gcc-arm-embedded downloads a precompiled binary for linux only

[copumpkin]

This tripped up someone trying to use it on Darwin on IRC today. Is it not possible for it to use our extensive existing gcc machinery, which already works on Darwin?

cc @bjornfor

[bjornfor]

Why not have both?

Nix still does not dominate the world, so I think it is nice to be able to use a toolchain on nix that is readily available on non-nix systems.

I'd love better cross compiler/ing support in nix though.

[cleverca22]

when i was helping stepcut on irc i got things worked down to just this:
http://pastebin.com/YmJPSDzc + nix-build default.nix -A gccCrossStageStatic
but that only works on linux based nix, the darwin nix uses an incompatible binutils

[stepcut]

I made some additional progress. There are two sets of problems -- things that are nix specific, and things that are related to building a GCC cross-compiler with clang.

These are the patches I have to nixpkgs files so far:

[nix-shell:~/nixpkgs]$ git diff
git diff
diff --git a/pkgs/development/compilers/gcc/4.9/default.nix b/pkgs/development/compilers/gcc/4.9/default.nix
index fecdd85..d50b618 100644
--- a/pkgs/development/compilers/gcc/4.9/default.nix
+++ b/pkgs/development/compilers/gcc/4.9/default.nix
@@ -303,8 +303,9 @@ stdenv.mkDerivation ({
     if SDKROOT=$(/usr/bin/xcrun --show-sdk-path); then
       configureFlagsArray+=(--with-native-system-header-dir=$SDKROOT/usr/include)
       makeFlagsArray+=( \
-       CFLAGS_FOR_BUILD=-F$SDKROOT/System/Library/Frameworks \
-       CFLAGS_FOR_TARGET=-F$SDKROOT/System/Library/Frameworks \
+       CXXFLAGS="-g -O2 -fbracket-depth=1024" \
+       CFLAGS_FOR_BUILD="-F$SDKROOT/System/Library/Frameworks" \
+       CFLAGS_FOR_TARGET="-F$SDKROOT/System/Library/Frameworks" \
        FLAGS_FOR_TARGET=-F$SDKROOT/System/Library/Frameworks \
       )
     fi
@@ -398,6 +399,7 @@ stdenv.mkDerivation ({
     CC_FOR_TARGET = "${stdenv.cross.config}-gcc";
     NM_FOR_TARGET = "${stdenv.cross.config}-nm";
     CXX_FOR_TARGET = "${stdenv.cross.config}-g++";
+    CXXFLAGS_FOR_TARGET="-g -O2 -fbracket-depth=1024";
     # If we are making a cross compiler, cross != null
     NIX_CC_CROSS = if cross == null then "${stdenv.ccCross}" else "";
     dontStrip = true;
diff --git a/pkgs/development/compilers/gcc/builder.sh b/pkgs/development/compilers/gcc/builder.sh
index dd77c67..9e59d30 100644
--- a/pkgs/development/compilers/gcc/builder.sh
+++ b/pkgs/development/compilers/gcc/builder.sh
@@ -24,7 +24,11 @@ if test "$noSysDirs" = "1"; then
         # Figure out what extra flags to pass to the gcc compilers
         # being generated to make sure that they use our glibc.
         extraFlags="$(cat $NIX_CC/nix-support/libc-cflags)"
-        extraLDFlags="$(cat $NIX_CC/nix-support/libc-ldflags) $(cat $NIX_CC/nix-support/libc-ldflags-before)"
+        if [ -e $NIX_CC/nix-support/libc-ldflags-before ]; then
+            extraLDFlags="$(cat $NIX_CC/nix-support/libc-ldflags) $(cat $NIX_CC/nix-support/libc-ldflags-before)"
+        else
+            extraLDFlags="$(cat $NIX_CC/nix-support/libc-ldflags)"
+        fi

         # Use *real* header files, otherwise a limits.h is generated
         # that does not include Glibc's limits.h (notably missing
diff --git a/pkgs/top-level/all-packages.nix b/pkgs/top-level/all-packages.nix
index 5917e71..621cee8 100644
--- a/pkgs/top-level/all-packages.nix
+++ b/pkgs/top-level/all-packages.nix
@@ -5469,7 +5469,7 @@ let

   binutilsCross = assert crossSystem != null; lowPrio (forceNativeDrv (
     if crossSystem.libc == "libSystem" then darwin.cctools_cross
-    else binutils.override {
+    else binutils_nogold.override {
       noSysDirs = true;
       cross = crossSystem;
     }));

However this patch is also required:

https://sourceware.org/bugzilla/show_bug.cgi?id=19222

I have not yet applied that patch so I am not sure what other issues lurk. It is possible that is the last issue though.

[dvc94ch]

building cross compilers is possible now, is this issue still relevant? or can we close this issue after #15867?

[matthewbauer]

Closing (see #39087)