Feature request: using custom target output as include directory

[fghzxm]

I have to use the output of a directory-producing custom_target() as an include directory to other targets.

---

Detailed situation: I want to use the cppwinrt tool in my project, which generates multiple header files into a directory. The number and names of the generated files cannot be predetermined, as they depend on the actual contents of the input files fed to the tool as well as the particular version of the Windows SDK being used.

I have set up a subproject that has a custom_target() executing the tool and taking its output directory as the target output, and a dependency declaration that specifies the target output as an include directory:

cppwinrt = find_program(...)

headers = custom_target('cppwinrt_headers',
  command : [cppwinrt, '-output', '@OUTPUT@', ...],
  output : 'headers_dir') # output is a directory

cppwinrt_headers_dep = declare_dependency(
  include_directories : include_directory(headers))

Meson doesn't like it:

meson.build:33:0: ERROR: Arguments must be strings.

Is there any possibility this might become supported?

[dcbaker]

This would be rather difficult to implement reliably. ninja especially depends on knowing what inputs map to which outputs so that it knows what to regenerate (using deps we can ensure that any file that imports a header generated by a source is rebuilt when appropriate. You could likely already do this with a little bit of scripting, but you're introducing unreliability into your build. I can't find any documentation of all of the cli options for cppwinrt, so I don't know if we could coerce it to do the right thing, but having 0 control over the output file names is pretty rude. This will be equally problematic for other build systems like cmake.