Investigate: Silent failure on large N3 files

[jeswr]

Discovered by @josd

On the extended deep taxonomy benchmark with --turtle we get

$ time eye --quiet --nope test-facts.ttl test-dl.ttl test-rules.n3 --query test-query.n3
@prefix : <http://eulersharp.sourceforge.net/2009/12dtb/test#>.
@prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#>.

:i100000 a :N100000.


real    0m14.959s
user    0m16.278s
sys     0m0.431s

With eye-js this doesn't produce the answer after running 10 times longer:

$ time eyereasoner --quiet --nope test-facts.ttl test-dl.ttl test-rules.n3 --query test-query.n3
@prefix : <http://eulersharp.sourceforge.net/2009/12dtb/test#>.
@prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#>.



real    0m51.283s
user    0m51.695s
sys     0m0.211s

[josd]

The used files are at https://github.com/eyereasoner/eye/tree/master/reasoning/edt

[jeswr]

Agent-assisted triage for @jeswr — could not reproduce on current eyereasoner (21.1.10). Regenerated the EDT inputs (same shape as eye/reasoning/edt) and ran the exact reported command at N=1000/5000/20000/40000 on a deliberately slow 2-core box: correct answer :iN a :NN. every time, exit 0, clean stderr, with reasoning time scaling linearly (19s@20k -> 38s@40k). Extrapolated N=100000 is ~2 min on this constrained machine — i.e., comfortably tractable on normal hardware today. This was reported May 2023 against a much older eye-js/EYE; the failure has almost certainly been fixed upstream since.

Two follow-ups before closing:

1. One confirmation run at the original N=100000 on an unconstrained machine (eye --quiet gentf.n3 / gentd.n3 from eye/reasoning/edt, then the command above).
2. The silent part is still real: lib/bin/main.ts never checks the reasoner outcome or sets process.exitCode, so a resource-error abort would still exit 0 with empty output. Worth a small fix (propagate failure to a non-zero exit code) plus the EDT perf case from chore: add performance tests for extended deep taxonomy benchmark #337 as a regression guard.