If I'm understanding your question correctly, the simulator should reject the timestep and choose a smaller one, such that the tolerances are met.

Spice simulators frequently reject timepoints -- they're trying to take the largest step they can, so as to finish the analysis faster, and often take too big a step such that they fail to converge or don't follow the shape of the waveform accurately.

Yes, if the abstol isn't met, the timepoint should be rejected and a smaller step taken.