On Mon, Jun 03, 2013 at 12:15:43AM +0200, Dimitry Andric wrote: > On Jun 2, 2013, at 23:31, Ted Faber <fa...@lunabase.org> wrote: > > On Sun, Jun 02, 2013 at 05:35:27PM +0200, Dimitry Andric wrote: > >> As a better fix, please try dropping the attached file into the > >> /usr/ports/www/firefox/files directory, then rebuild the firefox port > >> from scratch, and reinstall it. > > > > That works as well - firefox runs with compat6x installed, but compliled > > with the patch to www/firefox . > > > > Thanks again for debugging this! > > > Well, the mystery is not entirely solved, as Firefox does not crash with > a similar failure on -current, even with the compat6x libc.so.6 > installed, and without the patch to osfile_unix_allthreads.jsm. > > So *something* is different in the way multiple versions of libc.so are > handled, when they are loaded into the same process. Kostik, maybe you > have any idea?
The same library of different version loaded into the process address space causes all sort of funny effects which are not very useful to investigate in the details. E.g., non-hidden private symbols both function and data types which are also present in the earlier library are resolved to the symbols from earlier library. The split definitely causes complicated and contradictory behaviour. In other words, if not loading libc.so.6 makes the process behave, I consider the case closed.
pgpUwuiVftagM.pgp
Description: PGP signature
