dxf file parsing error, about spline curve

[libaineu2004]

I found a problem with parsing errors on dxf's spline curve. Here uploaded 3 dxf files, you can use klayout and AutoCAD for comparison. You can see that there are problems with spline curve analysis, either with more lines or less lines.

klayout version
0.28.8

dxf with spline
https://github.com/libaineu2004/myepolltest/releases/download/dxf/bo.dxf
https://github.com/libaineu2004/myepolltest/releases/download/dxf/potato.dxf
https://github.com/libaineu2004/myepolltest/releases/download/dxf/R.dxf

[klayoutmatthias]

I do not have AutoCAD for comparison. Could you paste some reference images and describe the expected output?

[klayoutmatthias]

I used an online viewer, so now your files are on their servers. I guess you don't care ...

I see the difference here:

KLayout:

Viewer:

So I think this is what you mean.

[klayoutmatthias]

I debugged the problem and I think I fixed it. Reason was an exceptionally high point density in certain segments.

Here are my outputs now:

"R":

"potato":

NOTE: this test case suffers from this reader limitation currently:

    Invalid SPLINE flag (code 70): 11. Only types 8 (non-rational) and 12 (rational) are supported currently. (line=4580, cell=)

I assume this is the reason for the straight lines sticking out. "11" means "periodic, non-rational SPLINE".

"bo":

I could not find an online viewer that displays "R" and "potato", so maybe you can confirm that now that is the correct rendering.

Matthias

[klayoutmatthias]

Ok, found the issue. Potato looks like this now:

I guess that fixes it.

[libaineu2004]

online viewer

1、bo.dxf and potato.dxf you found the difference. Congratulations. Those differences are the problem.
2、What is the web address of the online viewer you are visiting? What is the url?

[libaineu2004]

klayout version
0.28.8

The klayout software parameters I use are the default values after installation

"R.dxf" is:

[klayoutmatthias]

I tried that online viewer: https://products.groupdocs.app/viewer/total, but also others had difficulties with "R" and "potato".

Anyway, if the "R" with the circular frame is correct it looks like I have fixed the problem.

[klayoutmatthias]

This one (https://3axis.co/dxf-viewer/dxf-viewer/) shows the "R" like KLayout now:

But the potato looks strange:

I guess nobody is perfect ...

[libaineu2004]

Anyway, if the "R" with the circular frame is correct it looks like I have fixed the problem.

autocad is correct. Also, qcad is correct.
https://qcad.org/en/

[klayoutmatthias]

Thanks for pointing me to qcad ... I tried FreeCAD, but it did not like the splines.

[libaineu2004]

Have all the problems been solved? Can you tell me what has been changed in the source code?
thank you

[klayoutmatthias]

Code change is this :35e42a8

The PR is linked to this ticket and will be merged soon into master.

[libaineu2004]

Code change is this :35e42a8

The PR is linked to this ticket and will be merged soon into master.

I tried using PR's new code and found that the program would crash.
35e42a8

I'm already in the function spline_interpolate. Added：
assert(pe ! = curve_points.end());

The assertion will take effect and an error will be reported.

I opened the following three dxf files with the same problem. My programming environment is VS2019 x64.
bo.dxf
potato.dxf
R.dxf

[klayoutmatthias]

I am using exactly these files as unit test cases and they work in my case. They are called issue_1422a.dxf to issue_1422c.dxf. Tests pass for me on VS2017.

I will take a look. VS has iterator assertions which may be useful here. Are you using debug mode?

[libaineu2004]

1、yes, debug mode。I missed one change,here:

Wrong:
for (double t = t0 + dt; t < tn + 1e-6; t += dt) {

Correct:
for (double t = t0; t < tn + 1e-6; t += dt) {

Then:
bo.dxf , ok
R.dxf , ok

2、but, potato.dxf not good

here, debug model,
j=0, k=2, p=3,
j + k - p = -1, Array out of bounds, crash

if (k <= p) or if (k < p)?? which is right?

I use 'if (k <= p)', it can run, no crash, but there's an extra line

[klayoutmatthias]

Please use the wip2 branch tip. The code reference was just for your information, not for reproducing the solution.

wip2 branch works for me on Linux and Windows.

Find the golden data for the tests here: https://github.com/KLayout/klayout/tree/wip2/testdata/dxf (issue_1422a_au.gds.gz to issue_1422c_au.gds.gz).

Matthias

[libaineu2004]

potato.dxf

Could you just tell me. For these three dxf files, which source files have the source code changed? Besides dbDXFReader.cc, are there any other files that need to be modified?

I look at the 35e42a8, only dbDXFReader. Cc file is modified. But the potato.dxf problem was not solved.

[klayoutmatthias]

I'm sorry, but 35e42a8 are the only changes made. It should be enough to apply the changes of dbDXFReader.cc - the other modifications are only new tests to cover your case. The potato is reproduced without the sticks on MSVC2017, CentOS7 and Ubuntu 22 for me.

[klayoutmatthias]

I've put in an assertion and I can reproduce the array out of bounds issue. I will debug it.

[klayoutmatthias]

I have placed one more patch: 99df15a - this should fix the potato issue.

[libaineu2004]

I have placed one more patch: 99df15a - this should fix the potato issue.

Yes.
99df15a
The result of my test is correct.
Thank you!

[klayoutmatthias]

Very good. I'll merge the PR then.

Matthias