Languages & the code knowledge database.

What squeezy can answer about your code

For every language listed below, squeezy indexes your files into a code knowledge database: the things your project declares, and the links between them. It is built from your code on your machine, nothing is uploaded to build it, and it keeps up as you edit.

Each of these is one lookup rather than a hunt through files.

• Where a symbol is defined. By name, or by an id carried over from an earlier answer.
• What refers to it, what calls it, and what it calls. Callers and callees can be walked several hops out, not only one.
• What a module or class contains. The whole member roster in one answer, instead of a read per member.
• What a type inherits from, and which types inherit from it. The whole chain, not only the first generation, and which subtypes override a given method.
• What a change would affect. The files, symbols, and tests downstream of an edit, taken from a symbol you name or from the diff you already have.
• Which symbol is at a position. A line, a column, or a byte offset resolves to the symbol around it, so a compiler error or a search hit becomes a starting point.
• A signature on its own, or a body on its own. The model gets the lines that matter instead of the file around them.

squeezy is not a compiler and does not type-check your project. It reads structure, which is why every relationship it reports carries a confidence label: an exact match is marked differently from one candidate among several, and an answer that a macro or a build flag makes uncertain says so rather than reading as fact. When completeness matters more than precision, squeezy says so too, and falls back to plain text search.

Files in any other language still work. They fall back to bounded search, targeted reads, and shell commands, and squeezy lists them as unsupported instead of reporting a lookup it never made. What follows is the full supported list, grouped so it scans.

How a file gets classified

Usually by file extension. Compound endings are matched as one unit, so a .d.ts is indexed as TypeScript and a .rst.txt as reStructuredText rather than as a plain text file. Files with no extension at all are covered by name, so nothing useful is skipped for lacking a suffix: Gemfile, Rakefile, Podfile, Brewfile, and Vagrantfile are indexed as Ruby, Jenkinsfile as Groovy, and shell startup files such as .bashrc and .zshrc as Shell.

Where an extension is genuinely ambiguous, squeezy asks for evidence instead of guessing. A script with no extension is classified from a fixed list of shebang interpreters. Endings that several ecosystems share, such as .pl and .t, are indexed only when the start of the file shows which language wrote them. A .h header is decided by the C, C++, or Objective-C sources sitting beside it. A file that stays inconclusive is left out of the database and treated like any other unsupported file: still readable, still searchable, but not indexed.

Every indexed file also carries a role beside its language: implementation, test, configuration, DSL, stub, interface, and a few more. The role is separate from the grammar, which is what lets a Ruby Gemfile and a Ruby library file share a parser without being treated as the same kind of input.

How deep the coverage goes

Coverage is not uniform, and it is checked by language family: usually one language, sometimes a few close relatives indexed together. Where an outside authority for a family exists, squeezy is graded against it rather than against its own expectations: Rust against rust-analyzer, Python against CPython's own parser, Java against javac, C# against Roslyn, C and C++ against clang, JavaScript and TypeScript against the TypeScript compiler, and the same arrangement for Go, Ruby, PHP, Swift, Scala, and Dart. Every other family except the documentation formats is held to a checked-in fixture project with its own navigation questions and expected answers. The documentation formats have neither, and are covered by per-format extraction tests and doc-linking tests instead.

The benchmark page carries a measured cost result for 15 of these languages. squeezy answers the coverage question offline as well: squeezy help languages at a shell, or /help languages in a session, prints every supported language family from the registry inside the binary you installed.