ToPascalCase/ToCamelCase change their own output when run again on adjacent one-letter words: "vector x y" → "VectorXY" → "VectorXy"

[matt-edmondson]

What's wrong

Adjacent one-letter words turn into a run of capitals in PascalCase: "vector x y" → "VectorXY", "a b c" → "ABC", "x_y" → "XY". When that output is converted again, IsWordBoundary reads the run as a single all-caps word, and LowercaseAllCapsWords (the per-word all-caps normalisation added for #72) lowercases it. So the converters don't return their own output unchanged, and round trips silently lose word boundaries.

Relevant code, all in CaseConverter/CaseConverter.cs:

• ToPascalCase (~L370)
• LowercaseAllCapsWords (~L240)
• the acronym-tail rule in IsWordBoundary (~L150)

Reproduction

Against the net10.0 build of HEAD (1c1c555):

"vector x y".ToPascalCase()               = "VectorXY"
"VectorXY".ToPascalCase()                 = "VectorXy"   // expected "VectorXY"
"vector x y".ToCamelCase()                = "vectorXY"
"vectorXY".ToCamelCase()                  = "vectorXy"   // expected "vectorXY"
"a b c".ToPascalCase().ToPascalCase()     = "Abc"        // first pass "ABC"
"x_y".ToPascalCase().ToPascalCase()       = "Xy"         // first pass "XY"

"vector_x_y".ToPascalCase().ToSnakeCase() = "vector_xy"  // boundary between x and y lost
"a_b".ToPascalCase().ToSnakeCase()        = "ab"

A single one-letter word does survive, because the next letter is lowercase and the acronym-tail rule applies: "get_a_value" → "GetAValue" → "get_a_value".

Why it matters

Identifiers like coord_x_y, matrix_m_n and get_x_y are common. A code generator or rename tool that normalises names more than once, or goes snake → Pascal → snake, gets different identifiers on each pass: VectorXY becomes VectorXy, and coord_x_y becomes coord_xy.

Suggested fix / acceptance criteria

ToPascalCase and ToCamelCase should not produce output they would convert differently on a second pass. One option: when a one-letter word follows another one-letter word, lowercase it, so "vector x y" → "VectorXy" on the first pass. That output is stable, although vector_x_y → Pascal → snake would still come back as vector_xy. If that fold is acceptable, document it.

Tests:

• s.ToPascalCase().ToPascalCase() == s.ToPascalCase(), and the same for ToCamelCase, for "vector x y", "a b c" and "x_y"
• a tested expectation for "vector_x_y".ToPascalCase().ToSnakeCase()

[matt-edmondson]

Triage

• Category: Bug
• Priority: Medium. ToPascalCase and ToCamelCase are not idempotent for adjacent one-letter words (VectorXY → VectorXy), so identifiers drift across repeated normalisation. This is an edge case, but a real one for code generators.
• Area / suggested owner: CaseConverter.cs: ToPascalCase, LowercaseAllCapsWords (from Whether an all-caps word is normalized or kept as an acronym depends on what else is in the string #72), and the acronym-tail rule in IsWordBoundary.
• Duplicates / in progress: Not a duplicate. It is a side effect of Whether an all-caps word is normalized or kept as an acronym depends on what else is in the string #72's all-caps normalisation. The open PRs Keep the letter after a typographic apostrophe lowercase in ToTitleCase [patch] #91–Keep an all-caps word containing ß in one piece, so ToMacroCase is idempotent [patch] #94 all touch idempotence and acronym handling in the same file (possessives, ß, apostrophes), but none of them covers one-letter runs. Expect conflicts, so land this after those PRs or rebase onto them.
• Suggested next step: Decide whether the vector_x_y → vector_xy fold is acceptable and document it. Then add the ToPascalCase().ToPascalCase() idempotence tests from the issue.

---

Generated by Claude Code

[matt-edmondson]

Decision (maintainer, 2026-09-28)

• Take the suggested option: when a one-letter word follows another one-letter word, lowercase it. The first pass then gives "vector x y" → "VectorXy", which is stable on re-conversion.
• The vector_x_y → Pascal → snake → vector_xy fold is accepted. Document it in the README and the XML remarks as a known, deliberate loss of the boundary between adjacent one-letter words.

Next reader: implement the lowercase rule in ToPascalCase/ToCamelCase. Add the idempotence tests for "vector x y", "a b c" and "x_y", plus a pinned expectation that "vector_x_y".ToPascalCase().ToSnakeCase() == "vector_xy", and document the fold. Land it after, or rebase onto, the open PRs #91–#94 that touch the same code.

---

Generated by Claude Code