Generators: support multiple inputs

[gerion0]

It seems not to be possible to define a generator with multiple input files (at least meson complains that only string not list is supported). However, it would be helpful if generators would support that.

My current use case for this is a call to a custom linker, that obviously take a lot of inputs and link them to one file.

In pseudo code:

gen = generator(myprog,
                output  : '@BASENAME0@.o',
                arguments : ['@INPUT@', '-o', '@OUTPUT@'])
gen_src = gen.process([['in1.o', 'in2.o', 'in3.o'], ['inA.o', 'inB.o'])

Note the BASENAME0 to indicate the basename of the first input in the list should be used.

[jpakkane]

Unless I'm misunderstanding, what you want here is a custom_target, not a generator.

[gerion0]

I'm not sure if this fits my use case. I will try to explain my problem a little bit more.

I had a project, where I needed to call a specific linker a lot of times. Say here for example to the files (as in the example in the last post):

out1.o <- in1.o in2.o in3.o
out2.o <- inA.o InB.o

Currently, with custom_target I would write somethink similar to:

link_targets = []
foreach inputlist : [['in1.o', 'in2.o', 'in3.o'], ['inA.o', inB.o']]
  t = custom_target('linktarget' + inputlist[0],
                    input: inputlist,
                    output: inputlist[0] + '.o',
                    command: [mylinker, '@INPUT@', '-o', '@OUTPUT']
  link_targets += t
endforeach

With generators I could write exactly what is mentioned in my first post.

I find the generator solution more elegant because the generator allows me to call it easily from multiple place in my source (i.e. multiple meson build files).

Or did I understand you wrong?

Some meta toughts:
In general, as far as I understand, every generator can be replaced with foreach + custom_target but for me this feels a little bit like a workaround because I am using the false concept. I want a generic recipe but instead call a specific recipe multiple times.
Therefore, since both concepts (generator and custom_target) are kind of similar, but a generator provides a generic recipe and a custom_target is exactly one instance of such a recipe, won't it be possible or meaningful to combine the two concepts?: A custom_target takes a generator instead of the direct command and calls internally the generator with its inputs and outputs:

1. Specifing a recipe for calling an arbitrary external program -> use a generator
2. Make a target out of the recipe, specifing a concrete instance the recipe -> use a custom_target

(Btw: Thank you for writing the book/manual.)

[inklesspen]

I ran into a similar problem lately, when defining a target to produce c source for a Python extension module:

meson_cffi_buildtool = find_program('meson_cffi_buildtool.py')

cairopango_c = custom_target(
    command: [
        meson_cffi_buildtool,
        '--modulename', 'tabula.rendering._cairopango',
        '--cdef', '@INPUT0@',
        '--csrc', '@INPUT1@',
        '--output', '@OUTPUT@',
    ],
    output: '_cairopango.c',
    input: [
        'cffi/tabula/rendering/_cairopango.cdef.txt',
        'cffi/tabula/rendering/_cairopango.csrc.c',
    ],
)

The buildtool reads two input files (they are of different types, and each invocation needs both of them) and generates a c output file.

I know that generator.process can take multiple inputs, but it looks like this just runs the generator program on each of them individually, so that doesn't work for me here.

This one custom_target is not too bad, but I have a second custom_target for a second module, and if my project were larger I could see needing more. And each one of them is exactly the same except for the specific file paths and the modulename argument.