Hare aims to become a 100-year programming language November 8, 2023 by Drew DeVault
There are a number of languages attempting something similar to what Hare is doing by positioning itself as an alternative to C. However, there is an important trait which I think is necessary to understand Hare’s ambitions in this space: Hare is explicitly attempting to emulate C’s 50+ year staying power. One of Hare’s explicit design goals is to produce a programming language which will be stable over a very long period of time, exceeding the lifetime of its authors: I think of Hare as a 100-year programming language.
This goal imposes some important constraints on our design space, which is worth acknowledging and exploring a bit. What does our conscious emphasis on long-term stability imply for the language design?
We have identified a number of points which are important for this purpose:
- Conservatism in language design
- The importance of the standard
- The necessity of a feature freeze
- Defining long-term API stability goals
- Fostering a culture that values stability
By focusing on these objectives, we hope that Hare will outlive its designers. As I’ve said before, we want to build our language such that you can write a program on the day Hare 1.0 is released and it will still build in 100 years. Moreover, a Hare compiler written on day one will be able to compile contemporary Hare programs in a century’s time. I don’t think there is any other language project with these ambitions.
Conservatism in language design
One of the most important consequences of Hare’s long-term stability goals is a conservative approach to language design. Hare as a language is not actually particularly interesting or innovative. There are a number of language features being explored in other languages in this space (for instance, Zig’s experiments with compile-time semantics, or async/await in Rust) which are more innovative, and therefore more risky.
Hare instead aims to distill the as-of-this-decade state of the art in proven systems programming language design idioms to produce what we believe is the most robust language design right now. We have made the conscious choice to leave more novel design and experimentation to other languages, so that Hare can fill a niche wherein it may be suboptimal in respects other than stability. Our goal is to make Hare the tool of choice when writing a program which needs to be operational and maintainable for a long time, such that all of the design choices still make sense with a decade or two of hindsight.
Consequently, Hare aims to be a simple, elegant, and robust integration of only battle-tested design idioms. In other words, Hare is designed to be a boring language.
The importance of the standard
From the start, we have been developing Hare as a standardized programming language. The Hare specification aims to nail down the semantics of the language independently of its implementation, for the following reasons:
- Each feature is formally considered and specified, to avoid overlooking design pitfalls
- Our implementation is kept honest such that its behavior is constrained to a separate source of truth
- Third-party implementations can be developed in the future to bring Hare to future platforms and contexts beyond what we could have foreseen today
- Third-party implementations converge on a source of truth regarding how Hare is meant to behave
- Programmers have a formal reference to understand exactly how Hare is meant to behave such that they can plan out projects to be supported long-term
The specification also gets a hand on vendor extensions and C’s problem with liberal interpretations of undefined behavior, which is important for knowing for certain what any Hare program is expected to do, now, in the future, and in the past.
Defining policy for long-term API stability
We have drawn up (or are drawing up, or at least planning to draw up) a series of formally defined stability policies for various parts of the Hare ecosystem.
The standard library, for instance, will only make source-compatible changes indefinitely following Hare 1.0.1 This necessarily constrains the standard library mandate, so that we consciously balance the utility of any given module for standard library inclusion against our faith in our ability for it to have long-term staying power. The “extended library” (a set of important libraries in the Hare ecosystem which have explicit support from upstream Hare, this is for example where HTTP support lives) has separate, well-defined, but more flexible policies regarding long-term support and stability. We are also going to publish stability recommendations and best-practices for maintainers in the ecosystem at large.
Where we leave room for important changes to be made (for instance, the deprecation of obsolete cryptography modules), we are also preparing policies for effective communication and implementation of those changes throughout the ecosystem, so that maintainers of Hare software have an effective line of communication to take action to keep their software stable.
Freezing the language at Hare 1.0
We have made the bold decision to make a number of guarantees when we release Hare 1.0, chief among them being that the language grammar and semantics will be permanently frozen.
Following the release of Hare 1.0, the specification will be finalized and all future changes to the specification will be limited to clarifications, with a narrowly defined class of changes which will be accepted. Each trait of the Hare language has an agreed-upon policy and plan for stability. To give just one example, we have planned to release a separate specification defining Hare for 16- and 8-bit CPUs and other unconventional CPUs, which have a different set of agreed-upon portability and stability guarantees. A separate policy exists or will be written for API compatibility, compiler flags, and so on.
Prior to releasing Hare 1.0, and subsequently freezing the language and committing to long-term source-compatibility in the standard library, we will be undertaking a project of enumerating the Hare 1.0 “acceptance criteria”. We have identified a number of key focus areas which demand attention and deliberate review before committing to their implementation in perpetuity. These range from as broad as “the specification” and as narrow as “slices” or “IPv6”. Each of these focus areas will be assigned an evaluation team which will review them with a fine-toothed comb and produce a report regarding their soundness and long-term viability. This process will be completed before Hare 1.0 is released and is expected to take months or years.
Developing our culture
A more subtle part of our praxis comes down to how we cultivate a culture and community around Hare. Culture develops organically, but we can influence it by talking about our goals and what’s important to us. We attempt to instill these values in new contributors and users of the language, and propagate our beliefs through the community as it develops and grows.
Hare users are given resources on why stability is important and how we aim to achieve it, and we develop these values as a part of our consensus and consensual cultural development. We talk about best practices for libraries, API maintenance, deprecation, software design, packaging, and so on, with an eye towards fostering long-term thinking.
This is a vague goal and has a vague implementation, but it is important. You can see it in the way that people exposed to the Hare community start thinking about their work, and it’s meaningful and important.
The practical consequences and trade-offs
An intended consequence of these choices is that Hare is not going to be a language for everyone and everything. We know that we are leaving use-cases on the cutting room floor by making these trade-offs, and we think specializing is better than generalizing when it comes to picking which use-cases to include. We’re okay with the fact that Hare is not going to pick up any users in Python’s niches, for instance, something that Go for example is doing “better” than us.
Hare is necessarily going to become a time capsule of its era. Innovation in language design is important and moves the industry forward, but Hare is going to step away from those innovations. The cost of long-term stability is that Hare will inevitably become the less optimal choice over time, as a new set of language ideas becomes proven and tested and enters the canon of essential language idioms — essential idioms which might have been incorporated into Hare if they were proven today, but which will be left out in the interest of long-term stability. The next language can tackle these problems.
In this respect, Hare is unlikely to be a 100-year language in the sense that it will still be the best choice in a century — but rather because it will still work in a century. We believe that there is value in providing a language that can be depended on to work consistently for a very long time. If Hare succeeds in its ambitions, it will form an important part of the culture of long-term thinking and stability which provides a necessary counterweight to rapid innovation in software engineering.
This approach to innovation, and our near-rejection of it as a goal, is innovative in and of itself. If I were to point to one trait of Hare which sets it apart from the rest of the pack, this is it.
An exception is made for cryptography-related modules, which will necessarily change to track modern cryptographic algorithms and rotate out support for insecure algorithms. Obsolete algorithms are moved into a separate tree from the standard library and may still be manually installed to build software which depends on obsolete crypto. ↩︎