Issues found by sanitizer

1. owner is now initialized to -1, so we need to protect that. Could have been causing a wild qid assignment.

2. when sinks are disabled it is still possible to attempt to enable them. Caused random crashes with unexpected tasks.

added bug label

assigned to @pdraper

added 1 commit

• 00c1fe68 - Need to check if sinks are enabled, not just that the star_formation_sink task...

Compare with previous version

changed the description

added 1 commit

• e80f81d9 - Remove debugging

Compare with previous version

@matthieu no 2, had me blocked for a while. Seems that unionized task pointer (SINK_NONE defined) was being relied upon and touched by this second route (i.e. unskipping hydro related not just sink related). Wonder if that space saving magic has some other surprises.

Mmmmh... That's a cheeky one....

Indeed, it makes me worried now about other possible mishaps in many locations.

assigned to @matthieu and unassigned @pdraper

My tests seem to be running stably once more, so hopefully these are the last issues, so ready for merging.

The sink issue would have been around for quite a while. I guess we haven't run with sanitizer in a long time.

Thanks for fixing these!

merged

mentioned in commit ea9a4fbf

Pretty sure the sink issue was unskipping random tasks, I could see an attempt to run a none task which runner_main aborted on, so that was clearly randomly grabbed from the tasks array, so I'm surprised that it didn't come up before. I guess the big runs may not have that code merged in. May also have required deep cells (depth 10) and recycling, at least the code I ran needed that to happen.

A wrongly unskipped task will likely do nothing of consequence. The runner_xxx() operations will check whether the cell is active anyway. But it could trigger some of the consistency checks indeed.