Code completion is the editor feature that proposes likely continuations as a programmer types, ranging from finishing a partially typed identifier to offering the members of an object, the parameters of a function, or a whole code snippet. By reducing keystrokes and surfacing the available API at the point of use, completion both speeds up writing and reduces errors from misremembered names. It is among the most heavily used features of any modern development environment.
The simplest forms are textual. An editor can complete a word from other words already present in the buffer, matching on a typed prefix without understanding the language at all. This is useful but shallow, since it cannot tell whether a suggestion is valid in context or know about symbols defined elsewhere in the project or in libraries that have not yet been typed.
Microsoft’s IntelliSense, documented for Visual Studio, exemplifies the semantic approach. The documentation describes features such as List Members, which shows valid members of a type or namespace after a trigger character like a period; Parameter Info, which displays the number, names, and types of a method’s parameters; Quick Info, which shows a symbol’s full declaration; and Complete Word, which finishes an identifier once enough characters disambiguate it. These features require the tool to actually analyze types, scopes, and accessibility, so that only members legal at that point are offered, with icons indicating whether each is public, protected, or private.
The Language Server Protocol generalized semantic completion across editors. Its specification defines a textDocument/completion request that an editor sends with a document position; the language server replies with completion items carrying labels, kinds (such as method, field, or keyword), documentation, and the text to insert. Because this runs over a standard protocol, a single language server can deliver type-aware completion to any editor that speaks LSP, rather than each editor reimplementing analysis for each language.
Completion has continued to deepen. Early systems matched prefixes; later ones added fuzzy and camel-case matching, where typing the capital letters of a camel-cased name surfaces it. Semantic engines rank suggestions by relevance and context, and more recent systems extend completion to whole lines and multi-line predictions generated by machine-learning models that work alongside the symbol-based list. Across this evolution the core idea is constant: use what the tool knows about the code to anticipate what the developer is about to write.