A couple of weeks ago we had the fourth GNOME+Rust hackfest, this time in Thessaloniki, Greece. This is the beautiful city that will host next year's GUADEC, but fortunately GUADEC will be in summertime!
We held the hackfest at the CoHo coworking space, a small, cozy office between the University and the sea.
Every such hackfest I am overwhelmed by the kind hackers who work on [gnome-class], the code generator for GObject implementations in Rust.
Mredlek has been working on generalizing the code generators in gnome-class, so that we can have the following from the same run:
Rust code generation, for the GObject implementations themselves. Thanks to mredlek, this is much cleaner than it was before; now both classes and interfaces share the same code for most of the boilerplate.
GObject Introspection (
.gir) generation, so that language bindings can be generated automatically.
C header files (
.h), so the generated GObjects can be called from C code as usual.
So far, Rust and GIR work; C header files are not generated yet.
Mredlek is a new contributor to gnome-class, but unfortunately was not
able to attend the hackfest. Not only did he rewrite the gnome-class
parser using the new version of syn; he also added support for
passing owned types to GObject methods, such as
Variant. But the biggest thing is probably that mredlek made it a
lot easier to debug the generated Rust source; see the documentation
on debugging for details.
Speaking of which, thanks to Jordan Petridis for making the documentation be published automatically from Gitlab's Continuous Integration pipelines.
Alex Crichton kindly refactored our error propagation code, and even
wrote docs on it! Along with Jordan, they updated the
code for the Rust 2018 edition, and generally wrangled the build
process to conform with the lastest Rust nightlies. Alex also made
code generation a lot faster, by offloading auto-indentation to an
rustfmt process, instead of using it as a crate: using the
rustfmt crate meant that the compiler had a lot more work to do.
During the whole hackfest, Alex was very helpful with Rust questions
in general. While my strategy to see what the compiler does is to
examine the disassembly in gdb, his strategy seems to be to look at
the LLVM intermediate representation instead... OMG.
And we can derive very simple GtkWidgets now!
Saving the best for last... Antoni Boucher, the author of relm, has
been working on making it possible to derive from
this merge request is done, we'll have an example of
gtk::DrawingArea from Rust with very little code.
Normally, the gtk-rs bindings work as a statically-generated binding
for GObject, which really is a type hierarchy defined at runtime. The
static binding really wants to know what is a subclass of what: it
needs to know in advance that
Button's hierarchy is
Button → Bin →
Container → Widget → Object, plus all the
by any of those classes. Antoni has been working on making
gnome-class extract that information automatically from GIR files, so
that the gtk-rs macros that define new types will get all the
There are still bugs in the GIR pipeline that prevent us
from deriving, say, from
gtk::Container, but hopefully these will be
Sebastian Dröge has been refactoring his Rust tools to create GObject subclasses with very idiomatic and refined Rust code. This is now at a state where gnome-class itself could generate that sort of code, instead of generating all the boilerplate from scratch. So, we'll start doing that, and integrating the necessary bits into gtk-rs as well.
Finally, during the last day I took a little break from gnome-class to work on librsvg. Julian Sparber has been updating the code to use new bindings in cairo-rs, and is also adding a new API to fetch an SVG element's geometry precisely.
Oh, boy, I wish the weather had been warmer. The city looks delightful to walk around, especially in the narrow streets on the hills. Can't wait to see it in summer during GUADEC.
Special thanks to Jordan Petridis for being on top of everything build-wise all the time.