hasSubstuctMatch kernel crash with disconnected SMARTS

[vfscalfani]

Hi,

When trying the following minimal example:

smarts = "C.C"
smiles = ['CC', 'CCC', 'CCCC']

mols = [Chem.MolFromSmiles(smi) for smi in smiles]
pattern = Chem.MolFromSmarts(smarts)

config = SubstructSearchConfig(
    batchSize=1024,
    workerThreads=-1,
    preprocessingThreads=1,
    maxMatches=1,
    uniquify=False,
)

result_test = hasSubstructMatch(
    targets=mols,
    queries=[pattern],
    config=config,
)

print(f"GPU count: {result_test[:, 0].sum()}")

I get a helpful and expected output:

RuntimeError: Fragment queries (disconnected SMARTS patterns) are not supported. Query has 2 disconnected components: C.C

However, if I change the preprocessingThreads to >1 or -1, I get a kernel crash (using Jupyter Notebook), instead of the RunTimeError message.

Thanks,

Vin

[scal444]

Thanks for the report, will roll this fix into v0.5.1 as well.

Also, curious if you have a strong need for disconnected SMARTS support? If so, let us know and we can look into it.

[vfscalfani]

Thanks!

No, I don't have a strong need for disconnected SMARTS support, I had just noticed that these caused issues as I was testing and learning how to use hasSubstructMatch().