BasicMathTestSuite: Refactor from its current "one fact per function" format into the granular "one fact per scenario" #329

Activity

1. YYDan

   self-assigned this

   on Jul 27, 2026
2. YYDan

   added

   refactorAn improvement to existing test/module codeAn improvement to existing test/module code

   on Jul 27, 2026
3. YYDan

   added the

   Task

   issue type on Jul 27, 2026
4. YYDan commented on Aug 12, 2026

   

   YYDan

   on Aug 12, 2026

   ContributorAuthor

   More actions

   Progress update — Phases 1–3 complete (+ arctan2_test)

   53 → 124 facts. No test logic changed; every original assert is preserved.

   Phase 1 — verify the pattern (4 → 10)

   Test	Before	After
   math_set_epsilon_test	1	4
   ceil_test	1	2
   floor_test	1	2
   math_get_epsilon_test	1	2

   Phase 2 — trig basics (8 → 29)

   Test	Before	After
   tan_test	1	7
   arccos_test (incl. arccos_input_range_test)	2	5
   arcsin_test (incl. arcsin_input_range_test)	2	5
   arctan_test	1	4
   cos_test	1	4
   sin_test	1	4

   Phase 3 — degree trig (9 → 41)

   Test	Before	After
   darctan2_test	1	12
   dtan_test	1	7
   darccos_test (incl. darccos_input_range_test)	2	5
   darcsin_test (incl. darcsin_input_range_test)	2	5
   darctan_test	1	4
   dcos_test	1	4
   dsin_test	1	4

   Additional — arctan2_test (1 → 13)

   arctan2_test appears in the per-function table above but wasn't listed in any phase, so it would have been missed. Folded into the Phase 3 pass and split alongside darctan2_test (4 quadrant cases + 9 known-value cases).

   ---

   Conventions applied

   • Split rule: one fact per distinct input value, plus one fact per property/range loop. This overshoots the estimated counts in several places (tan_test 7 vs 4, darctan2_test 12 vs 4) — deliberate, since the point is granular pass/fail rather than hitting the planning estimate.
   • Section header comment per family (// DARCCOS TESTS), facts named <original>_test #N.
   • #macro declarations hoisted to section level — GML macros are global at compile time, so a macro left inside a split fact body becomes a duplicate definition.
   • Setup duplicated into each fact so facts are self-contained; commented-out show_debug_message cruft stripped.
   • Original assert messages preserved verbatim, including their #N prefixes (which no longer always match the new fact number).

   Epsilon handling

   Facts that opened with a defensive math_set_epsilon(0.00001) have that call replicated into each new fact rather than shared. This multiplies the calls, but keeps each fact self-contained and avoids depending on the unmerged epsilon set/restore fix (see #332). Those calls should be cleaned up wholesale when that lands.

   Verification

   Static, per phase, against HEAD:

   • Assert calls: 703 → 703 suite-wide — none lost.
   • Assert content (balanced-paren extraction, normalised, sorted): identical except three deliberate message-text fixes, listed below.
   • Bracket balance (string/comment-aware): curly/paren/square all zero, matching HEAD.
   • All 53 original base names still present; no per-function assert-count mismatch; no duplicate fact names.

   Runtime: all 80 facts passing after Phases 1–2. Log for the current 124 attached.

   AfterPhase3.txt

   Deviations from verbatim (message text only, no logic change)

   1. floor_test — two asserts both labelled #1 Local Negative; second renumbered to #2.
   2. darctan_test #3 — "#3 (darctan(-1) == -45" → stray open paren removed.
   3. arctan2_test #4 — "#4 arctan2(-2, -2) == -(pi - (pi/4)" → missing closing paren added.

   Also noted

   two_pi_Arctan2Test is declared but never referenced anywhere in the repo. Hoisted along with the used macro rather than deleted, as removing dead code is outside this refactor's scope.

   Remaining

   Phases 4–9: conversion & directional, dot products, point functions, logarithmic, aggregation, and the big ones.
5. YYDan commented on Aug 12, 2026

   

   YYDan

   on Aug 12, 2026

   ContributorAuthor

   More actions

   Phases 4–6 complete — conversion & directional, dot products, point functions

   Suite total is now 224 facts, up from 53 at the start of the refactor (124 after Phase 3).

   Phase 4 — Conversion & directional

   Test	Before	After	Estimate	Asserts
   degtorad_test	1	10	3	10
   radtodeg_test	1	10	3	10
   lengthdir_x_test	1	3	3	3
   lengthdir_y_test	1	2	3	2
   angle_difference_test	1	23	4	32
   Total	5	48	16	57

   Running total after Phase 4: 167 facts.

   angle_difference_test is the one place we deliberately did not go one-fact-per-assert. Its 32 asserts sit inside shared buffer_create/buffer_write/buffer_delete blocks, and the four asserts within each block are a single sign-handling matrix over one input pair rather than four independent scenarios. Splitting per assert would have meant duplicating buffer setup ~190 extra lines for no extra diagnostic value, so it's one fact per type (real / int / int64 / mixed pairs). Each fact builds its own minimal buffer, writing and reading only the values it needs, so every value read is identical to before.

   lengthdir_y_test lands under its estimate at 2 — it only ever had 2 asserts.

