pkg_rpm: Specify _topdir for install stanza

[kellyma2]

If we don't specify the _topdir as part of the install-from path we can't seem to find the correct files to install when using a set of rules like:

pkg_filegroup(
    name = "my_filegroup",
    srcs = [
      "//some/label:foo",
    ],
    prefixs = "/usr/bin",
)

pkg_rpm(
    name = "rpm",
    ...
    srcs = [
        ":my_filegroup",
    ],
)

Adding %{{_topdir}/BUILD/ as a prefix to the source path seems to solve this problem.

[aiuto]

Can you add a test for this and explain more clearly what is failing?
The RPM builder has been working for a long time, and some test input is from pkg_filegroup, so it is surprising that the code is so for off.

[kellyma2]

Can you add a test for this and explain more clearly what is failing? The RPM builder has been working for a long time, and some test input is from pkg_filegroup, so it is surprising that the code is so for off.

I'll try to rig up a test for this tomorrow and also provide more details. The tl;dr was that it couldn't find the expected inputs in the relative path, but putting the prefix in place means it becomes an absolute path and can be resolved.

[kellyma2]

Can you add a test for this and explain more clearly what is failing? The RPM builder has been working for a long time, and some test input is from pkg_filegroup, so it is surprising that the code is so for off.

Still working through adding a test for this, but figured I'd provide a bit more detail as to what the failure (and my rules) look like. I'm using https://github.com/bazelbuild/examples/tree/main/cpp-tutorial/stage1 as a simple scaffold for verifying the failure. My rules are as follows:

• main/BUILD:

load("@rules_pkg//pkg:mappings.bzl", "pkg_files", "pkg_attributes")

pkg_files(
    name = "exe",
    srcs = [
        ":hello-world",
    ],
    visibility = ["//visibility:public"],
    attributes = pkg_attributes(
        mode = "0755",
        user = "root",
        group = "root",
    ),
)

Root BUILD:

load("@rules_pkg//pkg:mappings.bzl", "pkg_filegroup")
load("@rules_pkg//pkg:rpm.bzl", "pkg_rpm")

pkg_filegroup(
    name = "rpm_filegroup",
    srcs = [
        "//main:exe",
    ],
    prefix = "/usr/bin",
)

pkg_rpm(
    name = "rpm",
    package_name = "example",
    license = "bar",
    summary = "baz",
    version = "1.0.0",
    release = "foo",
    description = "bam",
    architecture = "x86_64",
    srcs = [
        ":rpm_filegroup",
    ],
)

in WORKSPACE I'm configuring things with:

load("@rules_pkg//toolchains/rpm:rpmbuild_configure.bzl", "find_system_rpmbuild")

find_system_rpmbuild(name="rule_pkg_rpmbuild")

Where @rules_pkg is just pointing at a fresh clone of rules_pkg in my local worktree.

When I run bazel build rpm the output I get is:

> bazel build rpm
Loading: 
Loading: 
Loading: 0 packages loaded
Analyzing: target //:rpm (1 packages loaded, 0 targets configured)
INFO: Analyzed target //:rpm (2 packages loaded, 5 targets configured).
INFO: Found 1 target...
bazel: Entering directory `/tmp/mkelly/bazel/execroot/__main__/'
[0 / 8] [Prepa] BazelWorkspaceStatusAction stable-status.txt
INFO: From MakeRpm example-1.0.0-foo.x86_64.rpm:
Error calling rpmbuild:
Executing(%install): /bin/sh -e /tmp/tmp2s0xmxiy/TMP/rpm-tmp.u26xbD
+ umask 022
+ cd /tmp/tmp2s0xmxiy/bld-x86_64/example
+ '[' /tmp/tmp2s0xmxiy/BUILDROOT '!=' / ']'
+ rm -rf /tmp/tmp2s0xmxiy/BUILDROOT
++ dirname /tmp/tmp2s0xmxiy/BUILDROOT
+ mkdir -p /tmp/tmp2s0xmxiy
+ mkdir /tmp/tmp2s0xmxiy/BUILDROOT
++ dirname /usr/bin/hello-world
+ install -d /tmp/tmp2s0xmxiy/BUILDROOT//usr/bin
+ cp bazel-out/k8-fastbuild/bin/main/hello-world /tmp/tmp2s0xmxiy/BUILDROOT//usr/bin/hello-world
cp: cannot stat 'bazel-out/k8-fastbuild/bin/main/hello-world': No such file or directory
error: Bad exit status from /tmp/tmp2s0xmxiy/TMP/rpm-tmp.u26xbD (%install)


RPM build errors:
    Bad exit status from /tmp/tmp2s0xmxiy/TMP/rpm-tmp.u26xbD (%install)

No RPM file created.
ERROR: /tmp/test/examples/cpp-tutorial/stage1/BUILD:12:8: output 'example-1.0.0-foo.x86_64.rpm' was not created
ERROR: /tmp/test/examples/cpp-tutorial/stage1/BUILD:12:8: MakeRpm example-1.0.0-foo.x86_64.rpm failed: not all outputs were created or valid
bazel: Leaving directory `/tmp/mkelly/bazel/execroot/__main__/'
Target //:rpm failed to build
Use --verbose_failures to see the command lines of failed build steps.
INFO: Elapsed time: 0.255s, Critical Path: 0.17s
INFO: 2 processes: 1 internal, 1 linux-sandbox.
FAILED: Build did NOT complete successfully

[aiuto]

• cp bazel-out/k8-fastbuild/bin/main/hello-world /tmp/tmp2s0xmxiy/BUILDROOT//usr/bin/hello-world
  cp: cannot stat 'bazel-out/k8-fastbuild/bin/main/hello-world': No such file or directory
  I'm scratching my head a bit on this. That should be the path to hello-world.

[kellyma2]

• cp bazel-out/k8-fastbuild/bin/main/hello-world /tmp/tmp2s0xmxiy/BUILDROOT//usr/bin/hello-world
  cp: cannot stat 'bazel-out/k8-fastbuild/bin/main/hello-world': No such file or directory
  I'm scratching my head a bit on this. That should be the path to hello-world.

That would be my expectation and with this PR it actually correctly points to it because the path ceases to be relative. When you take the example repo and apply my BUILD + WORKSPACE setup what do you get?

Also, missed part of my WORKSPACE the correct variant also includes:

local_repository(
    name = "rules_pkg",
    path = "/home/mkelly/repos/bazel/rules_pkg",
)

load("@rules_pkg//:deps.bzl", "rules_pkg_dependencies")

rules_pkg_dependencies()

[aiuto]

Take a look at #770

I tried to prove your change works, but it failed in new and unexpected ways.
Perhaps you do something like that as a test case or example.

[kellyma2]

After a bit of thrashing around with the Fedora docker image it seems like I got this working and it's using rpmbuild 4.18.1 - I'll see if I can upgrade rpmbuild in my build environment to determine if this also happens there.

Secondary to this, should there be a minimum version check for the version of rpmbuild being used?

[cgrindel]

The change looks reasonable. However, I noticed that @aiuto requested a test for this. Were you able to create one? Also, could you ping me again when CI is green?