add wiggle function to fix an invalidated position #1
Conversation
Codecov Report
@@ Coverage Diff @@
## master #1 +/- ##
==========================================
- Coverage 98.75% 96.17% -2.58%
==========================================
Files 1 1
Lines 402 497 +95
==========================================
+ Hits 397 478 +81
- Misses 5 19 +14
Continue to review full report at Codecov.
|
|
Nice to see some activity on this ;)
Yes, in theory the main motivation of the data structure was to have something that allows fixing a position much faster than going through a whole remove + insert. Doing a remove + insert would also be possible on a tree, but from some preliminary benchmarks I felt that this is too costly, which is why I went for this array based solution. In practice we can assume that the rank fixes are small, so in terms of arrays we would only have to swap a small number of neighbouring elements. But I think we can see all that as a possible performance optimization and start with just going for remove + insert.
From the 21re/rust-geo-booleanop#17 perspective I was expecting the interface of the "fix rank" function not to operate on a single element, but rather on either two (because if the algorithm splits something, it typically splits two segments, which both need to be rank-fixed), or even three (because the current segment might overlap with both the "above" and the "below" segments, so from the perspective of the outer loop there may be even three segments that had been mutated). From an algorithmic perspective I was never fully sure if it is better to fix the ranks one-by-one (potentially inefficient) or if such a "fix all rank mutations at once" approach is feasible. But you seem to be more actively thinking about 21re/rust-geo-booleanop#17, so I'll simply support any attempt to get it fixed :). |
|
FYI I've published a new version (0.2.1) in case that simplifies things for you. |
|
That wasn't really needed, but is very much appreciated :) |
I've implemented a function (wiggle) to fix the element position for a single mutated element.
I'm not sure how bad the removal/insertion implementation is performance wise, but it seems to work okay across block-boundaries.
I'm going to try to use this in 21re/rust-geo-booleanop#17 where I think that one could know which element may offend the order criterion (namely the ones mutated by splitting)