[NOMERGE] llvm: fix cross-compilation, alternative approach #40711
Conversation
GrahamcOfBorg
added
10.rebuild-darwin: 0
| "-DLLVM_HOST_TRIPLE=${stdenv.hostPlatform.config}" | ||
| "-DLLVM_DEFAULT_TARGET_TRIPLE=${stdenv.targetPlatform.config}" | ||
| "-DTARGET_TRIPLE=${stdenv.targetPlatform.config}" | ||
| ] | ||
| ++ stdenv.lib.optionals (stdenv.hostPlatform != stdenv.buildPlatform) [ | ||
| "-DCMAKE_CROSSCOMPILING=ON" # inferred when setting CMAKE_SYSTEM_NAME, but is satisfying to do so explicitly |
There was a problem hiding this comment.
Shouldn't the cmake setup hook be setting this already? If not sounds like a bug.
|
(cc #25477) I don't think so? Not sure what setting this is "supposed" to be interpreted as anyway, for example: https://gitlab.kitware.com/cmake/cmake/issues/17653 . Basically it appears to be true when explicitly set to true (as we do here) or when CMAKE_SYSTEM_NAME was set directly or indirectly. Perhaps it's best to just avoid bothering with it at all? |
|
It should be set because various cmake users will check for it, e.g https://github.com/alisw/clang/blob/master/CMakeLists.txt#L433 (and https://codesearch.debian.net/search?q=CMAKE_CROSSCOMPILING in general) |
Well we should probably be setting FWIW upstream cmake added documentation about it's limitations[1] which is why I'm reluctant to say we should clearly set it. But maybe the right answer is to just set it since that seems right and look at the fallout and handle the cases where those limitations cause problems instead of ignoring it entirely O:). Alright, you've convinced me :). Regardless doing that is a big endeavor that may be worth handling separately. [1] https://gitlab.kitware.com/cmake/cmake/merge_requests/1672/diffs |
|
Here's the logic Buildroot uses for CMake packages:
They seem to use a toolchain file that eventually sets |
|
Nevermind for now, we can revisit if/when have more time for this :). |
This should be "ready" functionally (AFAIK),
but needs discussion to sort out how we want to do things.
Cleaned up version of the mess I made over on #40604 (again, sorry!)
providing a different way of reaching those goals.
Not sure if this is my preferred approach,
but hopefully is interesting and helps
move things along.
LMK what you think, thanks! :)
Submitted against 'master' since that's how I developed it
and because staging would only make evaluating this more difficult.
("is build failing because of this PR or because this is staging?" :P)