Re: interesting gc result wrt cse

[Richard Henderson <rth at cygnus dot com>]

On Fri, Sep 10, 1999 at 06:18:28PM +0100, Joern Rennecke wrote:
> Note that cse_basic_block calls new_basic_block, which frees the hash
> table and upts stuff from there in a free list.
> All this freeing up of stuff from the hash table should be zapped, and
> instead cse_basic_block should clear the hash table when it finishes.

The table_elt structs that new_basic_block plays with are not
garbage collected, so there's no chance of Bad Things happening
with that memory because of its placement wrt ggc_collect.

Now, the table does contain rtx that is subject to collection,
but as we zap the table every time we enter cse_basic_block, 
we'll not see old data.

Therefore, I believe the change to be safe.


r~