Hello,
This is a fallout of a patch of mines:
http://gcc.gnu.org/ml/gcc-patches/2004-01/msg00822.html
My fix to tsubst <INTEGER_TYPE> for array domains was assuming that subsequent
code (compute_array_index_type) was able to properly check for value
dependness. This was untrue in some cases, for instance while trying to order
between two different partial specializations. The following patch cures this,
in exactly the same way it's already cured in more_specialized for function
templates.
Tested on i686-pc-linux-gnu, OK for 3.4/mainline?