🚀 go-pugleaf

In <AMOSS.93Apr29165653@shuldig.cs.huji.ac.il> amoss@shuldig.cs.huji.ac.il (Amos Shapira) writes:

>bingh@sibelius.cs.byu.edu (Curtis Bingham) writes:

>2. As far as I understand IRIX 5.0 uses SysVr4 shared libs which are much
>   more standard and I think also are supported by GNU.

>I guess the only advice I can give you is "be patient".

Laugh!  I get the impression that SVR4 support is actually on the
"negative" priority list by the GNU people...  (Yes Ron, you've done
a great job!).  A quick perusal of gcc-2.3.3's README.DWARF might change
your mind.

I also understand that there is a pretty fundamental design
"mis-feature" in g++..  I understand that the MSB of function/method
pointers is used as a "virtual" flag.  This is fine, assuming that
the code runs in the bottom half of the memory map.  Guess where
SVR4 puts it's shared libaries?  Yup. In the third gigabyte.  So,
all the g++ code thinks all the methods in the library are virtual.
This aparently breaks things in a pretty big way.

It also means that any change to the "style" of indicating a virtual
function/method would make .o files that are 100% incompatable with
old .o files.

Sigh...

-Peter

PS:  I doubt that I understand 100% correctly how virtual methods
are handled, but I think that is "near enough" for the illustration.
-- 
Peter Wemm   peter@DIALix.oz.au -or- peter@zeus.DIALix.oz.au
NIC handle: PW65    Current maintainer of the "NN" newsreader.