Tried a 3.5.3 build to rule out source tree changes as a possible cause
of the illegal instruction faults and segfaults. The build with gcc/g++
4.2.4 and the latest unstable binutils (as of yesterday evening) works.
Time to retry 3.5.11 with gcc/g++ 4.4.5 since binutils has changed. If
that fails, retry with the older gcc/g++, and if that fails, increment
through the firefox source releases starting with 3.5.4 until I find the
earliest version that breaks.
On the plus side, we know it's possible to build a working 3.5.3 from
the standard mozilla source. Is there an approved replacement for the
gcjwebplugin functionality I had prior to the hard disk failure? If
not, looks like another thing on my "to do" list.