Zinc goals redefined and extended

Published on Jul 14, 2014

The scope of zinc is now slowly getting stabilised. Currently zinc is a platform that allows (would allow) one to build a basic application based on the same basic principles that Arduino uses. Zinc provides developer with access to MCU peripherals and a bunch of drivers for common hardware found in hobbyist projects.

How is zinc different from Arduino?

Rust allows to write much safer code than the usual arduino sketch. It means less problems in the runtime. You cannot bootstrap zinc with a simple setup() and loop() though, in a foreseeable future all supported platforms will be ported to platformtree, which means that a hardware specification must be written for an application. That one, while adding some typing to the “hello world” style application makes sure that hardware constraints are met at compile time (i.e. one cannot simply share same pin with a led and UART transmit and toggle it in different code paths, which makes sense for most of the cases).

How is zinc different from mbed?

Zinc is actually closer to mbed rather than Arduino now. It shares the same architecture (ARM being the only one zinc supports for now). Zinc tends to be more lightweight on the runtime side, as all the hardware-specific checks are done at compile time, lots of runtime code can be saved (there’s no reason to have code that computes UART clocks based on baud rate and system clock if the platform definition implies that system clock and baud rates will be constant).

Due to this, most of libzinc APIs, while maturing from experimental would be tagged as unsafe, that is — no runtime checks are done so it’s developer’s responsibility to make sure the arguments are valid. Those APIs are mostly for platform tree generator code, though, where PT parser can verify that all the arguments are in safe ranges.

So, getting back to zinc goals. The vision of zinc platform priorities is:

Platform trees in zinc

Published on Jun 27, 2014

Welcome to the zinc.rs blog, where I’ll post my discoveries on the path of using rust for embedded development. For today’s topic I’ll share my vision on platform trees.

While I’ve been prototyping some code for network stack, I figured that there’s no nice way to define IRQ callback functions in a way that would be compact, effective and safe. Rust is all about safety so I really do my best in terms of providing safe API to hardware. I went to device trees for some inspiration and figured that I can provide a safe way to initialize and manage hardware with some metaprogramming. And rust is extremely effective for metaprogramming allowing one to take a stream of tokens and produce some AST. While at that, I decided to hide all the platform-specific initializations and RTOS hacks, and that’s how platform tree was born.

Platform tree is a structure that contains nodes where each node can have some child nodes and/or attributes. It’s effectively XML without the namespaces and more human-friendly syntax without tags, which is a bit more like object declaration in some programming languages. The syntax does match device tree syntax rather closely. Apart from a few specific cases, nodes, child nodes and attributes are excessively checked for any inconsistencies. PT parser would not allow one to define PLL configuration that would burn the target MCU. All the verification checks are done in same rust code but at compile time. In the end, PT compiles down to hal::mcu::whatever calls, so there’s nothing you can do with PT that you cannot do without it.

The blink examples illustrates the usage. The implementation would probably server as a good reference to how macros and ast::TokenTree can be put to use.