On Fri, Sep 07, 2001 at 12:49:11PM +1000, matthew green wrote:
>    
>    On Fri, Sep 07, 2001 at 02:43:32AM +1000, matthew green wrote:
>    >    
>    >    I've found something interesting which may help tracking this down:
>    > [...]   
>    > is this the only difference?  it shouldn't matter.  the former
>    
>    at least in the output of 
>    objdump --all-headers
>    yes. And the assembly output for all sections except .eh_frame is the same, 
>    too. I've diffed the byte values and I can't see any fundamental 
>    differences except this one.
> 
> 
> hmm, also try --full-contents.  :-)  there must be other changes.

OK, got further: 
The problem seems to be ld. After looking closer at the resulting shared 
library (using objdump -R), I saw that R_SPARC_32 relocations became 
R_SPARC_RELATIVE relocations with a value counted from '*ABS*' (the 
beginning of the file?).
For the R_SPARC_UA32 relocations however the value was the same, but it now 
indicated that it was still a R_SPARC_UA32 relocation starting from the 
'.text' section. I decided to just try to change it to also a 
R_SPARC_RELATIVE relocation starting from *ABS* (a little bit trial and 
error changing byte values) and it suddenly worked. (Also tried just 
changing either, but that wouldn't work.)

I'm trying to find the relevant portion in binutils now ...


Michael