Case Converter

Nothing in the string says where the words are

Converting to snake or kebab case means finding the boundaries, and an identifier does not carry them — the capitals are the only clue there is. Two rules read them:

getUserById   -> get_user_by_id
        a capital that follows a lowercase letter starts a word

      HTTPServer    -> http_server
        a run of capitals followed by a capital and a lowercase
        letter splits before that last capital, so the acronym
        stays whole

      userid        -> userid
        no capitals and no separators. There is nothing to read,
        and no rule can tell whether that is "user id" or one word.
        Every tool that guesses here is guessing.

The third case is the one worth remembering, because it is where a converter quietly produces something wrong rather than something odd. A name written entirely in lowercase arrived that way for a reason — it is a package name, a plugin slug, a word that was already one word — and the only way to split it is a dictionary of the language's words, which is a different tool with a different failure rate. Where the boundaries matter and the capitals are gone, the answer is a rename in the source rather than a conversion here.

The same reading explains what happens at the ends of a name. A leading capital is a boundary whether or not anything precedes it, a trailing digit is usually part of the word before it, and a name that already mixes conventions — an underscore in one place and a capital in another — comes out joined consistently, which is the point. The converter is not preserving your intent; it is applying one reading of the capitals and joining the pieces the way the target convention wants.

Title case is a style, not a rule

Which words stay lowercase in a title is decided by a style guide, and the guides disagree. The short list everyone half-remembers — the articles, the conjunctions and the short prepositions — is broadly shared, but where a guide draws the line between a short preposition and a long one varies, and so does whether a verb like is counts as a small word. Nothing about the text says which guide it is being held to, so the tool picks one and applies it consistently.

Two decisions inside that are worth knowing about, because both produce a result that looks like a bug and is not. A hyphenated compound is converted part by part, so each half is capitalised as a word in its own right. And a full stop is read as the end of a sentence wherever it appears, which means a filename in a title comes out with its extension capitalised — there is nothing in a string of characters that distinguishes the dot that ends a sentence from the dot that separates a name from its type. A title containing a filename is better handled by converting the words and typing the name back in.

The choice between Title Case and sentence case for headings is a house decision rather than a correctness one, and the thing readers notice is the inconsistency rather than the choice. One level of heading in Title Case and the next in sentence case reads as an oversight even when both are defensible on their own. Pick one, write it into the style notes, and let the converter do the mechanical part.

Where case does not exist

Case is a property of some writing systems and not others. Chinese, Japanese, Korean, Arabic, Hebrew, Thai and Devanagari have no upper and lower forms at all, so a conversion over text in any of them returns the text unchanged — not a failure, and not something to work around. A string that mixes scripts has a title case in the Latin part and nothing to do in the rest, which is why a translated heading can look half-converted when it has been handled correctly.

Where the scripts do have case, the mapping is not always one character to one character. The German ß uppercases to SS, two characters where there was one, so an uppercase conversion can lengthen a string — and a length limit, a column width or a filename that fitted before now does not. Accented Latin letters convert cleanly, with é becoming É and back, but a few languages apply rules of their own rather than the Unicode defaults, and a converter that does not know which language it is reading cannot tell.

The practical consequence is that comparing two strings case-insensitively is a language question rather than a byte question, and a name that differs from another only in case is best avoided altogether. Where a renamed file or a re-cased identifier has to be checked against something else that reads it, the Diff Checker will show the two versions side by side and the JSON Formatter will confirm a payload still parses after its keys were changed.