Re: [xml] Thanks for excellent software

On Wed, Nov 12, 2003 at 05:03:15PM -0700, Eric S. Eberhard wrote:
Just wanted to share my experience and say thanks for the excellent software.
[... cool story removed ...]

  Hi Eric,

 thanks for sharing this, this is really an interesting story. To me
the key is that the economy that can be done by sharing code, development,
debugging and documentation effort is the reason that I think free
software will prevail in the long run for all the agreed-upon layers
found in the software stack. The fact that people coming from commercial
entities start adopting this model for those layers is significant.
Independantly of the advocacy or political message that usually follow
the use of free software, the development model is really showing its
strength (again for the layers where semantic and APIs are agreed upon
or based on widely accepted standards), the economy of scale directly
generate an economy of work from an individual perspective and 
reduced costs for commercial entities. It's interesting to see the 
developpers and the Pointy Haired Boss agree on a given approach, sure
it's for totally different reasons (as a developper I want my code to
be used and feedback, and the boss just want to minimize costs), but
this works.

  Watching the changes since the beginning of the Linux developpment
has been both interesting and fun ! Sustained growth in the computer
industry seems best achieved by going for the most standard and simplest
solutions, and unless there is a revolution in the way this industry
operates it looks like to be it's only the beginning of the fun, well
we will see :-)

My background is 20+ years 'C' programming experience and almost no xml 
experience.  One of the easiest packages to integrate I have ever found -- 
even for the xml ignorant.  The code is very readable and well structured 
and written, making anything not completely documented in text easy to find 
in the source.

  Well, so you're not affected by the main weakness which is the relatively
poor high-level documentation, not everybody is used to make its way into
foreign C code. That's one of the areas which really need work.

I would suggest a few changes to the "Makefile" -- for some reason mine was 
compiling everything with the "-g" option.  This is a debug option that 
makes the code slow and large.  I changed that to "-O2" for the level 2 
optimization (level 3 did cause some problems but I have seen no code 
including my own that works with level 3).  I also added 

  So I assume you compiled with gcc instead of the default compiler. Still
I'm surprised by the default use of -g, maybe someone with access to AIX
can check this out (Dodji?) I used to have an AM_MAINTAINER_MODE in configure,
maybe this was the cause of this.

"-qarch=com"   This is what is known as "common mode" and if compiled this 
way the result will run on ANY rs6000 hardware architecture and ANY version 
of AIX -- so long as they are equal or later than the machine compiled on 
(e.g. I compiled on a single processor running 4.3.3 and it works on 
multi-processors on 5.x just fine).  This should at least be an option (it 
is a hair slower in theory, I have not found it so in practice).

Depending on your application -qfuncsect may also make sense.  This makes 
the linking at the function level rather than object file level -- takes 
longer, smaller code.

  Hum, I'm unable to say whether these should really be default options,
seems both are reasonable options. Assuming it makes sense to use them
then that could be done directly by modifying in the switch
dnl Workaround for native compilers
dnl  HP  :
dnl  DEC : Enable NaN/Inf
if test "${GCC}" != "yes" ; then
    case "${host}" in

  and provide those for the AIX systems

  Again, thanks a lot for the feedback, it's appreciated !


Daniel Veillard      | Red Hat Network
veillard redhat com  | libxml GNOME XML XSLT toolkit | Rpmfind RPM search engine

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]