Fail stock III solutions that keep only two runs

Merged
PE
7d ago
+41-1

Problem and behavior

Every Spot Capacity Round Trips case gave the same answer to a correct solution and to one that splits the prices into rising runs and adds the two largest run gains (the C++ solution in #691). That solution never joins two runs into one trade across a dip, yet it passed all six cases. After this change it fails the new case best trades span a dip, [1,2,4,2,5,7,2,4,9,0]: the right answer is 13 (buy at 1 and sell at 7, then buy at 2 and sell at 9), and the two largest runs give 12.

Approach

One case appended to the end of the judge in problem-bank/judges.json, so the indexes of the existing cases do not move, and web/judges/spot-capacity-round-trips.json regenerated from it. The label joins the tricky-case list in tests/web/contract.rs, as the triangle cases did in d429ee70, so the case cannot be dropped silently. tests/golden/task-posed.json takes the problem's new judge hash; both the Rust and the Python golden checks compute the same value.

Validation

  • A correct state machine solution passes 7/7 in Python and in C++ (compiled locally with clang), and so does a split-point solution in Python.
  • The two-runs solution fails only the new case, with got 12, in both languages.
  • In the app, the C++ two-runs solution now shows 6/7 with the new case failing:
image
  • ./scripts/test.sh passes every gate except review-memory-tests, which fails the same way on main here because my gh 2.46 lacks the attestation flags, and shell-syntax, where the local shellcheck 0.11.0 warns about scripts/git-post-commit.sh, which this change does not touch. Not checked locally: cargo-audit, cargo-deny, actionlint and the desktop lane, none of which covers a file this change touches.

Closes #691


Summary by cubic

Adds a best-time-to-buy-and-sell-stock-iii case where the best two trades span a dip, so a flawed solution that splits prices into rising runs and keeps the two largest run gains now fails. Previously all six cases gave the same answer to correct and flawed solutions.

  • The new case [1,2,4,2,5,7,2,4,9,0] expects 13; the two largest runs give 12.
  • Appends the case to problem-bank/judges.json and regenerated web/judges/spot-capacity-round-trips.json from it.
  • Updates the golden hash in tests/golden/task-posed.json and adds the case label to the tricky-case list in tests/web/contract.rs so it cannot be dropped silently.