Inverting a 4x4 matrix with try_inverse_mut doesn't leave self unchanged if the inversion fails.
#1384
Conversation
jnyfah
changed the title
try_inverse_mut doesn't leave self unchanged if the inversion fails.
|
I guess we don't have any unit tests validating this. Want to add one? |
|
@Ralith, I realized that I had mistakenly omitted the cofactor |
|
Oh, nice, looks good then. |
|
@Andlon oh yes, added the test now |
tests/core/matrix.rs
Outdated
| #[test] | ||
| fn test_inversion_failure_leaves_matrix_unchanged(m in matrix4()) { | ||
| let original_matrix = m.clone(); | ||
| if m.try_inverse().is_none() { |
There was a problem hiding this comment.
Does this branch actually get hit? I'm not sure how often proptest generates non-invertible matrices. Would a hardcoded case work better?
There was a problem hiding this comment.
Agree with @Ralith, I think the likelihood of matrix4() producing singular matrices is likely (very) small (though I haven't checked), so a test with a fixed input is better in this case.
|
I was actually just asking about a test to make sure this isn't causing a regression, but that was already covered. |
tests/core/matrix.rs
Outdated
| 4.0, 8.0, 12.0, 16.0 | ||
| ); | ||
| let expected = mat.clone(); | ||
| if !mat.try_inverse_mut() { |
There was a problem hiding this comment.
This test is currently very brittle. Imagine that some change in an implementation detail somewhere caused mat.determinant() to return something not exactly 0.0. Then mat.try_inverse_mut() might return true and the test would succeed, although it's no longer testing what we want to test it.
Perhaps replace the if by an assert? E.g. assert!(!mat.try_inverse_mut()) followed by the assert_eq that you already have.
tests/core/matrix.rs
Outdated
| @@ -1263,6 +1263,20 @@ fn column_iterator_double_ended_mut() { | |||
| assert_eq!(col_iter_mut.next(), None); | |||
| } | |||
|
|
|||
| #[test] | |||
| fn test_inversion_failure_leaves_matrix_unchanged() { | |||
There was a problem hiding this comment.
Maybe rename to _matrix4_unchanged to indicate that this was specifically an issue with 4x4 inversion and what we're testing here
|
@jnyfah: thanks for fixing the last remaining points! We're all good to go now, except CI is failing due to formatting. I think you just need to run |
|
Thanks for your help with this @jnyfah! The CI failed due to an unrelated issue (a brittle test). |
Fixes #1380
Modified the
do_inverse4function to calculate the determinant of 4x4 matrix before proceeding with the computation of the adjugate matrix to ensure thatoutis not modified if the determinant is zero.