Fix turns where segments arrive at an intersection point, or leave #773
Conversation
barendgehrels
force-pushed
the
fix/fraction-by-arrival
branch
from
b03b111 to
2dcaa2d
Compare
| // Not yet fully tested for float and long double. | ||
| // The difference algorithm can generate (additional) slivers | ||
| BoostGeometryWriteExpectedFailures(10, 11, 24, 14); | ||
| BoostGeometryWriteExpectedFailures(10, 11, 24, 15); |
There was a problem hiding this comment.
(this was already)
There was a problem hiding this comment.
This means that there is no extra error here?
There was a problem hiding this comment.
Nope, these statements are mainly for myself, during transition of removing rescaling. This is one for long double. I probably didn't update it correctly last time. It's not updated automatically.
For this PR, I commented the code (updating the fraction) and it was still there - so it's not this PR for sure.
Main purpose is to see if we get there. For set operations, the version without rescaling is (nearly) on par, or better. For buffer it's still not good enough so I'm concentrating on those cases.
| // Not yet fully tested for float and long double. | ||
| // The difference algorithm can generate (additional) slivers | ||
| BoostGeometryWriteExpectedFailures(10, 11, 24, 14); | ||
| BoostGeometryWriteExpectedFailures(10, 11, 24, 15); |
There was a problem hiding this comment.
This means that there is no extra error here?
|
@barendgehrels Do you want it to be released with 1.75? |
Hi @awulkiew , no, it's not necessary - only for without rescaling, but that's currently not the default. |
| { | ||
| ti.method = method; | ||
|
|
||
| // Intersection points are calculated on both segments, either one can be taken. |
There was a problem hiding this comment.
This comment was wrong, I'll fix it and merge. Thanks for your reviews
Then their fractions should be 1 or 0 (and not nearly so)
barendgehrels
force-pushed
the
fix/fraction-by-arrival
branch
from
2dcaa2d to
42bd7cf
Compare
|
@barendgehrels since you merged it now I'll push it together with other fixes to master for 1.75. |
@awulkiew , fine, thank you! |
3 participants
This happens in rare cases. When they arrive there (or leave), their fraction should be 1 (or 0). The fraction is calculated and is sometimes nearly 1 (or ), but not considered so.
This can be fixed without any threshold because the information on arrival is already there.
It fixes ~10% of the remaining errors in the
recursive_polygons_buffertest.Several cases (added last PR) are fixed now. And I added 3 new cases (not yet fixed by this)