Fails for the Union of these paths

[photopea]

The library throws an error when computing the Union of these two paths:

M 572.559987828135,1413.760003894567 L 523.119988933206,1413.760003894567 L 572.559987828135,1413.760003894567
M 570.879987865686,1389.280004441738 L 523.119988933206,1389.280004441738 L 570.879987865686,1389.280004441738

It probably fails for all paths of this structure (a line "there and back").

[photopea]

Have you tested your library on some well-known dataset of vector paths?

[r-flash]

Thank you for reporting! No, I'm not aware of such dataset. Do you know any? I'm building my own dataset in the visual-tests directory.

[r-flash]

@photopea, I just replicated the error and added your example to the tests (did not push it t othe repository yet). Now, the question is how should this case be handled? So far, I am discarding all lines that do not enclose any area. If I fixed this issue the most straightforward way, all lines in your example would be discarded too. Would that be an acceptable result for you?

[photopea]

This case is very stange and I can not prevent my users from entering such a shape. I think whatever way you decide to handle it, is ok.

You can look at how paper.js handles it: http://paperjs.org/examples/boolean-operations/ . It seems like it keeps both paths untouched.

[r-flash]

Started discarding path segments that go there and back, as in this case. New builds are available at https://github.com/r-flash/PathBool.js/tree/master/dist .

[photopea]

I just started to use your latest release of PathBool.js

Whenever I start using your library at www.photopea.com , within a few hours, someone reports a bug, and I see a case which your library cannot process :(

Have you thought about testing your library on some prepared samples? Maybe there is a dataset somewhere online. You can also make a generator of random paths and test your library on them. Or find a database of SVG files and extract paths from them for testing your library.

[r-flash]

I'm sorry to hear that, but I'm afraid that this project is simply too new and untested to be used in production like this (hence the disclaimer in README). That said, I really appreciate you testing it out and submitting the bug reports!

The library is currently tested on a limited set of samples which cover some corner cases (see the visual-tests folder). I'm not aware of any existing dataset. What I can do though is to test on some publicly available icon sets, for example. That would at least ensure that the program is not crashing. Generating random paths is also a good idea. That would, however, not ensure the correctness of outputs. (I tried generating the ground truth with Inkscape, but surprisingly it fails on most of the corner cases.)