Stephen McCamant wrote:
> It appears there were two problems with walkoptree and pmreplroot:
> first, there was a logic error in the previous addition of pmreplroot
> support: the code was claiming that the pmreplroot kid of the OP was a
> always PMOP, while in general it isn't: it's only the *parent* we know
> to be a PMOP in this branch. This was causing a segfault on a s///.
>
> The second problem was analogous to the change needed in
> walkoptree_slow: when the OP is an OP_PUSHRE (the regex in a split()),
> the replroot is sometimes reused to store a pointer to the array being
> split into, either as a GV* or as a PADOFFSET under ithreads. In that
> case, walkoptree shouldn't recurse. The second hunk adds a comment to
> op.h mentioning this weirdness. As the comment implies, it would be
> even better documentation to change the code to achieve the reuse with
> a union rather than with casts, but maybe that's more work than it's
> worth.
Thanks, applied as change 23983 to bleadperl (after a long delay, sorry...)