You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
After PR #384 was merged, I thought I would look around and try to close as many easy issues as I could while my brain is still familiar with compiler-builtins. I realized that issues #162 and #168 was caused by the fact that fixing the exponent would enable the generation of subnormal numbers, and the soft multiplication does not support subnormal numbers (at least not yet). I haven't messed with build.rs much because I thought it was doing some really special testing. After looking it over, I realize all it is is a hackier version of the tests now under tests/. It doesn't go anywhere near the level of rigorousness my new tests achieve (it hasn't found any of the subnormal or rounding edge cases #384 found).
The only thing build.rs has that the new testing suite doesn't is a few more tests related to the special intrinsics that ARM gets. Should I open a PR to test the remaining functions and delete build.rs and tests/generated.rs altogether? Also, I think the references to panic-handler and utest_macros crates can be deleted. Some documentation says they were introduced as a workaround for rust-lang/rust#51647, but that issue was closed a long time ago.
The text was updated successfully, but these errors were encountered:
I think as long as we test basically the same amount of things it's fine to refactor/clean up as necessary personally. I forget the exact intricacies of how this is all tested, but I see no reason to keep one specific way of testing over another!
After PR #384 was merged, I thought I would look around and try to close as many easy issues as I could while my brain is still familiar with
compiler-builtins
. I realized that issues #162 and #168 was caused by the fact that fixing the exponent would enable the generation of subnormal numbers, and the soft multiplication does not support subnormal numbers (at least not yet). I haven't messed withbuild.rs
much because I thought it was doing some really special testing. After looking it over, I realize all it is is a hackier version of the tests now undertests/
. It doesn't go anywhere near the level of rigorousness my new tests achieve (it hasn't found any of the subnormal or rounding edge cases #384 found).The only thing
build.rs
has that the new testing suite doesn't is a few more tests related to the special intrinsics that ARM gets. Should I open a PR to test the remaining functions and deletebuild.rs
andtests/generated.rs
altogether? Also, I think the references topanic-handler
andutest_macros
crates can be deleted. Some documentation says they were introduced as a workaround for rust-lang/rust#51647, but that issue was closed a long time ago.The text was updated successfully, but these errors were encountered: