I'm building a simple program for ARM. GPRbuild outputs the following:
gprbuild -p -P avocado_test.gpr (in directory: /home/rodeo/Projects/AvocadoTest)
Bind
[gprbind] run_avocado_test.bexch
[Ada] run_avocado_test.ali
Link
[link] run_avocado_test.adb
/opt/GNAT/arm-elf/bin/arm-eabi-ld: unrecognised emulation mode: cpu=cortex-m3
Supported emulations: armelf
gprbuild: link of run_avocado_test.adb failed
Compilation failed.
From what I could find online, ld has a switch -m which specifies an emulation mode. However, I am not using this switch in the linker of my project. I am using the -mcpu=cortex-m3 switch in the compiler, though, and it looks like somehow GPRbuild is passing this switch to the linker as well. Here is my .gpr file:
project Avocado_Test is
for Source_Dirs use ("source");
for Object_Dir use "build";
for Main use ("source/run_avocado_test.adb");
for Target use "arm-eabi";
for Runtime ("Ada") use "ravenscar-sfp-sam3x8e";
package Builder is
for Executable_Suffix use ".elf";
for Switches ("Ada") use ("-j0");
end Builder;
package Compiler is
for Driver ("Ada") use "arm-eabi-gcc";
for Switches ("Ada") use (
"-mthumb",
"-mcpu=cortex-m3",
"-O2");
end Compiler;
package Binder is
for Driver ("Ada") use "arm-eabi-gnatbind";
end Binder;
package Linker is
for Driver use "arm-eabi-ld";
end Linker;
end Avocado_Test;
Why is the linker taking a switch from the compiler? How do I prevent this?
I wouldn’t use Driver at all, in any of your GPR packages. Once gprbuild sees for Target use "arm-eabi"; it knows to prepend arm-eabi- to each tool.
There’s no reason to avoid passing compiler switches to the default linker driver, which is gcc (in this case arm-eabi-gcc), because it knows which of them to pass to the linker.
You might find these interesting from my cortex-gnat-rts project: runtime.xml (but note, details may differ from AdaCore’s runtimes), testbed.gpr.