Manual: future

Long term goals & strategies

Having more than one contributor, it is time to write out clear goals for Lever.

Communication of goals and strategies is important for meaningful movement as a group. This page is the place for that communication. The contents are revised every once in a while.

Table of contents ↑
Table of contents
01. Exploration
01.1. Standardization
02. Practicality
02.1. Code manipulation / analysis framework
02.2. Improved intermediate-results/module code memoization
02.3. Extensible abstract arithmetic
02.4. Live editing
02.5. Concurrency
02.6. Custom formats and protobuf
02.7. Pretty printing
03. Penetration

01. Exploration

People did a lot of computer science related discoveries prior to the 1990s. A recurring theme was to make the machine assist the user in programming. For a long time it was impractical to implement such tools because computers were simply too slow.

A lot of programming language implementations exist in a calcified state. The internals that are left undocumented, as if deliberately to discourage any significant changes to them. The user is not supposed to change the implementation even if the source code is available.

This means that our programming tools have advanced by revolution rather than evolution. The possibility for a real revolution shrinks every year a little bit. You just can't throw everything away once in a decade.

The internals of Lever are highly flexible. Our goal is to retain this plasticity through the lifetime of the language.

01.1. Standardization

Having seen how node.js software breaks despite version numbering makes us skeptical that version numbering is sufficient to ensure compatibility.

In short: It is likely that semantic versioning doesn't really address issues related to changing APIs. Node.js, as an example, follows semver but is messed up on a regular basis.

As the result, we're picking up something that has been proven to communicate compatibility. We will establish a standards board for our language. The standards established by the board are used to ensure compatibility.

Version numbers will be still used, but only for identifying releases of software.

02. Practicality

We strive to create a practical tool for solving difficult problems. Supremely abstract and sophisticated programming languages usually perform badly. For this reason the practical code is frequently written in crappy programming languages.

02.1. Code manipulation / analysis framework

We are designing a framework to allow manipulation and analysis of source code after it has been compiled into a bytecode or loaded into the runtime environment.

This framework will be supplied with optimization, partial evaluation, static analysis, symbolic execution, theorem proving, validation, formal verification tools and libraries.

02.2. Improved intermediate-results/module code memoization

Once we have software with very good runtime behavior, we do not want to compile it every time when a program is loaded. We require user-accessible intermediate formats to make most out of machine code that is produced.

On GPUs some of this has been already solved for us, in form of the SPIR-V binaries. So we need to come up with something that works with CPUs as well.

02.3. Extensible abstract arithmetic

We want the arithmetic to be extensible from the user side, but it is not very usable if you have to worry about whether you just broke half of your language by introducing a bad implicit conversion.

More about the subject in the "How should the addition -expression behave?"

02.4. Live editing

Smalltalk demonstrated functionality where you could edit your software while it was running. This is an incredibly powerful tool for design of software that is highly interactive.

Today live programming doesn't appear to be supported on programming languages or their implementations. Perhaps perceived to be too volatile for real-world use? While it allows software to be developed quickly, it also allows to write the kind of programs that never start again if you attempt to reset them.

02.5. Concurrency

We have great tools for implementing concurrency inside a single thread.

There are concerns that the provided concurrency is too much and leads people to writing unreliable software. We need to study these ideas formally and see what kind of conclusions we get. It is possible there is something we would have to consider even if our approach worked out apparently well.

The paper "Concurrency Among Strangers" will be used as a starting point.

02.6. Custom formats and protobuf

Machine-readable logs and how the data is stored are going to be more important in the future.

I think that protobuf got some of the details right, mainly by selecting good design goals. But the resulting tools were clumsy and difficult to use with dynamically typed languages.

Protobuf also ignores the fact that sometimes we want to observe and modify the data directly. The text editing was perfected through efficient data structures such as the piece tables, those could be also used here. There are more ways to diff tree-structured data than there are ways to diff plain text, which also could be useful with tightly packed non-compressed persistent log data that protobuf attempts to cater for.

We are going to need much better tools for handling binary data. Also need to decide on formats. Likely we would like to derive something from protobuf that's better for the kind of persistent storage needs that arise in the runtime.

02.7. Pretty printing

Debugging a program often involves observing a state. To allow this we could use really good set of tools for pretty printing. Since we have environments where we can show 3D graphics, it would be preferable that our pretty printing would match with this environment while being backwards compatible with terminal views.

We would somehow need to handle both passive and interactive forms of prints. The interactive prints do not necessarily need to be terminal-friendly.

03. Penetration

Since we have a really nice language, we want to use it for something and see how it fares.

Our current focus is on the graphics related programming. To find a bit stronger niche we focus to software that runs in Virtual Reality.

Ultimately we would like to develop software in virtual reality and produce all assets while we are "in the program" ourselves.