[Kingpython123]

check_Tid() in algorithmic/problems/23/check.cpp called quitf(_ok, "Test 0.") when Tid % 100 == 0, before the submitted program was run. Any output that parses as an A=B program was therefore accepted on such tests, including an empty program and a single rule like a=b. At the commit pinned by Inspect Evals (c7ec558) all three test ids were 0, 100 and 200; on main none of the 16 ids is a multiple of 100, but the shortcut was still there.

The shortcut existed because test set 0 was the empty generator, so running the program on it would have divided by zero. This PR removes the shortcut and gives set 0 real cases: the four fixed samples plus the 100 seeded random cases (the same generators as sets 1 and 2, seeded by Tid / 100). Every test id now runs the program. As a side effect, test ids whose Tid % 100 had no generator (21 to 99) also used to yield an empty set and a score of nan; they now run set 0 as well.

The 16 existing test ids are not affected: for them the generated cases and the scores are identical before and after (for example test 1, id 795913901, scores Ratio: 0.613672131215 with the program in 1.ans on both).

Reproduction (g++ 11.4, -O2 -std=gnu++17, repo testlib; same result with testlib 1e4e8a24):

g++ -O2 -std=gnu++17 -I algorithmic/judge/include -o chk algorithmic/problems/23/check.cpp
echo 0 > tid0.in; : > empty.out; printf 'a=b\n' > onerule.out
./chk tid0.in empty.out algorithmic/problems/23/testdata/1.ans
./chk tid0.in onerule.out algorithmic/problems/23/testdata/1.ans
./chk tid0.in algorithmic/problems/23/testdata/1.ans algorithmic/problems/23/testdata/1.ans

Before (main), for ids 0, 100 and 200:

ok Test 0.                      (exit 0)   empty program
ok Test 0.                      (exit 0)   a=b
ok Test 0.                      (exit 0)   program from 1.ans

After:

wrong answer WA on Test case 1. abcabbabcScabb 1   (exit 1)   empty program
wrong answer WA on Test case 1. abcabbabcScabb 1   (exit 1)   a=b
points 0.615449792 Ratio: 0.615449791957, ...     (exit 7)   program from 1.ans

algorithmic/tests/test_problem23.py is a regression test in the style of the existing checker tests: for ids 0, 100, 200 and 50 the empty and one-rule programs are WA and the program in 1.ans scores, and an existing test id behaves as before. All 13 tests pass with this change; on main 12 of them fail (only the existing-id test passes).

Fixes #205.

[Kingpython123]

Thanks for the quick fix! This is superseded by #210, which landed the fix for #205, so I'm closing this PR.