Inclusion of "*.css.js" css name mapping as the output files of closure_css_binary

[adam-rocska]

Problem Statement

closure/stylesheets/closure_css_binary.bzl declares %{name}.css.js as its output :

outputs = {
        "bin": "%{name}.css",
        "map": "%{name}.css.map",
        "js": "%{name}.css.js",
    },

however, the returned struct's files list was files which did not include ctx.outputs.js :

...
if not css.srcs:
        fail("There are no CSS source files in the transitive closure")
    inputs = []
    files = [ctx.outputs.bin, ctx.outputs.map]
...
return struct(
        files = depset(files),
        closure_css_binary = struct(
            bin = ctx.outputs.bin,
...

By changing depset(files) to depset(output) we have the artifact included when it's built.

My specific use case for which this is a solution

I have a Macro for a genrule which takes :js_bin, :css_bin, etc. as input, and creates a copy of the input files with the exact same extensions, but with the name provided to the macro.

An example renaming :

expose_as(
    name = "options",
    srcs = [
       ":js_bin",
       ":css_bin",
       ":html_bin"
    ],
    visibility = BINARY_BUILD_VISIBILITY
)

Original name	Renamed / Copied name
js_bin.js	options.js
js_bin.js.map	options.js.map
css_bin.css	options.css
css_bin.css.map	options.css.map
css_bin.css.js	options.css.js
...	...

Since %{name}.css.js wasn't part of the actual outputs, it failed. With this fix it works, and to the best of my knowledge it shouldn't break any existing behavior, and isn't a flawed expression of intent either.

If it's problematic, sorry for the inconvenience, and I'll attempt to find a work-around.

[adam-rocska]

@alexeagle : Good point! Sure.

I'll come back to this problem in a few days, and will figure out some test. At first thought it smells tricky, but will worth the effort.

[gkdn]

There hasn't been any activity in this PR for very long time. Please let me know if there are any intentions to pursue this change. Otherwise I will close it. Thanks.