riot-sys

Bindings for RIOT system calls

This crate contains dynamically generated Rust FFI bindings to the RIOT Operating System.

Those bindings are inherently unsafe; it is recommended that their safe abstractions in the riot-wrappers crate are used in most applications.

For a newcomer's starting point, see RIOT's documentation on using it with Rust. This also contains installation instructions / dependencies.

RIOT integration

Which functions and structs are present in this crate, and sometimes their details, inherently depends on the RIOT configuration this will be used with. For example, RIOT's thread_t only has a member name if DEVHELP is set for a build, and its flags member is only present if the thread_flags module is in use.

All the relevant information -- including the location of the actually used RIOT header files and flags influencing the ABI -- is conveyed to riot-sys by passing on the compiler and the CFLAGS. This can either be done by passing in the path to a "compile commads" file through the RIOT_COMPILE_COMMANDS environment variable (accompanied by a RIOT_USEMODULES, as that part of CFLAGS is missing from the compile commands), or alternatively by passing in the C compiler as RIOT_CC and the CFLAGS (both their CFLAGS_WITH_MACROS and the INCLUDES part from RIOT's build system) in. When called from within RIOT's build system, care must be taken to clear CC and CFLAGS, as these would be interpreted by Cargo (Rust's build system) to refer to the host compiler and flags. The flags will be interpreted by libclang based tools; care must be taken to pass in flags suitable for clang and not for GCC.

These steps are automated in RIOT's build system.

The RIOT_CC and RIOT_CFLAGS are made available to dependent crates through Cargo (as DEP_RIOT_SYS_CC etc); see riot-wrappers's build.sh for an example. Similarly, custom markers are made available based on the presence of certain defines or features in RIOT as downstream crates require that information (typically to allow a crate to work across a wider range of RIOT versions); see the section below for details.

Extension

Currently, only a subset of all the RIOT headers is processed; all the relevant header files are included in this crate's riot-headers.h header file. If you need access to more RIOT APIs, more includes can be added there.

Versioning

riot-sys is versioned using SemVer, and efforts are made to not make breaking changes even while in the 0.x phase. Note that as it passes on RIOT internals, any of the SemVer guarantees only hold when built on the same RIOT -- once the underlying C code is changed, all bets are off. Users of riot-rs can introspect the DEP_RIOT_SYS_... variables that are available to crates that set links = "riot-sys" to affect the symbols those crates use. Typical variables to inspect are DEP_RIOT_SYS_BINDGEN_OUTPUT_FILE (to determine whether a symbol is imported in the first place, eg. when RIOT renames something) and DEP_RIOT_SYS_CFLAGS which includes the enabled modules.

Markers

Deprecated, see below.

Some decisions of downstream crates need to depend on whether some feature is around in RIOT. For many things that's best checked on module level, but some minor items have no module to mark the feature, and checking for versions by numers is not fine-grained enough, so it's easiest to check for concrete strings in the bindgen output.

The build.rs of this crate contains a list of marker conditions. These lead to MARKER_foo=1 items emitted that are usable as DEP_RIOT_SYS_MARKER_foo=1 by crates that explicitly links = "riot-sys". They are stable in that they'll only go away in a breaking riot-sys version; downstream users likely stop using them earlier because they sooner or later stop supporting old RIOT versions.

For example, in PR #17957, an argument to a particular handler function changed fundamentally; no amount of .into() would allow writing sensible abstractions. The marker coap_request_ctx_t was introduced, and is present automatically on all RIOT versions that have that particular pull request merged. Code in riot-wrappers uses conditions like #[cfg(marker_coap_request_ctx_t)] to decide whether to use the old or the new conventions.

These markers are currently checked against bindgen's output, but could query any property that riot-sys has access to. The markers are defined in terms of some change having happened in RIOT; the way they are tested for can change. (In particular, when riot-sys stops supporting an older RIOT version, it can just always emit that marker).

Crates building on this should preferably not alter their own APIs depending on these, because that would add extra (and hard-to-track) dimensions to them. If they can, they should provide a unified view and degrade gracefully. (For example, riot-wrappers has the unit T of the phydat_unit_t in its enum, but converts it to the generic unspecified unit on RIOT versions that don't have the T type yet -- at least for as long as it supports 2022.01).

Deprecation note / successor: As of 2023-02, no new markers will be added, because implementing this mechanism here has shown to be impracitcal: Changes need to go into riot-sys first before they can be use (and tested in) with riot-wrappers. Instead, BINDGEN_OUTPUT_FILE is exported (usable as DEP_RIOT_SYS_BINDGEN_OUTPUT_FILE), which points to the Rust file generated by bindgen. Downstream crates can inspect that file, possibly using the same string-based methods as riot-sys uses, but without any cross-crate dependencies.

Crates accessing the BINDGEN_OUTPUT_FILE should exercise the same caution that is recommended above for the use of markers.

The types and constants of RIOT are translated in two forms: through bindgen (to be linked together), and through C2Rust (transpiled, to be inlined). This is necessary because neither can express the full set of useful RIOT APIs.

All bindgen types are reexported in the main module and exclusively public through there. The C2Rust types largely reside in the [inline] module, with some pub used into the root module as necessary or convenient.

License

This crate is licensed under the same terms as of the LGPL-2.1, following the license terms of the RIOT Operating System.

It is maintained by Christian Amsüss chrysn@fsfe.org as part of the etonomy project, see https://etonomy.org/, and the RIOT team.