Merge #18779: doc: Better explain GNU ld's dislike of ld64's options

cd24f37ea9 doc: Better explain GNU ld's dislike of ld64's options (fanquake)

Pull request description:

  There's also now more than a single option being special cased for
  darwin. If we didn't special case these options they would still end
  up on the link line and the binaries produced would just segfault.

  I'm going to plug #17874 here as well, because adding
  `-fatal-warnings` to our `AX_CHECK_LINK_FLAG` calls would
  mostly prevent this sort of option mangling from happening.

  An example of the warning behaviour:
  ```bash
  echo "int main() {}" | g++ -x c++ -std=c++11 -Wl,-dead_strip -
  /usr/bin/ld: warning: cannot find entry symbol ad_strip; defaulting to 0000000000001040

  nm -C a.out
  0000000000001000 t _init
  0000000000001040 T _start
                   U ad_strip
  ```

ACKs for top commit:
  dongcarl:
    ACK cd24f37ea9

Tree-SHA512: 8c5ff11b647e7d44dbb3f509a07caf8606a6b481c114403f0de72b3ad65395dbe9a3436e731ae1b46a823431ed23c3c6aacab8942d78629d59cd8c258c5dbf02
This commit is contained in:
fanquake 2020-04-28 17:19:00 +08:00
commit 9fb95ae8e3
No known key found for this signature in database
GPG key ID: 2EEB9F5CC09526C1

View file

@ -757,7 +757,9 @@ if test x$use_hardening != xno; then
esac
fi
dnl this flag screws up non-darwin gcc even when the check fails. special-case it.
dnl These flags are specific to ld64, and may cause issues with other linkers.
dnl For example: GNU ld will intepret -dead_strip as -de and then try and use
dnl "ad_strip" as the symbol for the entry point.
if test x$TARGET_OS = xdarwin; then
AX_CHECK_LINK_FLAG([[-Wl,-dead_strip]], [LDFLAGS="$LDFLAGS -Wl,-dead_strip"])
AX_CHECK_LINK_FLAG([[-Wl,-dead_strip_dylibs]], [LDFLAGS="$LDFLAGS -Wl,-dead_strip_dylibs"])