firmly.

Now everything appears to work great.

Did a little bit more research on the article, but really not much for lack of time And for bandwidth waste: playing now Terra Nola from the album "Have A Nice Trip" by Die Apokalyptischen Reiter
Actually I don't use sudo, as that's for wimps. I use root shells :)

It didn't work.

In the Console I could see that the installer tries to LaunchCFMApp but fails. I looked into the folder but LaunchCFMApp was there. Starting LaunchCFMApp manually gives a bus error. Is my LaunchCFMApp binary corrupt or is it something else, you be the judge:

bash-2.05a$ otool -L /System/Library/Frameworks/Carbon.framework/Versions/A/Support/LaunchCFMApp /System/Library/Frameworks/Carbon.framework/Versions/A/Support/LaunchCFMApp: /System/Library/Frameworks/Carbon.framework/Versions/A/Carbon (compatibility version 2.0.0, current version 122.0.0) /System/Library/Frameworks/QuickTime.framework/Versions/A/QuickTime (compatibility version 1.0.0, current version 172.0.0) /usr/lib/libSystem.B.dylib (compatibility version 1.0.0, current version 63.0.0)

Coincidence ??? :) :) :)

In the end I found MPlayer which could play the WMV file. So my recommendation goes to MPlayer.

Windows Media Player running under Mac OS X, what a smart naming scheme ...

I figured to advance with baby steps into this uncharted territory to write a malloc tracer.

So I want to take the libMallocDebug.a replace the debug_malloc.o file with my code and then link it back together into a dylib.

The first baby step was to simply unar the .a archive into its constituent .o files and then naively try to link them back together into a dylib with:

ld -flat_namespace -dylib -o libTestMalloc.dylib _mallocProtocolUser.o _mallocProtocolServer.o excUser.o excServer.o exceptionCatcher.o debug_malloc.o PortListen.o -F/System/Library/PrivateFrameworks -framework vmutils

as the original dylib appears to be linked like that:

bash-2.05a$ !534
otool -L /usr/lib/libMallocDebug.A.dylib
/usr/lib/libMallocDebug.A.dylib:
        /usr/lib/libMallocDebug.A.dylib (compatibility version 1.0.0, current version 1.0.0)
        /System/Library/PrivateFrameworks/vmutils.framework/Versions/A/vmutils (compatibility version 1.0.0, current version 1.0.0)

unfortunately that didn't work, because the linker complains thusly:

ld: _mallocProtocolUser.o illegal undefined reference for multi module MH_DYLIB output file to symbol: dyld_stub_binding_helper from section (__DATA,__la_symbol_ptr) relocation entry: 0

which already sees me at the end of my wits :( Help!

Ciao
	Nat!

Another day of failure and missed chances. :)

So I just needed to create a dylib with an -init routine that sets the malloc_logger global variable and then start outputting the events to a file. As easy said as done and as I typed the magic incantations:

export DYLD_INSERT_LIBRARIES=/usr/local/libMulleMallocTracer.dylib
export DYLD_FORCE_FLAT_NAMESPACE=1

and then

ls

They were not.

As I searched the /tmp directory in vain for the log file to be generated, I began to understand that the no-show of my debug message on stderr was not just coincidence. I wrote a little test program and linked the library directly. Still nothing happened. Reading through the ld documentation, it became apparent that the -init function is only called, when the shared library is referenced for the first time.

ARGH!

Back to square one. Next up, perusal of the Darwin dyld. I am searchin' for clues.

Playing now: Ghost of the Sun from the album "Viva Emptiness" by Katatonia

I only have an hour tonite, to figure out what to actually do now with the dyld :)

At first I thought that it was merely a linker option to set a different dyld. But not so. So I will just read up on the MachORuntime and then call it a day.

My current theory is, that the dyld path is coming solely from /usr/lib/crt1.o. So I will try to find crt1.o in Darwin somewhere and link that against my dyld and see what happens.

So no more updates the next 10 days or so.