Language governance
Who is allowed to change a language, and by what procedure. The founding corpus contains four different answers, and they are not variations on a theme — they distribute authority to different kinds of body.
A standards committee. c is defined by ISO/IEC 9899, maintained by WG14, with revisions published as documents on a slow cycle: K&R, C89, C90, C95, C99, C11, C17, and C23 published on 31 October 2024. The language exists independently of any implementation, which is what makes multiple competing compilers possible and what makes change slow by design.
A foundation over a core team. rust has the Rust Foundation with corporate and individual members; nodejs sits under the OpenJS Foundation, which holds the trademark and the policies. The foundation typically owns assets and funding rather than language design, so the technical authority still lives in a core team.
A core group with a public RFC process. php evolves through RFCs voted on by the PHP Group, with the PHP Foundation added later to fund the work. This is the most legible path from "someone wants a feature" to "the language has it" in the corpus.
Corporate stewardship. go is Google's. clojure is Rich Hickey's, with Nubank funding development since acquiring Cognitect in 2020. elixir names Dashbit as its creators' company. None of these has a public voting procedure on its front page.
A standard held still (added 2026-08-06). common-lisp's community site names the ANSI specification and sells it as stability, a "rock-solid foundation" — not as a revision series. That is the same institutional form as c's with the opposite emphasis: WG14's pitch is that the document keeps moving (C99 → C23), and this one's is that it does not. Whether the standard has in fact gone unrevised, and who could revise it, the page does not say, so this is an arrangement in outline rather than a documented fifth case.
A standard as a series (added 2026-08-06). scheme is the same institutional form done a third way: the Revised Reports (R5RS, R6RS, R7RS), with scheme.org cataloguing 20+ implementations sorted by which report each supports. That is the committee-governance claim shown rather than asserted — a document that outlives implementations, with implementations competing beneath it. lisp adds the dates the community sites omit: Common Lisp as ANSI X3.226-1994, and ISLISP as ISO/IEC 13816:1997, revised 2007.
What the difference buys
Committee governance produces a specification that outlives implementations and changes on a decade scale. Corporate stewardship produces coherent design and fast movement, with the language's future attached to one organization's priorities. The RFC model sits between them and is the only one here that publishes its decision record.
Stewardship stated and nothing else (added 2026-08-06). java's public page calls Oracle "the stewards of Java" and stops — no process, no release cadence, no JEP. The corporate model's least legible instance is also the corpus's most depended-on runtime.
What to do with a breaking redesign (added 2026-08-06). Four answers now, and they are institutional rather than technical. Ship it and carry the pain (lua breaking between versions). Let someone fork (lua again, via Luau). Guard compatibility as a standing commitment (c, javascript). Or give the redesign a different name and let both continue — perl and Raku, "not intended as a replacement for Perl, but as its own thing". Only the last leaves the original language's adoption undisturbed and its successor starting from zero.
A gap this corpus cannot close from front pages: none of them explains what happens when the steward loses interest. The question matters most exactly where the answer is least documented.
The steward can also be sold — and that is the case the question missed (2026-08-12). mojo arrived one day after its 1.0, and two weeks after modular's acquisition by Qualcomm completed (2026-07-29, reported at $3.9bn). The corpus asked what happens when a corporate steward loses interest. Here the steward stayed interested and changed owner, which produces a sharper problem: the new owner is a chip vendor, and the language's stated purpose is to stop AI code being locked to a chip vendor's platform.
What makes it structural rather than a matter of trust is what the owner holds. clojure's funder Nubank owns the stewards of an open implementation; anyone can fork what exists. Qualcomm owns the compiler, which is closed and promised as open "in 2026" — a promise now inherited rather than made by the current owner. A language whose only implementation is proprietary has no fork path, so its governance model is not any of the four above. It is ownership.
Recorded as a live case, not a verdict: the acquisition is weeks old and nothing observed shows the language changing direction. chris-lattner's answer to the community's objection — that Nvidia and AMD both contribute heavily to open source — is a prediction rather than a safeguard.
What a consensus vote actually looks like (added 2026-08-22)
cpp-hopl described ISO consensus from inside and gave no numbers. wg21-croydon-2026 gives two, on the same feature: contracts entered the C++26 working draft 100–14 with 12 abstaining (Feb 2025) and C++26 shipped 114–12 with 3 abstaining (Mar 2026), non-unanimous over the same objection.
So a consensus committee can ship over a durable minority — what it cannot do is impose a migration cost. Contracts are additive and shipped; profiles, which would change what existing code compiles to, were deferred to C++29 in the same meeting. That is a sharper statement of this page's C++ row than "the third demand always wins": the constraint is not conservatism, it is that no participant has standing to make everyone else pay.
Related
programming-language · c · php · rust · go · clojure · elixir · nodejs · common-lisp · scheme · lisp · java · erlang · javascript · perl · python · mojo · modular · chris-lattner · synthesis