Dennis den Brok wrote:
>>>Yes, 1.99 fixes pango for me also.  I don't understandy why 1.100 
>>>doesn't work, do you?  There is just one error statement removed, 
>>>which warns of a missing symbols.
>>
>>It seems to be crashes in glib2. glib2 uses return value of 
>>dlerror() as error status. If dlerror() returns NULL, it is 
>>assumed that there is no error even if dlsym() returned NULL. 
>>Therefore undefined symbol is treated as defined symbol.
>>
>>However, dlerror() must return the most recent error. If 
>>symbol is undefined, dlsym() must set error before it returns 
>>NULL, isn't it? Is revision 1.100 of 
>>src/libexec/ld.elf_so/rtld.c broken?
> 
> 
> While I do also think that dlsym() should behave better in case of =
> error, I
> checked pango with revision 1.99-ld.elf_so on 1.6Y i386 and it did not =
> work.
> Appearantly, there are probably further traps (or I am doing something
> completely wrong).

What steps did you use to go back to 1.99?  I did it by downloading
rtld.c 1.99 from the cvsweb repository and rebuilding the entire
system from scratch and it DID work.  But I only did it that way
because I tried re-compiling just the code for ld.elf and got a
bunch of compiler complaints about missing headers and such.

Is there a simple and reliable way to build a small section of the
system manually?  I'll be happy to try it again if you can tell me how.

This morning I just saved the binary /libexec/ld.elf_so from the build
I did yesterday (the 1.99 sources) and copied it over this morning's
binary and that also worked perfectly.  Before I did that I tried to
run pango-querypackages and confirmed that it still segfaults with
version 1.100 and still runs fine with yesterday's 1.99 binary.