This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [patch] Fix PR26859
On 3/29/06, Roger Sayle <roger@eyesopen.com> wrote:
> Thanks. One aspect that I'm tackling is what the constraints
> on types in polynomial CHRECs should be. A reasonable requirement
> might be that the types of the left and right operands are the
> same as the type of the POLYNOMIAL_CHREC. Indeed, much of the
> code seems to assume this as an invariant.
>
right, when converting the step, we're using this invariant.
Here is a comment from the chrec_convert function:
The following rule is always true: TREE_TYPE (chrec) ==
TREE_TYPE (CHREC_LEFT (chrec)) == TREE_TYPE (CHREC_RIGHT (chrec)).
> However, a simple sanity check such as the following breaks
> bootstrap...
>
Okay, I know where this can come from: basically we don't convert
the step when building the chrecs in the analyzer, but once these
are built, the conversion of sequences from a type to another are safer.
I'll try to include your assert patch and bootstrap with it.
Sebastian