Dylan is in an unusual position for a programming language.
It has a pretty mature implementation, a specification, and a lot of high quality documentation. It has a long history and many people remember it from the early days of its history when it was being developed at Apple and other institutions.
On the other hand, the current community is fairly small as is the available set of libraries. This is typically a disadvantage, however it can be used to the community's advantage in a couple of ways.
One way is that we can look forward and take advantage of what we know and want now without having to worry about backward compatibility.
On the opposite side of this are languages like Python where there is no one solution for handling non-blocking I/O and so some libraries do one thing, others do something else, and large parts of the standard library don't deal with it at all.
In Scala and Java, there's no coherent strategy for tracing execution, metrics gathering or concurrency. Twitter (as an example) has solved this for themselves by producing a large set of libraries that stand on their own and integrate together, but they don't integrate well with other frameworks from other companies (like, say, Akka). Logging used to be an issue, but this has largely been solved by slf4j-api.
In these (and many other) cases, you see fragmentation amongst the libraries as newer considerations that weren't present when the initial libraries were written get dealt with separately rather than as part of the language's core libraries.
Clojure is a demonstration of how a language can benefit from being forward-looking by having a rich and comprehensive set of concurrency primitives that are available as part of the standard library.
How might this work with Dylan? What core concepts do we want to support throughout our software stacks to give us an advantage over others?
A few things come to mind:
- Fully integrated tracing framework.
- Coroutine-based solution to non-blocking and asynchronous work.
- A solid approach to Unicode strings and byte buffers.
We have just produced a new tracing library (based on Google's Dapper) for use in Dylan software. We should be able to readily integrate it with the HTTP, concurrency, nanomsg and database libraries and other libraries in the future. Any server side software using Dylan should be able to have awesome tracing support from day one.
Event-driven software often means having to use callbacks. While we can support callbacks in Dylan without any issue, it is also nice to be able to use coroutines to have more natural looking code. We can integrate coroutines as a threading model within the compiler, runtime and common libraries, including the I/O libraries.
We should have a clear distinction between buffer (byte vector) types and strings. Strings should be Unicode and this should be supported by all libraries. This was something that was done in Python 3.0 and is a great improvement over the past.
We should look at improving our concurrency support and our usage of (functional) data structures that are better suited for concurrency. But there are probably many other ways that we can help build libraries that are well suited for the future.
Another way that we can benefit from having a small community and set of libraries now is where we set the bar for quality and lead the next generation of Dylan developers to follow.
We should require that all packages / libraries follow some guidelines:
- Consistent documentation tooling.
- Consistent testing framework usage.
We should be conservative about accepting new packages and generous in offering up resources like continuous builds to help everyone achieve these goals.
This is an area where Perl has done very well with CPAN. Packages have a consistent build, test, and documentation procedure: something we should aim to emulate.
All of our documentation is written using Sphinx using our Dylan extensions. By having a single toolset for all of our documentation, we'll be able to build new tools on top of that. An example would be writing a JSON exporter and providing searchable, browsable documentation using something like ElasticSearch on the back-end.