CI test failure in tsk_treeseq_init()

[bhaller]

This is a one-off test failure that puzzles me. Here's the output:

Test project D:/a/SLiM/SLiM/Debug
    Start 1: testSLiM
1/3 Test #1: testSLiM .........................***Failed   45.15 sec
_RunMultitraitTests() done
Error:tsk_treeseq_init: A mutation's parent is not consistent with the topology of the tree. Use compute_mutation_parents to set the parents correctly.(TSK_ERR_BAD_MUTATION_PARENT)
[2111] initialize() { initializeTreeSeq(recordMutations=T, simplificationRatio=0.0, checkCoalescence=T, runCrosschecks=T); } initialize() { initializeMutationRate(1e-7); initializeMutationType('m1', 0.5, 'f', 0.0); initializeGenomicElementType('g1', m1, 1.0); initializeGenomicElement(g1, 0, 99999); initializeRecombinationRate(1e-8); } 1 early() { sim.addSubpop('p1', 10); } 100 early() { stop(); } : FAILURE : stop() not reached
   raise message: tsk_treeseq_init: A mutation's parent is not consistent with the topology of the tree. Use compute_mutation_parents to set the parents correctly.(TSK_ERR_BAD_MUTATION_PARENT)

The script in question, prettyprinted:

initialize() {
	initializeTreeSeq(recordMutations=T, simplificationRatio=0.0, checkCoalescence=T, runCrosschecks=T);
}
initialize() {
	initializeMutationRate(1e-7);
	initializeMutationType('m1', 0.5, 'f', 0.0);
	initializeGenomicElementType('g1', m1, 1.0);
	initializeGenomicElement(g1, 0, 99999);
	initializeRecombinationRate(1e-8);
}
1 early() {
	sim.addSubpop('p1', 10);
}
100 early() {
	stop();
}

I re-ran the test and it succeeded (i.e., stop() was reached without an error). This was on Windows, mingw64, x86_64, CLI not SLiMgui.

This was in the multitrait branch. I can think of no reason why this test would fail. So maybe SLiM has some issue with compute_mutation_parents() in rare cases? Might be worth pounding on this a few million times to see if I can catch it in the debugger, get a seed, etc. Anyhow, I wanted a record of it in case it recurs.

[petrelharp]

I hit that error using

$ slim -v
SLiM version 5.1, built Nov 30 2025 08:01:58
Git commit SHA-1: 2f6ce2368a1d352eda3c2c5cc59d6a1d8ef0d394

with seed 3199.

The "not consistent with the topology" is TSK_ERR_BAD_MUTATION_PARENT; this occurs within check_mutation_parents, which is called by tsk_table_collection_check_integrity; this is called a few places in SLiM's code. However, in none of those places is it called with the TSK_CHECK_MUTATION_PARENTS flag that would be required to check the mutation parents, so it must be from one of the times that check_integrity is called by some other tskit function.

Looking at the CHANGELOG a first guess might be that we didn't pass TSK_NO_CHECK_INTEGRITY to tsk_table_collection_compute_mutation_parents; however, that's not right (we do always pass that).

There is only one place in the tskit code that check_integrity is called with TSK_CHECK_MUTATION_PARENTS: in tsk_treeseq_init. (It does this unless the flag TSK_TS_INIT_COMPUTE_MUTATION_PARENTS is passed, so maybe we need to be passing that flag to init?) We call init in four places, so I think it must be one of those places.

Note that the error is triggered by the mutation parents not matching the re-computed mutation parents. So this could happen if simply we hadn't yet computed mutation parents when we do init. Perhaps this is triggered by the combination of arguments to initializeTreeSeq somehow? (eg usually we compute mutation parents while simplifying, but here it's turned off?) And probably we just need to do TSK_TS_INIT_COMPUTE_MUTATION_PARENTS but it'd be nice to verify what's going on.

