Skylark ctx.command doesn't incorporate runfiles from input executables; py_binary/java_binary executables fail

[dfabulich]

(I forked this bug off of issue #1136, because #1136 turned out to be a mix of several bugs.)

Consider this repository https://github.com/dfabulich/cannot-find-runfiles

There's a simple python script in build_tools/whatevah.py that accepts one CLI argument and writes a file to that location.

import sys

with open (sys.argv[1], "w") as OUT:
    OUT.write("hello world")

It's declared as a simple py_binary in build_tools/BUILD.

py_binary(
    name='whatevah',
    srcs=['whatevah.py'],
    visibility=['//visibility:public'],
)

There's a simple Skylark rule in build_tools/example.bzl; the example rule runs whatevah to generate a %{name}.txt file as output, but it uses the command parameter to ctx.action instead of the executable parameter.

def _impl(ctx):
    ctx.action(
        inputs = [ctx.executable._whatevah],
        outputs = [ctx.outputs.text],
        command = "{} {}".format(ctx.executable._whatevah.path, ctx.outputs.text.path),
    )

example = rule(
    implementation = _impl,
    attrs = {
        "_whatevah": attr.label(executable=True, allow_files=True, default=Label("//build_tools:whatevah"))
    },
    outputs = {
        "text": "%{name}.txt",
    }
)

Finally, foo/BUILD declares a foo target with the example rule.

load('/build_tools/example', 'example')

example(
    name='foo',
)

When I bazel build foo, I expect it to succeed and generate foo.txt, but instead I get this error:

$ bazel build foo --verbose_failures
INFO: Found 1 target...
ERROR: /private/tmp/repro/foo/BUILD:3:1: error executing shell command: 'bazel-out/local_darwin-fastbuild/bin/build_tools/whatevah bazel-out/local_darwin-fastbuild/bin/foo/foo.txt' failed: bash failed: error executing command
  (cd /private/var/tmp/_bazel_dfabulich/92410d0e3373f265902564b48493ac12/repro && \
  exec env - \
  /bin/bash -c 'bazel-out/local_darwin-fastbuild/bin/build_tools/whatevah bazel-out/local_darwin-fastbuild/bin/foo/foo.txt'): com.google.devtools.build.lib.shell.BadExitStatusException: Process exited with status 1.
Traceback (most recent call last):
  File "bazel-out/local_darwin-fastbuild/bin/build_tools/whatevah", line 110, in <module>
    Main()
  File "bazel-out/local_darwin-fastbuild/bin/build_tools/whatevah", line 66, in Main
    sys.argv[0])
AssertionError: Cannot find .runfiles directory for bazel-out/local_darwin-fastbuild/bin/build_tools/whatevah
Target //foo:foo failed to build
INFO: Elapsed time: 0.113s, Critical Path: 0.03s

As discussed in issue #1136, you can workaround this issue by using the executable parameter to ctx.action instead of command.

def _impl(ctx):
    ctx.action(
        executable = ctx.executable._whatevah,
        arguments = [ctx.outputs.text.path],
        outputs = [ctx.outputs.text],
    )

[dfabulich]

Beware, if you try to workaround this issue #1192 by converting your ctx.action's command into a sh_binary, you'll probably also hit issue #1147, in which sh_binary scripts have an incorrect working directory when run via Skylark.

[umap341]

Hi,

Is there a way now to invoke py_binary through skylark rule?

Thanks,

[comius]

This issue is pretty old and API has changed a lot.
ctx.actions.run incorporates runfiles when somebody uses ctx.executable.xxx in it. So I believe this is solved.