Behavior on hardlinks different from tar

[hasufell]

Some GHC bindists have a weird quirk:

-rwxr-xr-x root/root 4642048 2020-07-20 17:24 ghc-8.0.2/utils/haddock/dist/build/tmp/haddock
hrwxr-xr-x root/root 0 2020-07-20 17:24 ghc-8.0.2/utils/haddock/dist/build/tmp/haddock link to ghc-8.0.2/utils/haddock/dist/build/tmp/haddock

The cli tar will unpack this somehow as one file... my haskell-tar fork tar-bytestring also manages to unpack this. libarchive will not unpack this file at all, leaving us with an empty directory.

I'm wondering if this is a bug or expected behavior.

libarchive/libarchive#1381

[vmchale]

Oh dear. Hm. I wonder what Schily tar does; I'll have to dig into it.

[vmchale]

Looks like Schily tar does create the file but it has

star: current 'kexec-tools-2.0.17/purgatory/arch/i386/entry32-16.S' newer.

...i.e. it doesn't extract if the on-disk entry is newer. Hm :o

vanessa@vanessa-desktop /tmp 🌸 exa --all kexec-tools-2.0.17/purgatory/arch/i386/entry32-16.S
kexec-tools-2.0.17/purgatory/arch/i386/entry32-16.S
vanessa@vanessa-desktop /tmp 🌸 readlink -f kexec-tools-2.0.17/purgatory/arch/i386/entry32-16.S
/tmp/kexec-tools-2.0.17/purgatory/arch/i386/entry32-16.S

[vmchale]

Oh, and the Rust tar library core dumps :p

vanessa@vanessa-desktop /tmp 🌸 ~/.cargo/bin/star x kexec-tools-2.0.17.tar.xz 
thread 'main' panicked at 'called `Result::unwrap()` on an `Err` value: Custom { kind: AlreadyExists, error: TarError { desc: "failed to unpack `/tmp/kexec-tools-2.0.17/purgatory/arch/i386/entry32-16.S`", io: Custom { kind: AlreadyExists, error: "File exists (os error 17) when hard linking /tmp/kexec-tools-2.0.17/purgatory/arch/i386/entry32-16.S to /tmp/kexec-tools-2.0.17/purgatory/arch/i386/entry32-16.S" } } }', /home/vanessa/.cargo/registry/src/github.com-1ecc6299db9ec823/star-0.1.2/src/main.rs:271:5
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
Aborted (core dumped)

[vmchale]

The go tar library seems to skip with a message...

vanessa@vanessa-desktop /tmp 🌸 arc unarchive ghc-8.0.2-x86_64-deb8-linux.tar.xz 
2020/07/20 16:33:34 [ERROR] Reading file in tar archive: file already exists: ghc-8.0.2/utils/haddock/dist/build/tmp/haddock
2020/07/20 16:33:34 [ERROR] Reading file in tar archive: file already exists: ghc-8.0.2/utils/hsc2hs/dist-install/build/tmp/hsc2hs
2020/07/20 16:33:34 [ERROR] Reading file in tar archive: file already exists: ghc-8.0.2/utils/ghc-pkg/dist-install/build/tmp/ghc-pkg
2020/07/20 16:33:34 [ERROR] Reading file in tar archive: file already exists: ghc-8.0.2/utils/hpc/dist-install/build/tmp/hpc
2020/07/20 16:33:34 [ERROR] Reading file in tar archive: file already exists: ghc-8.0.2/utils/runghc/dist-install/build/tmp/runghc
2020/07/20 16:33:34 [ERROR] Reading file in tar archive: file already exists: ghc-8.0.2/ghc/stage2/build/tmp/ghc-stage2

vanessa@vanessa-desktop /tmp 🌸 exa --all ghc-8.0.2/utils/haddock/dist/build/tmp/haddock
ghc-8.0.2/utils/haddock/dist/build/tmp/haddock
vanessa@vanessa-desktop /tmp 🌸 readlink -f ghc-8.0.2/utils/haddock/dist/build/tmp/haddock
/tmp/ghc-8.0.2/utils/haddock/dist/build/tmp/haddock

[hasufell]

Seems this is fixed by the latest sources:

libarchive/c/archive_write_disk_posix.c

Line 594 in a78dd47

/*