Re: RFA: [PATCH]: MIPS enhancements to libgcj...

[Andrew Haley <aph at redhat dot com>]

David Daney writes:
 > Andrew Haley wrote:
 > 
 > >David Daney writes:
 > > > Rainer Orth wrote:
 > > > 
 > > > >
 > > > >This shouldn't be configure options in my opinion, but the problem should
 > > > >be detected by autoconf if at all possible.  Otherwise, just enable it in
 > > > >configure.{in, host} for the affected platforms.
 > > > >
 > > > >	Rainer
 > > > >  
 > > > >
 > > > The only way I know of to detect the got overflow problem is to build 
 > > > libgcj and see what happens.  I don't think that can reasonably be done 
 > > > with autoconf (chicken and egg problem).
 > > > 
 > > > The only way to test for broken dladdr is likewise to build libgcj, 
 > > > install it on the target and run it.  Again something that cannot be 
 > > > done with autoconf
 > >
 > >Can't you do this in configure.host?  i.e. if this is a MIPS target,
 > >build with -xgot and don't use dladdr().  I don't like adding all
 > >these new configure options, especially for temporary workarounds.
 > >
 > It looks like that is the way everyone thinks that it should be. 
 >  However I think that dladdr may only be broken for me, so disabling it 
 > for everyone may not be the right thing to do.

Oh, I get it.  Well, if it really isn't a known problem anywhere else
then maybe you should just use a local patch or use a make variable:
"make BROKEN_DLADDR=1' or somesuch.  No need to complicate our
configury for a bug that might only have one instance!

 > Likewise it seems possible that N32 and N64 ABI targets may not need -mxgot.

Yeah, but it won't hurt.

 > If you still think that this should go in configure.host, I will update 
 > the patch.

Well, I don't feel terribly strongly either way.

On balance, I reckon the -xgot configure option is OK, but unless we
know there is a ton of MIPS boxen out there with broken dladdr() we
shouldn't add an option for that.

Andrew.