Resolve xtl in user cmake#1351
Conversation
|
@drew-parsons does that work for you? |
serge-sans-paille
left a comment
There was a problem hiding this comment.
LGTM too, thanks
|
I'm not certain how to judge this PR. It explicitly preserves the current behaviour of xsimd 14, i.e. reproduces the "inconsistent" behaviour that I reported, adding edit: wait, no, I think it's working fine for me. I didn't have the tests in my demo test organised as clearly as it could be. The "2 tests" it reported were for type float vs complex, not test_load_store vs test_load_store_xtl. doctest didn't report the SUBCASES clearly. Placing test_load_store_xtl in a separate TEST_CASE_TEMPLATE rather than a SUBCASE shows the patch is working. Compile flag So, looks like this patch is working fine with xsimd 14. If I understood right, the ENABLE_XTL_COMPLEX handling for tests at Line 62 in 510b4f1 |
|
Incidentally, I added an extra test run to the debian CI tests, "test xsimd-test-float" (and test "xsimd-test-clang-float"), which runs the tests with this patch with an explicit |
Instead of having the packager (Debian, Conda recipe) decide on xtl support, let the final user decide on wheter to use it.
The current behaviour is maintained and overridable until xsim 15.
In xsimd 15, the configure time option has no effect for installed targets.
Close #1338