You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
clear[] after loadData crashes since #182: the propagative layer is freed twice #205
@HugoFara, since #182, ~FireDomain deletes propagativeLayer. After loadData, the DataBroker also registers that layer (registerLayer(domain->getPropagativeLayer()->getKey(), ...), DataBroker.cpp:1397), and ~DataBroker deletes every registered layer, so clear[] frees it twice.
From tests/runff, this segfaults on dev (3eb20d5) and passes on 9e06ade, just before #182:
ASan: heap-use-after-free at FireDomain.cpp:340, freed by ~PropagativeLayer ← ~DataBroker ← ~FireDomain. tests/runff/run.bash fails on rungeojson.ff for the same reason (it passes on 9e06ade). CI runs ff-run.bash, whose two cases (real_case.ff, reload_case.ff) have no clear[], so it stays green.
With addLayer("propagation", ...) (pyforefire) the layer is never registered and ~FireDomain is its only owner, so the delete is needed there. Either loadData should register the layer for lookup without handing it to the broker, or ~DataBroker should skip it.
Activity
HugoFara commentedon Oct 22, 2026
Thanks for signalling @yanissrairi , I opened #206 to fix the issue
fix(core): stop the data broker freeing the domain's propagative layer (
HugoFara commentedon Oct 22, 2026
Merged, thanks @filippi !
Dev to master -> v2.6.0 (#224)