This is great. It will make the sample node much simpler! If you want, the commit message could mention that it's inspired by (or at least similar to) OpenVDB's ValueAccessor.

This will probably be more beneficial when up-sampling, since the hint will be useful more often is the segments are long compared to the samples.
We could probably improve it for the down-sampling case by looking 5 or 10 segments ahead. But it's fine to leave that for later, since this is a nice simplification either way.

Unfortunately the two tests I added in D15340 still fail. That doesn't need to block this patch, it should be easier to look into after anyway.

source/blender/blenlib/BLI_length_parameterize.hh
135
139

This revision is now accepted and ready to land.Jul 2 2022, 8:35 PM

Unfortunately the two tests I added in D15340 still fail. That doesn't need to block this patch, it should be easier to look into after anyway.

I was looking at the tests now. I might be wrong but it feels like the test cases are just wrong. They are mixing up point values with segment lengths.

It doesn't work even with a much simpler example like lengths = {0, 1}.
If I create actual point values that result in the given lengths (like below), it works fine.

Vector<float> point_values;
point_values.append(4.0f);
for (float &length : lengths) {
  point_values.append(point_values.first() + length);
}

Jacques Lucke (JacquesLucke) marked 2 inline comments as done.Jul 2 2022, 9:47 PM

typos

Harbormaster completed remote builds in B22834: Diff 53191.Jul 2 2022, 9:48 PM

Closed by commit rBab444a80a280: BLI: refactor length parameterization (authored by Jacques Lucke (JacquesLucke)). · Explain WhyJul 2 2022, 9:52 PM

This revision was automatically updated to reflect the committed changes.

Jacques Lucke (JacquesLucke) added a commit: rBab444a80a280: BLI: refactor length parameterization.

In D15348#415194, @Jacques Lucke (JacquesLucke) wrote:

I was looking at the tests now. I might be wrong but it feels like the test cases are just wrong. They are mixing up point values with segment lengths.

It doesn't work even with a much simpler example like lengths = {0, 1}.

If I create actual point values that result in the given lengths (like below), it works fine.

I was suspicious about that but not sure. Makes sense in retrospect. And good news! Thanks for looking into it.