Split outputs because there's no point in keeping a reference to Python
and it causes trouble during the Darwin stdenv bootstrap. There's also
an unnecessary dependency on LLVM in libc++ which causes us to rebuild
LLVM several more times than necessary during bootstrap, and an awkward
dependency on XPC in the TSAN that we turn off. This is in preparation
for using LLVM 4 in the Darwin stdenv and by default across nixpkgs.
deriving Yojson is a ppx_deriving plugin that generates JSON serializers and
deserializers that use the Yojson library from an OCaml type definition.
Homepage: https://github.com/whitequark/ppx_deriving_yojson
To reproduce the issue:
```
nix-build -E 'import <nixpkgs> {}' -A python27Packages.spambayes
result/bin/sb_filter.py </dev/null
```
Before this commit, this fails with
```
[...]
File "/nix/store/vlgr1wrxd2wiw5ay5ch54365dzqvcwgq-python2.7-spambayes-1.1b1/lib/python2.7/site-packages/spambayes/dbmstorage.py", line 20, in open_dbhash
return bsddb.hashopen(*args)
AttributeError: 'NoneType' object has no attribute 'hashopen'
```
After this commit, it doesn't.
The wrapper for Atom was loading libraries via LD_PRELOAD, for example
libxkbfile. Now, if you installed atom via nix-env and happened to use a newer
nixpkgs for that than what your system environment is build against, you could
end up with an error like this:
```
uname: relocation error:
/nix/store/68sa3m89shpfaqq1b9xp5p1360vqhwx6-glibc-2.25/lib/libdl.so.2:
symbol _dl_catch_error, version GLIBC_PRIVATE not defined in file libc.so.6
with link time reference
```
This happens because atom calls the `uname` executable from the system to
determine the platform. Because that inherits the `LD_PRELOAD` environment
variable, so the libxkbfile library that the `atom` wrapper was build against
is loaded into `uname`. But since `atom` comes from `nix-env`, the `libxkbfile`
it was built with might be compiled against a newer version of `glibc` than
`uname`, which comes from the system, was! Having two versions of glibc loaded
into the same processes results in chaos.
To fix this, we avoid setting `LD_PRELOAD` and instead use patchelf to set the
correct RPATH. RPATH is not inherited by child processes, so the above issue
can no longer occur.
The only small complication here is that the library that actually loads
libxkbfile is not the atom binary itself, but a node extension that atom uses.
So instead of setting the RPATH on `atom` only, we also set the `rpath` on all
node extensions (`*.node`) the output.
