Handle a corner case.

Author: Paul Prescod

snowfakery/standard_plugins/_math.py

@@ -42,10 +42,10 @@ def __init__(self, repeat, iterable):
 
 
 def parts(
-    total: int,
-    min_: int = 1,
-    max_: Optional[int] = None,
-    requested_step: float = 1,
+    user_total: int,
+    user_min: int = 1,
+    user_max: Optional[int] = None,
+    user_step: float = 1,
     rand: Optional[Random] = None,
 ) -> List[Union[int, float]]:
     """Split a number into a randomized set of 'pieces'.
@@ -64,22 +64,24 @@ def parts(
     will be inconsistent with them. e.g. if `total` is not a multiple
     of `step`.
     """
-    max_ = max_ or total
+    max_ = user_max or user_total
     rand = rand or Random()
 
-    if requested_step < 1:
+    if user_step < 1:
         allowed_steps = [0.01, 0.5, 0.1, 0.20, 0.25, 0.50]
         assert (
-            requested_step in allowed_steps
-        ), f"`step` must be one of {', '.join(str(f) for f in allowed_steps)}, not {requested_step}"
+            user_step in allowed_steps
+        ), f"`step` must be one of {', '.join(str(f) for f in allowed_steps)}, not {user_step}"
         # multiply up into the integer range so we don't need to do float math
-        total = int(total / requested_step)
+        total = int(user_total / user_step)
         step = 1
-        min_ = int(min_ / requested_step)
-        max_ = int(max_ / requested_step)
+        min_ = int(user_min / user_step)
+        max_ = int(max_ / user_step)
     else:
-        step = int(requested_step)
-        assert step == requested_step, f"`step` should be an integer, not {step}"
+        step = int(user_step)
+        min_ = user_min
+        total = user_total
+        assert step == user_step, f"`step` should be an integer, not {step}"
 
     pieces = []
 
@@ -88,47 +90,55 @@ def parts(
         smallest = max(min_, step)
         if remaining < smallest:
             # mutates pieces
-            handle_last_bit(pieces, rand, remaining, min_, max_)
+            success = handle_last_bit(pieces, rand, remaining, min_, max_)
+            # our constraints must have been impossible to fulfill
+            assert (
+                success
+            ), f"No way to match all constraints: total: {user_total}, min: {user_min}, max: {user_max}, step: {user_step}"
 
         else:
-            pieces.append(generate_piece(pieces, rand, smallest, remaining, max_, step))
+            pieces.append(generate_piece(rand, smallest, remaining, max_, step))
 
     assert sum(pieces) == total, pieces
     assert 0 not in pieces, pieces
 
-    if requested_step != step:
-        pieces = [round(p * requested_step, 2) for p in pieces]
+    if user_step != step:
+        pieces = [round(p * user_step, 2) for p in pieces]
     return pieces
 
 
 def handle_last_bit(
     pieces: List[int], rand: Random, remaining: int, min_: int, max_: int
-):
+) -> bool:
     """If the piece is big enough, add it.
     Otherwise, try to add it to another piece."""
 
     if remaining > min_:
         pos = rand.randint(0, len(pieces))
         pieces.insert(pos, remaining)
-        return
+        return True
 
     # try to add it to some other piece
     for i, val in enumerate(pieces):
         if val + remaining <= max_:
             pieces[i] += remaining
             remaining = 0
-            return
+            return True
 
-    # just insert it despite it being too small...our
-    # constraints must have been impossible to fulfill
-    if remaining:
-        pos = rand.randint(0, len(pieces))
-        pieces.insert(pos, remaining)
+    # No other piece has enough room...so
+    # split it up among several other pieces
+    for i, val in enumerate(pieces):
+        chunk = min(max_ - pieces[i], remaining)
+        remaining -= chunk
+        pieces[i] = max_
+        assert remaining >= 0
+        if remaining == 0:
+            return True
+
+    return False
 
 
-def generate_piece(
-    pieces: List[int], rand: Random, smallest: int, remaining: int, max_: int, step: int
-):
+def generate_piece(rand: Random, smallest: int, remaining: int, max_: int, step: int):
     part = rand.randint(smallest, min(remaining, max_))
     round_up = part + step - (part % step)
     if round_up <= min(remaining, max_) and rand.randint(0, 1):

tests/tests_math_partition.py tests/test_math_partition.pytests/tests_math_partition.py renamed to tests/test_math_partition.py

@@ -5,10 +5,11 @@
 from snowfakery.data_gen_exceptions import DataGenError
 
 REPS = 1
+SEEDS = [randint(0, 2 ** 32) for r in range(REPS)]
 
 
-@pytest.mark.parametrize("seed", [randint(0, 2 ** 32) for r in range(REPS)])
-class TestSummation:
+@pytest.mark.parametrize("seed", SEEDS)
+class TestMathPartition:
     def test_example(self, generated_rows, seed):
         generate_data(
             "examples/math_partition/math_partition_simple.recipe.yml", seed=seed
@@ -19,8 +20,11 @@ def test_example(self, generated_rows, seed):
             c["Amount__c"] for c in children
         ), (parents, children)
 
-    def test_example_pennies(self, generated_rows, seed):
-        generate_data("examples/math_partition/sum_pennies.recipe.yml", seed=seed)
+    regression_seeds = [824956277]
+
+    @pytest.mark.parametrize("seed2", regression_seeds + SEEDS)
+    def test_example_pennies(self, generated_rows, seed, seed2):
+        generate_data("examples/math_partition/sum_pennies.recipe.yml", seed=seed2)
         objs = generated_rows.table_values("Values")
         assert round(sum(p["Amount"] for p in objs)) == 100, sum(
             p["Amount"] for p in objs
@@ -31,7 +35,7 @@ def test_example_pennies_param(self, generated_rows, seed, step: int):
         generate_data(
             "examples/math_partition/sum_pennies_param.recipe.yml",
             user_options={"step": step},
-            seed=1,
+            seed=seed,
         )
         objs = generated_rows.table_values("Values")
         assert round(sum(p["Amount"] for p in objs)) == 100, sum(
@@ -151,3 +155,21 @@ def test_bad_step(self, generated_rows, seed):
         """
         with pytest.raises(DataGenError, match="step.*0.3"):
             generate_data(StringIO(yaml), seed=seed)
+
+    def test_inconsistent_constraints(self, generated_rows, seed):
+        yaml = """
+    - plugin: snowfakery.standard_plugins.Math
+    - object: Obj
+      for_each:
+        var: child_value
+        value:
+          Math.random_partition:
+            total: 10
+            min: 8
+            max: 8
+            step: 5
+      fields:
+        Amount: ${{child_value}}
+        """
+        with pytest.raises(DataGenError, match="constraints"):
+            generate_data(StringIO(yaml), seed=seed)