Non-atomic keylenx_has_ksuf check could lead to assertion failure

[huangyihe]

The following two lines of code in scan seems to be giving us issues:

masstree-beta/masstree_scan.hh

Lines 270 to 271 in 6b6c118

	if (n_->keylenx_has_ksuf(keylenx))
	keylen = ka.assign_store_suffix(n_->ksuf(kp));

Basically if the content of n_->keylenx_ changes between the check and the n_->ksuf() call, the precondition check in ksuf() might fail because the precondition check loads the latest value. Is this a bug?

[kohler]

Ugh.

[thawk105]

How about change
"keylen = ka.assign_store_suffix(n_->ksuf(kp));"
into
"keylen = ka.assign_store_suffix((n_->ksuf_ ? n_->ksuf_->get(kp) : n_->iksuf_[0].get(kp)));"
?