Call reset before initializing the task

[pac48]

@henrygerardmoore @rhaschke I have been investigating this issue where calling plan() on the same task more than once can lead to a segfault. The is some internal state the task that is modified during planning, which causes subsequent planning to fail. For example, the states_ field of StagePrivate contained some values from the last run on the subsequent run. My solution is to call the reset() method on the MTC Task object before calling init() and planning. This fixed the segfault we were seeing and also generated valid plans.

Let me know if this is a suitable solution or if there is another recommended way to avoid stale task objects causing failures.

[pac48]

Looks like CI failed due to a segfault caused by this change

moveit::task_constructor::InterfaceState::Priority::operator<(moveit::task_constructor::InterfaceState::Priority const&) const
    Segmentation fault (Address not mapped to object [0xed])

[rhaschke]

In principle, it should be possible to (re)run plan() without resetting beforehand - for example to continue planning when a previous run prematurely stopped, e.g. due to a timeout. If that's not working for you, I would like to see a minimal example reproducing this.

Unfortunately, your PR revealed another issue: calling reset() before plan() should be possible as well - namely, if you want to plan from scratch instead of continuing.

[pac48]

Unfortunately, I do not currently have the bandwidth to investigate this further. Closing for now.