2026.05.17

  • Bumped literalizer to 2026.5.17.

  • :variable-name: (with :existing-variable:) on literalizer-call now produces a call-result binding for ada, bash, c, d, elixir, erlang, forth, fortran, nim, objectivec, systemverilog, tcl, and zig. These languages previously failed the build because they could not bind a call result; they now emit an idiomatic binding (for example set my_data [make_widget ...] for tcl and auto my_data = make_widget(...); for d), following upstream literalizer extending variable_form to those languages. No directive change was needed.

  • :heterogeneous-strategy: record (including the auto default’s record fallback) and :record-struct-name-prefix: now work for c, csharp, cpp, crystal, d, nim, odin, swift, v, and zig. This follows upstream literalizer adding the RECORD strategy to those languages; the directives expose it generically from the per-language capability flags, so no directive change was needed.

  • :heterogeneous-strategy: now defaults to auto instead of falling through to literalizer’s per-language default (e.g. error for Rust). auto renders the input with its natural representation first – so homogeneous and genuinely map-shaped data render identically to before – and genuinely unrepresentable input still raises, so this only ever turns a representable input that previously broke the build into a representable one. Projects that forced auto globally via a directive subclass in conf.py can drop that override. auto makes a representation choice implicitly (record vs tuple vs tagged_enum changes the generated API), so set :heterogeneous-strategy: explicitly – including :heterogeneous-strategy: error to keep a mixed-scalar collection a hard build failure – when a specific behavior is wanted.

  • The documentation examples now use sphinx-toolbox’s rest-example directive, which executes each literalizer and literalizer-call directive at build time, so the rendered output can no longer drift from the pinned literalizer behavior.