The problem is not just for older versions, it's also every time a structure is altered as dmods don't use CTF to locate the offsets.
As an alternative, why don't we just bundle up mdb, dmods and macros from a target system when taking a dump, then we should normally have a debugger that matches the dump. The size of the binaries is insignificant in comparison to the dump size. My preference would be to have savecore grab them and keep them with the dump. Cheers, Greg John Levon wrote: > On Thu, Aug 14, 2008 at 01:38:47AM +0100, John Levon wrote: > > >> http://cr.opensolaris.org/~johnlev/mdb-kb-hvm/ >> > > Related to the mdb code, I'd like to get people's thoughts on something. > > With these changes plus an upcoming S10 putback, we will be able to > debug Solaris 10 HVM domains when they go a bit wonky. > > Generally, when a change that affects mdb is made, usually that change > is made directly in mdb as well, and it's expected to use the correct > version of mdb matching the OS version in the dump you're examining. > Having said that, a bunch of stuff does still work reasonably well. > > The nature of xVM means that we're going to be debugging Solaris 10 > dumps for a long time to come. Should we be looking at making (and > keeping) mdb more robust for older OS versions? Or should we just be > backporting mdb_kb to Solaris 10's MDB, and expect services to use the > correct S10 MDB version? > > regards > john > _______________________________________________ > mdb-discuss mailing list > mdb-discuss at opensolaris.org >