[bhaller]

Hmm, for me with:

bhaller@glass SLiM % slim -v
SLiM version 5.1, built Sep 12 2025 17:18:04
Git commit SHA-1: unknown (not defined under Xcode)

seed 3199 does not reproduce the bug. Does it reproduce for you consistently with that seed? If so, you should be able to run it under gdb or similar and get a backtrace for the error, I imagine. Did you find that seed by just searching forward from 1, and it took 3199 tries? I imagine I can do that on my end and find a seed that works for me (got a script I can just run to do it?).

The behavior might be platform-dependent, since it might depend upon the particular values present in uninitialized memory or something (maybe in the parents column). I ran it with ASan / UBSan / MallocScribble and that didn't find any problems, though, so maybe unitialized memory is not the issue.

My guess, since end users have never reported this, is that it is related to checkCoalescence=T or runCrosschecks=T, and probably involves the way that we create the tree sequence for that. And yeah, it does seem to be inside a tsk_treeseq_init call; the log quoted above says Error:tsk_treeseq_init: A mutation's parent is not consistent.... The only places we call tsk_treeseq_init directly (that would be hit by this model) are in Species::CheckCoalescenceAfterSimplification() and Species::CrosscheckTreeSeqIntegrity(), so that supports the bug being in one of those two places.

I imagine that playing with the flags to initializeTreeSeq() would illuminate the source of the problem. Try turning coalescence checking and crosschecks off, individually. I like your hypothesis that simplification normally sets up the parents but it has been turned off; that seems very likely. You clearly understand the ins and outs here better than I do, and you've got a reproducible case I guess, so maybe you can track this down? I'm happy to help if I can, though!

[petrelharp]

seed 3199 does not reproduce the bug. Does it reproduce for you consistently with that seed?

Yes, it's consistent for me. And yes, I searched forward from 1 to find it, which I did like so:

for x in $(seq 10000); do slim -s $x test.slim; if [ $? -eq 1 ]; then echo $x; break; fi; done

[petrelharp]

If so, you should be able to run it under gdb or similar and get a backtrace for the error, I imagine.

Running it with gdb gets the error but then it says

tsk_treeseq_init: A mutation's parent is not consistent with the topology of the tree. Use compute_mutation_parents to set the parents correctly.(TSK_ERR_BAD_MUTATION_PARENT)
[Inferior 1 (process 1175084) exited with code 01]
(gdb) bt
No stack.

If I understood gdb better I'd know what was going on, probably? (Perhaps because this is a exit, not a crash?)

[bhaller]

If so, you should be able to run it under gdb or similar and get a backtrace for the error, I imagine.

Running it with gdb gets the error but then it says

tsk_treeseq_init: A mutation's parent is not consistent with the topology of the tree. Use compute_mutation_parents to set the parents correctly.(TSK_ERR_BAD_MUTATION_PARENT)
[Inferior 1 (process 1175084) exited with code 01]
(gdb) bt
No stack.

If I understood gdb better I'd know what was going on, probably? (Perhaps because this is a exit, not a crash?)

Ah yes. You'd need to set a breakpoint on the place where the exit happens, to get the backtrace. No worries. Using your little script (thanks for that, I suck at shell scripting) I'll get it to reproduce on my side, if possible, and get a backtrace so we know where it happens exactly. At that point I'll hand it back to you, and I imagine you can diagnose and fix the bug simply by visual examination of the relevant code; I doubt the debugger will be any further use. I'll have a go at it tomorrow, perhaps. Thanks!

[bhaller]

Well, I ran through all 10,000 iterations of your script and it didn't find a reproducible case for me. So maybe it doesn't reproduce on macOS, who knows why. I'm trying 100,000 now, lol. But this might be an opportunity for you to learn gdb better!

[bhaller]

Well, it's made it to 70,000 without an error now, and I want to work on battery so I'm going to stop it now since running this will run my battery down fast. Doesn't seem to reproduce on my machine. :-<