NB: This PR leaves unresolved the underlying misbehaviour that leads to _config.transient.hitlogfile closing before the scan completes. That behaviour is potentially a side-effect of insufficient state management of _config, tracked in #646. We might consider undoing this PR when #646 is resolved, or blocking it in order to testing #646. Adding an assert to monitor hitlogfile state may be sufficient when completing #646.
All that said: pending the larger change to _config state, this PR should enable hitlog file writing to at least recover from inconsistent state.
evalutors/base is now more defensive and can recover from closed hitlogfiles
cf. https://github.com/leondz/garak/pull/637#pullrequestreview-2044150946 - testing now passes with a
NIM_API_KEY
setresolves #599
NB: This PR leaves unresolved the underlying misbehaviour that leads to
_config.transient.hitlogfile
closing before the scan completes. That behaviour is potentially a side-effect of insufficient state management of_config
, tracked in #646. We might consider undoing this PR when #646 is resolved, or blocking it in order to testing #646. Adding an assert to monitorhitlogfile
state may be sufficient when completing #646.All that said: pending the larger change to
_config
state, this PR should enable hitlog file writing to at least recover from inconsistent state.