Add perl to nix-shell

[gdeest]

This PR fixes a mostly innocuous error when running tests regarding missing Perl:

[nix-shell:~/code/rules_haskell]$ bazel test tests/haskell_test/...
INFO: Analysed 2 targets (0 packages loaded).
INFO: Found 1 target and 1 test target...
haskell_test fired
external/bazel_tools/tools/test/test-setup.sh: line 160: perl: command not found
INFO: Elapsed time: 0.247s, Critical Path: 0.11s
INFO: 1 process: 1 processwrapper-sandbox.
INFO: Build completed successfully, 2 total actions
//tests/haskell_test:haskell_test                                        PASSED in 0.0s

INFO: Build completed successfully, 2 total actions

It seems Bazel expects Perl to be in PATH for minor things, such as regexp-based editing.

@Profpatsch mentioned that we could instead file a patch upstream to remove the dependencies. We could have that discussion here.

[mboes]

Cannot reproduce.

$ nix-shell --pure
[nix-shell:~/src/tweag/rules_haskell]$ bazel test //tests/haskell_test/...
/nix/store/rxl8zia74r3sz9ndy624shhp2kfzxh0a-all-haskell-packages.bzl
/nix/store/jzl6qh6wrvzwfrgw6i6jra17j1bdw9k4-bazel-cc-toolchain
INFO: Build options have changed, discarding analysis cache.
/nix/store/whh4d9yplnpm7pykmhb7v1fn3y63v9rp-glibc-locales-2.27
/nix/store/npy34yz3g7fnxrjh75kv5xxjk3cihsdi-base-bazel
/nix/store/8ybvb7g8b1g7as2jsafxhrl3qs8r35rp-ghc-prim-bazel
/nix/store/gmlsj5ia6yzjxf71hvm665y83c9wq330-rts-bazel
/nix/store/wg5py8hw4pfff2hq0xva4bm3vwwhy4yx-ghc-bazel
/nix/store/6imipp19z62ybv3xpr5wqgrpampdsgzd-c2hs-bazel
/nix/store/ciicimzyc3gy792h34lijgrv82z3cgwv-integer-gmp-bazel
INFO: Analysed 2 targets (32 packages loaded).
INFO: Found 1 target and 1 test target...
INFO: Elapsed time: 38.687s, Critical Path: 0.08s
INFO: 1 process: 1 processwrapper-sandbox.
INFO: Build completed successfully, 2 total actions
//tests/haskell_test:haskell_test                                        PASSED in 0.1s

INFO: Build completed successfully, 2 total actions

On latest master.

[gdeest]

I still witness the issue on latest master. @regnat - I think you witnessed it as well ?

[Profpatsch]

This regex is used twice in the bazel_tools to escape the content of a CDATA field for test output:

# Unused if EXPERIMENTAL_SPLIT_XML_GENERATION is set.
function encode_output_file {
  if [ -f "$1" ]; then
    # Replace invalid XML characters and invalid sequence in CDATA
    # cf. https://stackoverflow.com/a/7774512/4717701
    perl -CSDA -pe's/[^\x9\xA\xD\x20-\x{D7FF}\x{E000}-\x{FFFD}\x{10000}-\x{10FFFF}]+/?/g;' "$1" \
      | sed 's|]]>|]]>]]<![CDATA[>|g'
  fi
}

According to https://stackoverflow.com/questions/7772430/how-to-remove-invalid-characters-from-an-xml-file-using-sed-or-perl/7774512#7774512 it can be replaced with a sed script.

As far as I saw that is also the last remaining usage of perl in bazel for now.

[Profpatsch]

We can remove it again once bazelbuild/bazel#4691 is in bazel.