Upgrade to shapely 2.0

[La0]

We (Teklia) are upgrading our Kraken worker to it inside Arkindex.

Our base worker library depends on recent shapely 2.0.3

Do you have any blocker to stay on shapely 1.8.5 (released in october 2022) ?

[mittagessen]

Yes, there are numerical differences between 1.8.4 and 2.x that break polygon calculation in some cases. See here for more details.

It's on the radar but low priority. Frankly, anytime we bump up shapely versions we're finding undocumented behavior changes in corner cases for months afterwards. I'd like to avoid that if possible because it's a severe drain on my time.

[La0]

Would you consider a patch from Teklia if we apply the official migration guide on your codebase ?

[mittagessen]

The migration guide only accounts for API changes and not for behavior changes largely due to different versions of the underlying GEOM library. In fact, I already did the migration almost two years ago (eb89ab6) but had to roll it back 6 months later as hunting down all the new corner cases became intractable.
Apparently, now there are some new functions to deal with precision which might or might not resolve a lot of these issues but I haven't tried them yet. The big issue is really the polygon calculation which is rather more complex than most uses of the library.
I'd absolutely love a PR but it is much more work than just mechanically applying the API changes.

[colibrisson]

I recently installed Kraken 5.2.3 along with Shapely 1.8.5 (GEOS 3.11.1) and encountered some issues related to polygonization. This is the first time I've faced such problems with Shapely 1.8.5. I'm unsure if these issues stem from the different GEOS version or the new computational algorithm.

…

On Thu, May 9, 2024 at 11:52 AM mittagessen ***@***.***> wrote: The migration guide only accounts for API changes and not for behavior changes largely due to different versions of the underlying GEOM library. In fact, I already did the migration almost two years ago (eb89ab6 <eb89ab6>) but had to roll it back 6 months later as hunting down all the new corner cases became intractable. Apparently, now there are some new functions to deal with precision which might or might not resolve a lot of these issues but I haven't tried them yet. The big issue is really the polygon calculation which is rather more complex than most uses of the library. I'd absolutely love a PR but it is much more work than just mechanically applying the API changes. — Reply to this email directly, view it on GitHub <#601 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAZKGRKPHAVCVCYKOQB5ZFTZBNBNZAVCNFSM6AAAAABHJOXFNCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCMBSGMZTIMBRGE> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[mittagessen]

Do you have a test case I can run to see what's going on? The polygon calculation is untouched (apart from faster rotation code), only the extraction of line images with baseline/polygons has been improved.