Expose set_process_dispatcher to capi

[sensille]

In order to be able to add debuginfod support to bpftrace, it looks like the c api has to expose set_process_dispatcher(). Maybe a more specific solution would also do.

[d-e-s-o]

The generic process dispatcher facility is a Rust-only API. I consider it pretty much impossible to express it in C, as it requires exposure of the Resolve trait, ElfResolver type, etc in some form. My recommendation would be to interface with the Rust API.

Short of that, the one thing we could alternatively do is have a specific instance of this API that basically only triggers for the ProcessMemberType::Path variant, invokes a user provided C callback with the path, and instantiates an ELF resolver for the potentially returned path. This way, users would be given control over the following bits:

blazesym/examples/sym-debuginfod/main.rs

Lines 132 to 150 in 96156f7

	let build_id = if let Some(build_id) = read_elf_build_id(&path.maps_file)? {
	build_id
	} else {
	// The binary does not contain a build ID, so we cannot
	// retrieve symbol data. Just let the default resolver do
	// its thing.
	return Ok(None)
	};
	let path = if let Some(path) = client
	.fetch_debug_info(&BuildId::raw(build_id))
	.map_err(Box::from)?
	{
	path
	} else {
	// If we were unable to find debug information for the provided
	// build ID we let the default resolver see what it can do.
	return Ok(None)
	};

In effect it would allow for path to path translation, assuming the target path represents an ELF file.

[sensille]

That sounds great. I think I can add this for testing myself for now, so once the rest works, I'll let you know.