On 23-05-23, 13:25, Erik Schilling wrote: > As of now, the Rust bindings are only consumable as git dependencies > (and even then come with some restrictions when wanting to control > the build and linkage behaviour). > > This series does some (hopefully) cleanup and then proposes a change > in how the Rust bindings are built and linked. > > Since the changes may require people hacking on the bindings to set some > additional environment variables (at least if they want to avoid running > make install), I am sending this as an RFC in order > to hear opinions. > > For gpiosim-sys the situation is slightly more complex. Right now, > libgpiosim installs without a pkg-config. If it is desireable to add > one, that could be done and the same mechanism could be used. Otherwise, > if packaging that lib is desirable (it looks like it?), we could either > still query for libgpiod (and hope that the linker and include paths are > matching) or need some other way to acquire the linker and include paths > (and flags). > > So... The open questions: > - Is this OK at all? Are people depending on this building against > relative paths? Not sure if the build of the libgpiod git repo depends on that relative path, did you try to do `make` in the libgpiod directory ? Like: $ ./autogen.sh --enable-tools=yes --enable-bindings-rust --enable-examples --enable-tests; make > - What to do with gpiosim-sys (see above)? The only user for the cargo tests for libgpiod stuff is for the vhost-device crate with the help of rust-vmm containers. I am installing libgpiod there currently and so it works, not sure of how it may be required to be used later on though. > - Is there interest into getting this published on crates.io after > packaging is fixed? Yes. -- viresh