   Phase 5 — Dot products

   Test	Before	After	Estimate	Asserts
   dot_product_test	1	7	4	7
   dot_product_3d_test	1	12	4	12
   dot_product_normalised_test	1	12	4	12
   dot_product_3d_normalised_test	1	12	4	12
   Total	4	43	16	43

   Running total after Phase 5: 206 facts.

   One fact per assert throughout — each of these was already one distinct input scenario per assert.

   Phase 6 — Point functions

   Test	Before	After	Estimate	Asserts
   point_direction_test	1	7	4	10
   point_distance_3d_test	1	7	4	7
   point_distance_test	1	7	4	7
   Total	3	21	12	24

   Running total after Phase 6: 224 facts.

   Same shared-buffer shape as angle_difference, so the same per-type rule applies: real / int / int64 / three mixed-type pairs / zero-case = 7 facts each. In point_direction_test the direct and negated-target asserts stay paired within their type fact (same buffer, same input pair, opposite sign).

   Verification

   Static verification battery run after each phase, comparing against HEAD:

   Check	Result
   Total assert calls	703 → 703
   Per-test assert counts	no mismatches
   Assert content (normalised, sorted diff)	identical except 3 deliberate typo fixes
   Bracket balance, string/comment-aware	() {} [] all balanced, matching HEAD
   Base names lost / added	none / none
   Duplicate fact names	none

   The three typo fixes are the previously-noted duplicate #1 Local Negative label in floor_test, and two unbalanced parens in assert message text (darctan_test #3, arctan2_test #4). No assertion logic was changed anywhere.
6. YYDan commented on Aug 13, 2026

   

   YYDan

   on Aug 13, 2026

   Last edited by YYDan

   ContributorAuthor

   More actions

   Phases 7–9 complete — suite now at 626 facts

   Three phases since the last update, taking the suite from 224 to 626 facts with no change to assertion logic.

   Phase 7 — Logarithmic (5 → 110)

   Test	Before	After
   exp_test	1	16
   ln_test	1	17
   log10_test	1	19
   log2_test	1	19
   logn_test	1	39

   All 1:1 (one fact per assert). Running total: 329, confirmed passing.

   Side finding: logn_test asserts #34–#36 are behaviourally identical to #19–#21 — they sit under a // negative values heading where every other assert negates the value, so they were clearly meant to be logn(-2, -_vReal) and instead repeat the earlier negative-base cases. Preserved verbatim here per the refactor's no-logic-changes rule; written up separately. Issue #334 added for this.

   Phase 8 — Aggregation (4 → 99)

   Test	Before	After
   max_test	1	24
   mean_test	1	23
   median_test	1	28
   min_test	1	24

   Twelve in-body #macro declarations hoisted to section level (GML macros are global at compile time, so leaving them inside a split fact is a duplicate-definition error). One dead statement dropped from max_test — instance_create_depth(0,0,0,oTest) whose result was never used, and oTest has no event scripts, so no side effects. Running total: 424, confirmed passing.

   Phase 9 — The big ones (9 → 211)

   Test	Before	After
   abs_test	1	16
   clamp_test	1	62
   frac_test	1	16
   lerp_test	1	24
   power_test	1	32
   round_test	1	22
   sign_test	1	16
   sqrt_test	1	7
   sqr_test	1	16

   sqrt_test wasn't on the phase list but was still a single 7-assert fact, so it's folded in — the file now has no unrefactored multi-assert tests. Running total: 626.

   Conventions applied throughout

   • Section header per family (// POWER TESTS), facts named <original> #N, original assert messages preserved verbatim including their #N prefixes and existing typos.
   • Setup duplicated inline into each fact — every fact builds its own minimal buffer, writing and reading only the types it needs, so values are bit-identical to the original shared buffer.
   • Default granularity is one fact per assert. Grouping is kept only where asserts form a single coherent scenario: property/range while loops, and the int64 precondition-plus-observation chains in abs_test.
   • Epsilon: the defensive math_set_epsilon(0.00001) is replicated into each new fact rather than blocking on the epsilon set/restore fix. That adds call sites for the epsilon issue to sweep up later, but keeps facts self-contained and independent of an unmerged change.

   Verification

   Check	Result
   Assert calls, base vs now	703 → 703
   Assert content (normalised, sorted diff)	identical bar 3 message typo fixes from Phases 1–2
   Base test names	53 → 53, none lost or added
   Per-test assert-count mismatches	none
   Duplicate fact names	none
   Bracket balance () {} []	balanced, matching base
   In-body #macro declarations	0 of 21
   Buffer create/delete pairing per fact	0 issues
   Diff scope	BasicMathTestSuite.gml only

   Assert distribution across the 626 facts: 598 single-assert, 24 with 2–4, and 4 outliers — the math_get/set_epsilon facts, which assert on epsilon state and don't split further.

   AfterPhase9.txt
7. YYDan

   added a commit that references this issue on Aug 13, 2026

   Fixes #329 - BasicMathTestSuite split correctly

   ...

   This issue will close once commit 2fd5351 is merged into the 'develop' branch.

   2fd5351
8. YYDan

   linked a pull request that will close this issueFixes #329 - BasicMathTestSuite facts split into a more accurate count #335on Aug 13, 2026
9. DiasFranciscoA

   closed this as completedin b8bc0e2on Aug 13, 2026