use ape awk to replace gawk

[comicfans]

this allow creating sysroot that directly referenced by cc toolchain.

in my test I have following setup

apt.install(
    name = "jammy_x86_64_sysroot",
    lock = "//sysroot_creator:jammy_x86_64.lock.json",
    manifest = "//sysroot_creator:jammy_x86_64.yaml",
)
use_repo(apt, "jammy_x86_64_sysroot")

and

def _unpack_impl(ctx):
    out_dir = ctx.actions.declare_directory(ctx.label.name + "_unpacked")

    ctx.actions.run_shell(
        inputs = [ctx.file.tar],
        outputs = [out_dir],
        command = """ tar -xf {src} -C {out}""".format(src = ctx.file.tar.path, out = out_dir.path)
    )

    return [DefaultInfo(files = depset([out_dir]))]

unpack = rule(
    implementation = _unpack_impl,
    attrs = {
        "tar": attr.label(mandatory = True, allow_single_file = True)
    }
)

...

unpack(
    name = "jammy_unpack",
    tar = "@jammy_x86_64_sysroot//:jammy_x86_64_sysroot",  # or @jammy_x86_64_sysroot//:flat
    visibility = ["//visibility:public"]
)

# and jammy_unpack is used as cc toolchain sysroot

#124 (comment) mentioned that gawk is only used by flatten, however directly using @jammy_x86_64_sysroot//:jammy_x86_64_sysroot or @jammy_x86_64_sysroot//:flat as my unpack rule input trigger similar errors:

     @@rules_distroless++apt+jammy_x86_64_sysroot//:jammy_x86_64_sysroot (5c3ac65ce1c8329cc5f7202959c1aceda00db4bbc8a0148931c8e5ed92b61cf2)
    @@rules_distroless++apt+jammy_x86_64_sysroot//:jammy_x86_64_sysroot (fcc255dcd92f8a7c8464439d77cfb04da9617fbc87079de9a2f3462df62874c2)
    @@rules_distroless++apt+jammy_x86_64_sysroot//:dpkg_status (fcc255dcd92f8a7c8464439d77cfb04da9617fbc87079de9a2f3462df62874c2)
.-> @@gawk+//:gawk (38c6768e959f095860edbca0f0d106bbb7ba105ca378e880c8947a6ed3a0b464)
|   @@bazel_tools//tools/cpp:current_cc_toolchain (38c6768e959f095860edbca0f0d106bbb7ba105ca378e880c8947a6ed3a0b464)
|   @@rules_cc+//cc:current_cc_toolchain (38c6768e959f095860edbca0f0d106bbb7ba105ca378e880c8947a6ed3a0b464)
|   @@toolchains_llvm++llvm+llvm_toolchain//:cc-clang-x86_64-linux (38c6768e959f095860edbca0f0d106bbb7ba105ca378e880c8947a6ed3a0b464)
|   @@toolchains_llvm++llvm+llvm_toolchain//:module-x86_64-linux (38c6768e959f095860edbca0f0d106bbb7ba105ca378e880c8947a6ed3a0b464)
|   @@toolchains_llvm++llvm+llvm_toolchain//:include-components-x86_64-linux (38c6768e959f095860edbca0f0d106bbb7ba105ca378e880c8947a6ed3a0b464)
|   @@toolchains_llvm++llvm+llvm_toolchain//:sysroot-components-x86_64-linux (38c6768e959f095860edbca0f0d106bbb7ba105ca378e880c8947a6ed3a0b464)
|   //sysroot_creator:jammy_unpack (38c6768e959f095860edbca0f0d106bbb7ba105ca378e880c8947a6ed3a0b464)
|   @@rules_distroless++apt+jammy_x86_64_sysroot//:flat (38c6768e959f095860edbca0f0d106bbb7ba105ca378e880c8947a6ed3a0b464)
`-- @@gawk+//:gawk (38c6768e959f095860edbca0f0d106bbb7ba105ca378e880c8947a6ed3a0b464)

[thesayyn]

hmm whats the runtime cost here?

[comicfans]

I use script in flatten.sh, input mtree is created by duplicate same file list twice, contains 5553536 lines, got following result:

ape              53.77s user 14.54s system 99% cpu 1:08.32 total
fastbuild_gawk   52.43s user 15.15s system 99% cpu 1:07.69 total
opt_gawk         33.76s user 13.70s system 99% cpu 47.631 total

ape is prebuilt so no differences between compilation_mode

[LittleHuba]

@thesayyn for my work regarding sysroot support with rules_distroless I'd like to know if you prefer:

1. this approach, or
2. only shell commands and no awk at all.

Edit:
There is a third option:

3. Leave all rules that are not used for sysroot support as they are and duplicate necessary logic in sysroot support without awk.

[LittleHuba]

Unfortunate news:
ape awk requires a build action for conversion of the cosmopolitan binary into a native binary. This makes it unusable as part of a repository rule. So from what I can see this is not a viable path for sysroot creation.