Prelinking

I have now become yet another fan of pre-linking. It works wonders on the startup times of most applications. Many thanks to Jakub Jelinek for creating prelink for Linux.

(Originally posted on Advogato.)

100 Inane Ramblings

I have put in more than 100 inane entries into my Advogato diary by now.
Here is a nice weblog entry (try to ignore the awfully distracting background image) by James Gosling talking about the problems in "correctly" implementing sin() and other trigonometric functions in Java.

Yumpee had pointed me to some nice papers by some of his fellow Googlers a while ago. They make for some great reading. Try to at least read the papers on Google FS, MapReduce and Sawzall (in that order).


Some day, I hope to be able to decipher what Graydon says about programming languages.


(Originally posted on Advogato.)

GCC v/s glibc

I had configured and built GCC 3.4.4 with "--with-arch=pentium3" and was happily using it for everything till I decided to compile glibc 2.3.5 with it. The build failed with a weird assembler error message as noted in glibc bug-701. That bug however, had been been closed unceremoniously and marked a duplicate of bug-333, which itself looks unnecessarily obnoxious.

After a bit of poking around, I found out that GCC pre-defines "__i686" if "-march=pentium3" is used (see $GCC_SRC_DIR/gcc/config/i386/i386.h) and this was causing the problem with glibc (see $GLIBC_SRC_DIR/sysdeps/i386/elf/setjmp.S). So I had to explicitly use "-march=i386" with GCC to be able to compile glibc. However, "make check" still fails. I will now build and install a vanilla GCC 3.4.4 and see if that is able to properly build glibc.


By the way, glibc 2.3.5 can not be built with GCC 4.0.1, though Dan Kegel seems to have patches to enable you to do it. It can not be built with anything lower than GCC 3.2 either.


(Originally posted on Advogato.)

Miscellany

I come back from a two-week vacation in Bhopal and find that GNU Classpath crosses a million lines of source code and that Tom has done the big GNU Classpath merge into libgcj that should make importing changes from GNU Classpath much easier than before. I also found out that people were still discussing the semantics of overflow of signed integers in C on the GCC list and that Zack Weinberg has left GCC development.

Daniel Berlin has set up an automated patch queue for GCC patches. All you have to do now is to include a line like:

:ADDPATCH java:

in your patch message and it would be added to the patch queue for the area "java". Cool!

Nvu, the new incarnation of Mozilla Composer and a standalone program like Firefox and Thunderbird, has reached version 1.0. I took it for a spin and was able to easily create a few nice-looking HTML documents. Apparently it also supports CSS, XHTML, etc. but I don't know much about them yet to find out how good it is at supporting them.


After having ignored all this while the RSS and other feed aggregation capabilities of Mozilla Thunderbird, I now find this feature absolutely indispensible and liberating. Now I can waste even more of my time reading weblogs and sites that I otherwise would not have bothered to visit with such regularity.


Barry Andrews became yet another person to let out a primal cry of joy and pride after he successfully built GCJ 4.0.1 for Win32 by following my document. Based on his suggestions, I have refactored the document a bit and have highlighted some of the more important points.


(Originally posted on Advogato.)

GCC Developers' Summit

The proceedings of the 2005 GCC Developers' Summit are now available (PDF). As usual, they make for very interesting reading. I was particularly interested in reading Vladimir Makarov's paper on the new register allocator infrastructure that he is working on since register allocation, especially for register-starved architectures like x86-32, is GCC's Achille's Heel. It seems that the new Tree-SSA infrastructure particularly aggravates the register pressure.

Dan Kegel also maintained a diary of his visit to the summit.


(Originally posted on Advogato.)

Projects for GCJ

Here are some random projects for GCJ of varying complexity that I would personally like to see implemented as soon as possible:

• 
  Generics - Although Ken Arnold says generics are a mistake, they are still very useful in my personal opinion and GCJ should be able to handle them. Of course, it would mean fixing PR9861 first.
  
• 
  Assembler - a Java assembler is a very useful tool to have, especially when you want to create invalid class files to test bytecode verification. We should either get Jasmin integrated into Classpath Tools, or create one ourselves. The latter option would be eased if we could get ASM integrated into Classpath Tools via Archit Shah's work on a new RMI compiler (which is still waiting to be checked in for some reason).
  
• 
  JIT Interpreter - I think QEMU's portable dynamic translation is a really nifty idea and it seems to work rather well judging by the speed at which QEMU emulates a guest operating system. This can perhaps be used by gij to implement a "real" JIT.
  
  

(Originally posted on Advogato.)