Conan integration ?

[darkrift]

I have been thinking about this idea for quite a while and even started experimenting around this but I was wondering if this project should not be the one that host and provide an integration with Conan.

The current Bazel generator of conan has the major drawback that it produces a generated dependencies file along with the binaries at the moment of the conan install is run instead of being integrated in the Bazel build from a proper ruleset and binaries built when requested.

I feel this would fill a major gap in the use of existing CC libraries with already alot of defined recipes almost removing the need entirely for CC bzlmod modules that is being done around. Those recipes already have alot of configurability and would open up a huge list of libraries and prevent re-inventing the wheels while reducing a major pain to build C/C++ projects with Bazel that requires open source dependencies.

Any thoughts on this ?

[jsharpe]

Happy to have this hated in rules_foreign_cc if someone wants to contribute it! However I don't think it completely removes the need for native CC build files as RBE typically only makes a single cpu core available to each build action by default which limits the speed at which a Conan dep would be able to be built unless it hits a binary prebuilt version. I also think we should be looking at what can be done with CMake dependency providers for better interop between bazel built artifacts and CMake project deps as a bazel CMake sandwich type target is tricky to configure at the moment.

[darkrift]

What I have been thinking was to introspect the recipe and make a target for each of the required dependencies, similar to how conan would compute the graph and build the dependencies before the requested recipe.

There are a bunch of commands to make that introspection (such a conan graph), however I suspect some recipes will be more challenging than others (especially when the recipe does filesystem listing to figure out what libraries were produced by the recipe).

But for a lib with 2 dependencies, that would be 3 targets that could be spreaded to multiple REB nodes. However, this could be done in a later iteration.

It wouldn't use the most of what Bazel could offer (especially at the action level granularity), but could be a good middle ground