[flang-commits] [flang] [flang] Handle Hollerith in data statement initialization in big endian (PR #103451)

[Rainer Orth via flang-commits]

rorth wrote:

I've done some looking myself and found that the multi-target nature of LLVM causes massive problems here.  Imagine a test with several `RUN:` lines for different targets with different endianess.  An endianess `lit` feature would have to be different per `RUN:` line.  I don't see how this can easily be done.  I fear for the moment we'll have to follow the approach in `Lower/default-initialization-globals.f90`, however repetetive and unmaintainable (imagine someone adding yet another big-endian target like `s390x`) it is.

At the same time we may want to file a `lit` bug: maybe the maintainers have ideas how to handle this.  The GCC testsuite is way simpler in that regard since it can perform whatever tests are necessary to determine the properties of the single target under test since those are guaranteed to be fixed.

https://github.com/llvm/llvm-project/pull/103451