Increase test coverage

[basil]

The PDF at http://www.eprg.org/G53DOC/pdfs/knuth-plass-breaking.pdf is missing some information that is in the printed version of Digital Typography; namely, the widths of the characters in the examples. This information allows us to write tests to verify that the behavior is exactly as specified in the Knuth & Plass paper, all the way down to the adjustment ratios being identical to the examples given in the paper. The relevant pages from the printed book are enclosed.

CC @robertknight

[robertknight]

This is great, thank-you! I'm going to merge this, although I had a minor query about the last adjustment ratio value for each of the two test cases.

[robertknight]

I'm not certain it is guaranteed that these float literals will be converted to the exact same float values as the result of parsing the truncated number, in all environments. If a problem is encountered, assert.closeTo could be used.

[basil]

I thought about using closeTo, but there wasn't a deepCloseTo like there is a deepEqual so I would have had to iterate over the list. Happy to switch to that approach if you think it is preferable.

[robertknight]

I'm happy to leave this as-is. My concern is only theoretical at the moment.

[robertknight]

I notice the book image shows 0.000 as the final adjustment ratio rather than 0.049. Do you know the reason for that? Does it matter at all?

[basil]

This is because you define MAX_COST to be 1000, but Knuth & Plass's calculations are based on a MAX_COST of 100,000. I chose to be consistent with the rest of the code in this repository, but if you want to be truly faithful to the book I can submit this PR with passing tests:

diff --git a/test/layout-test.ts b/test/layout-test.ts
index 2a0f21d..1e0d274 100644
--- a/test/layout-test.ts
+++ b/test/layout-test.ts
@@ -232,9 +232,9 @@ function frogPrinceItems(): TextInputItem[] {
     }
   };
   const epilogue: TextInputItem[] = [
-    { type: 'penalty', width: 0, cost: 1000, flagged: false },
-    { type: 'glue', width: 0, stretch: 1000, shrink: 0, text: '' } as TextGlue,
-    { type: 'penalty', width: 0, cost: -1000, flagged: true },
+    { type: 'penalty', width: 0, cost: 100000, flagged: false },
+    { type: 'glue', width: 0, stretch: 100000, shrink: 0, text: '' } as TextGlue,
+    { type: 'penalty', width: 0, cost: -100000, flagged: true },
   ];
   return frogPrinceItemsImpl(frogPrinceText, prologue, betweenWords, epilogue);
 }
@@ -277,7 +277,7 @@ describe('layout', () => {
       );
       assert.deepEqual(
         adjRatios,
-        [0.857, 0.0, 0.28, 1.0, 0.067, -0.278, 0.536, -0.167, 0.7, -0.176, 0.357, 0.049],
+        [0.857, 0.0, 0.28, 1.0, 0.067, -0.278, 0.536, -0.167, 0.7, -0.176, 0.357, 0.0],
       );
     });
 
@@ -305,7 +305,7 @@ describe('layout', () => {
       );
       assert.deepEqual(
         adjRatios,
-        [0.774, 0.179, 0.629, 0.545, 0.0, 0.079, 0.282, 0.294, 0.575, 0.353],
+        [0.774, 0.179, 0.629, 0.545, 0.0, 0.079, 0.282, 0.294, 0.575, 0.004],
       );
     });
 

[robertknight]

This is because you define MAX_COST to be 1000, but Knuth & Plass's calculations are based on a MAX_COST of 100,000.

Ah, thanks for the clarification. That's fine to leave as-is for the moment then.

[robertknight]

The last adjustment ratio value also differs from the book here.

[basil]

Ditto