Support cross-compilation via -target ... option

[umbertov]

Hi, as discussed in the issue #71 , I propose the following two changes:

• Support the -target ARCHITECTURE in shared/parser.go
• Default to llvm-objcopy because that's what comes with the LLVM toolchain, which supports any target. This changes the pre-existing default of objcopy from the GNU toolchain, which only supports the native compilation target (i.e. x86 on x86 machines, arm on arm machines).

If someone still wishes to use GNU objcopy, they can still use it by setting the GLLVM_OBJCOPY=objcopy environment variable for native targets, or the prefixed version for cross-compilation toolchains, e.g. GLLVM_OBJCOPY=aarch64-linux-gnu-objcopy for arm64.

[ianamason]

Got any idea why the tests are failing?

[ianamason]

I think objcopy should remain the default, ONLY because it is a good idea to maintain backward compatibility. Then you won't need to alter the documentation. Maybe also this is the cause of the test failure?

[umbertov]

Hi, I noticed the tests failing. I'm heading back home now, but tomorrow morning I'll edit the PR! Il mer 23 ago 2023, 17:12 Ian A Mason ***@***.***> ha scritto:

…

I think objcopy should remain the default, ONLY because it is a good idea to maintain backward compatibility. Then you won't need to alter the documentation. Maybe also this is the cause of the test failure? — Reply to this email directly, view it on GitHub <#72 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AC6ICSASIZ7OE6XKHWT4XWLXWYMW3ANCNFSM6AAAAAA33NFSZI> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[umbertov]

Hi, I just implemented what you mentioned in the issue #71:

• removed the llvm-objcopy default (dirty old force push method)
• added the ^--target=.*$ pattern to support --target=ARCH flag in addition to -target ARCH
• wrote a bunch of lines in the README.md to document the use of GLLVM_OBJCOPY when cross-compiling

Let me know if everything looks fine to you, and thanks for all your help along the way!

[ianamason]

Thanks @umbertov I’ll merge it in when I get home. Thanks again.

[umbertov]

hey @ianamason github noticed me that a test is failing; specifically, Test_file_type.

I tried reproducing it on my machine with go test -v ./tests, and it worked on my machine ™️

gllvm_cross git:(master-rename) go test -v ./tests

=== RUN   Test_basic_functionality
Loading '/home/umberto/dev/gllvm_cross/tests/.helloworld.c.o.bc'
Linking in '/home/umberto/dev/gllvm_cross/tests/.helloworld.c.o.bc'
Writing bitcode...
Bitcode file extracted to: ../data/hello.bc.
--- PASS: Test_basic_functionality (0.27s)
=== RUN   Test_more_functionality
Loading '/home/umberto/dev/gllvm_cross/data/.bhello.o.bc'
Linking in '/home/umberto/dev/gllvm_cross/data/.bhello.o.bc'
Writing bitcode...
Bitcode file extracted to: ../data/bhello.bc.
--- PASS: Test_more_functionality (0.05s)
=== RUN   Test_obscure_functionality
Loading '/home/umberto/dev/gllvm_cross/data/.bhello.notanextensionthatwerecognize.bc'
Linking in '/home/umberto/dev/gllvm_cross/data/.bhello.notanextensionthatwerecognize.bc'
Writing bitcode...
Bitcode file extracted to: ../data/bhello.bc.
--- PASS: Test_obscure_functionality (0.08s)
=== RUN   Test_file_type
--- PASS: Test_file_type (0.00s)
=== RUN   Test_env_and_args
--- PASS: Test_env_and_args (0.01s)
=== RUN   Test_parsing
--- PASS: Test_parsing (0.02s)
PASS
ok  	github.com/umbertov/gllvm/tests	0.427s

[ianamason]

I'll look into it once I have merged it in. The current, premerge, fails on my M1 so I guess it is unrelated.

[ianamason]

Yes it succeeds on my M1 too, so it is probably just bitrot with the test. I'll look into it.