Text case, sorted instantly

Naming conventions are arbitrary in isolation but load-bearing in practice. Most languages have a dominant style, and following it is what makes unfamiliar code readable. Fighting it produces code that works but reads as foreign to everyone else on the project.

Converting between cases means splitting a string into words and rejoining them. The rejoining is trivial. The splitting is where information gets lost, because some formats do not record where the boundaries were.

Going from camelCase to snake_case is reliable, since every capital marks a boundary. Going the other way is fine too. But a round trip through a lossy step does not always return the original.

Capitalising every word is straightforward but not what most style guides mean by title case. The common convention capitalises the first and last words, plus everything except short articles, conjunctions, and prepositions.

The exact list of exceptions differs between guides. Chicago, AP, and MLA disagree on where the cutoff sits, particularly for prepositions of four or more letters, so there is no single correct output.

For headings and titles this is worth doing by hand. For identifiers in code it is irrelevant, since the convention there is mechanical.

Case conversion depends on locale, and one case in particular breaks naive code. In Turkish, the uppercase form of i is a dotted capital I with a dot above, and the lowercase form of plain I is a dotless small i.

This means that lowercasing a string under a Turkish locale can produce a value that no longer matches an ASCII comparison, which has caused real bugs in configuration parsing and case-insensitive string matching.

Most casing bugs appear at the seams between systems that follow different conventions. A JavaScript client uses camelCase, a Python service uses snake_case, and a Postgres database folds unquoted identifiers to lowercase.

The reliable approach is to convert once, at a single well-defined boundary, rather than sprinkling conversions through the code. A serialisation layer that maps between the wire format and the internal format keeps the rule in one place, and makes it obvious which side of the boundary any given name belongs to.

Postgres deserves particular care: unquoted identifiers are folded to lowercase, so a column created as userId is stored as userid. Quoting preserves the case but then requires quoting everywhere thereafter, which is why snake_case is the path of least resistance in SQL.

Kebab-case is the standard for URL paths for practical reasons rather than aesthetic ones. Search engines have long treated hyphens as word separators and underscores as word joiners, so my-blog-post reads as three words while my_blog_post can read as one.

Producing a slug involves more than changing the separator. Text needs lowercasing, accented characters need transliterating or stripping, punctuation needs removing, and runs of separators need collapsing. Trailing hyphens left behind by stripped punctuation should go too.

Keep a stored slug stable once it is published. Regenerating slugs when a title is edited breaks every existing link, so the usual pattern is to generate once on creation and treat later changes as a deliberate redirect rather than an automatic rename.