Philosophy

TCL was conceived at a time when most networked software was written in C at the core, the internet was young, user expectations were looser.

It is a tiny language full of great ideas, but features that were useful 20 years ago are a hindrance today:

• I/O in the language is an obstacle, as the host is more than likely to already have taken a stance on how it wants to handle I/O,
• a built-in event loop for multiplexing I/O and orchestrating timers was useful when no host could easily provide this, but event loops are widespread and having to integrate multiple event loops in one application is error-prone.
• reference counting with lots of calls to malloc and free works great for standalone TCL, but the emergence of zig and wasm incentivizes being in control of allocations.

So what ideas are worth preserving?

A pure form of metaprogramming, syntax moldable like clay, with meaning to be added at a time and in a form that is convenient for that particular use case.

A transparent execution environment: every aspect of a running TCL program can be inspected from within that program, and often even modified.

A focus on expressing computation in the form of instructions to carry out.

The latter point is key: agentic coding benefits from an inspectable and moldable environment. Having the agent talk to your running program gives it highly valuable feedback for a small amount of tokens.

The browser is one example of this model being successful, but what about all the other applications? Your job runner, web server, database, your desktop or mobile app.

feather wants to be the thin glue layer that's easy to embed into your programs, so that you can talk to them while they are running.

Another way to look at TCL is this: it is a Lisp-2 with fexprs that extend to the lexical syntax level. Maybe that is more exciting.