Problems in 2.95 with g++ -frepo?

Problems in 2.95 with g++ -frepo?

[Jerry Quinn]

Oops.  Sent this to cygnus.com the first time.

 
> On hpux 10.20 hppa2.0:
> 
> I'm tying the latest gcc-2.95 snapshot on a largish project and having
> problems with template instantiation.  I've built the library with
> -frepo and then try to perform closure on the library templates by
> linking everything together.  However, when the link finishes, there are
> still unsatisfied template symbols.  If I try again a couple of times,
> more symbols go away, but I have a set of symbols that I can't get to
> resolve.
> 
> When I get stuck, if I delete all the objects that are being recompiled
> in collect2 and rebuild, and perform this task a few times I can
> eventually kill all the missing template symbols.
> 
> Shouldn't the complete closure be performed by a single link?
> 
> I'm not sure how to put a bug report together for this since it's
> proprietary code and about 14M of source code in the library.
> 
> One thing I've noticed about the most recent snapshot in particular is
> that it doesn't seem to do more than one recompile/relink pass in the
> link stage (as opposed to 2-3? snapshots ago).
> 
> --
> Jerry Quinn                             Tel: (514) 761-8737
> jquinn@nortelnetworks.com               Fax: (514) 761-8505
> Speech Recognition Research

Problems in 2.95 with g++ -frepo?

[Jerry Quinn]

Oops.  Sent this to cygnus.com the first time.

 
> On hpux 10.20 hppa2.0:
> 
> I'm tying the latest gcc-2.95 snapshot on a largish project and having
> problems with template instantiation.  I've built the library with
> -frepo and then try to perform closure on the library templates by
> linking everything together.  However, when the link finishes, there are
> still unsatisfied template symbols.  If I try again a couple of times,
> more symbols go away, but I have a set of symbols that I can't get to
> resolve.
> 
> When I get stuck, if I delete all the objects that are being recompiled
> in collect2 and rebuild, and perform this task a few times I can
> eventually kill all the missing template symbols.
> 
> Shouldn't the complete closure be performed by a single link?
> 
> I'm not sure how to put a bug report together for this since it's
> proprietary code and about 14M of source code in the library.
> 
> One thing I've noticed about the most recent snapshot in particular is
> that it doesn't seem to do more than one recompile/relink pass in the
> link stage (as opposed to 2-3? snapshots ago).
> 
> --
> Jerry Quinn                             Tel: (514) 761-8737
> jquinn@nortelnetworks.com               Fax: (514) 761-8505
> Speech Recognition Research