[llvm-branch-commits] [mlir] [MLIR][Standalone] test Standalone against install distributions (PR #157944)

[Maksim Levental via llvm-branch-commits]

makslevental wrote:

> I don't fully understand the rationale for adding this test. I followed the links, and it seems to stem from [this comment](https://github.com/llvm/llvm-project/pull/157583#issuecomment-3274164357). But the relationship between that failure and this test is not clear to me. What does Standalone project have to do with testing `install-distributions`?

The existing Standalone test already performs a nested build command: 

https://github.com/llvm/llvm-project/blob/d75b837ff4c27c8ab39a11a50ff64db3687503a7/mlir/test/Examples/standalone/test.toy#L1-L8

That's why adding another similar test (a nested build but simply against a different `MLIR_DIR`) occurred to me.

> The test appears to be launching a CMake configuration process operating on the same build/source directory, and then also a nested build command...but the parent process is itself a cmake build command in the same build directory. Is such a thing even supported by CMake? Seems like the result of doing that would be undefined behavior.

You're right that reconfiguring is suspect but then we can just take the other choice and go with the version that tests `"%cmake_exe" --build "%llvm_obj_root" --target install` instead of `install-distribution`.

> Wouldn't the failure mentioend by @ftynse be caught by exercising `install-distributions` in any normal CI test?

Yes but then you're proposing we should add to the `monolithic` scripts an invocation of `install-distributions`? That seems fine too.

> This patch can probably do what libc++ does and install to a directory inside the build directory.

@boomanaiden154 can I get a link to this - I can't find it.



https://github.com/llvm/llvm-project/pull/157944