- Inkscape Devel - Inkscape Mailman

Backports into 1.0.2
by Marc Jeanmougin 28 Nov '20

28 Nov '20

Hi team, If you have ideas or proposals to backport stuff that is already in master, and want it into our next 1.0.2 release, could you mention it in https://gitlab.com/inkscape/inkscape/-/issues/2067 Input from the bugs team would be particularly appreciated in this matter :) Thanks everyone! -- Marc

1 0

finally, finished using boolean operators
by NASA Jeff 15 Nov '20

15 Nov '20

finally, i finished off by trimming the lhsf eyebrow to the face using boolean operations. I couldn't seem to get them to work how I wanted, to trim the eyebrow bit. And selection order didn't seem to change the precedence in the boolean operation. In the end I ended up doing a copy and paste replacing the face, then a copy of the eyebrow with no paste, then the boolean operation between the eyebrow and one face copy, then paste from the eyebrow copy the style (opacity/blur didn't seem to copy, just the colour, not sure it stroke styles copied or not), then deleted the original eyebrow. And renamed everything correctly again using the XML editor. I noticed you can't do a multiple select in the XML editor, only select a single element. So, I think I'll raise three bugs from this, boolean operation precedence, style paste not pasting blur and transparency and maybe stroke styles etc.. and inability to select multiple shapes/groups via the XML editor dialogue. I'll leave the snapping as an enhancement to look into and not raise as a bug as the enhancement I suggested may be best as a different snap modulator.

1 0

path snap
by NASA Jeff 15 Nov '20

15 Nov '20

As I mentioned in my last email, I found snap to path, but it didn't work that great. It seemed to fit curve to a curve well some of the time and didn't seem to snap vertices to the path curve. What would be great is if it could snap a curve to a curve all of the time by inserting vertices as needed to make the snap happen. I'm also going to play around with grow and shrink, to see if that's a better option for multilayer hair, that just adjusting the paths manually. There would still be some alignment of paths between objects with higher/lower z-orders needed and it still wouldn't solve the eyebrow problem. I may end up just doing a boolean operation.

1 0

found path snap
by NASA Jeff 15 Nov '20

15 Nov '20

I found path snap and it worked okish, though not that great. An example of where matching two paths is required because you can't just use z-order and have a blob underneath is the eyebrow, where I have a semi transparent top layer for the hair and an opaque background that has to be beneath or fit the face and above or fit the eyebrow. https://imgur.com/4oHHL6Z

1 0

simple edge detection of a bitmap
by NASA Jeff 15 Nov '20

15 Nov '20

I can't remember exactly how to do an edge detection mask for a bitmap but it's something like 0 -1 0 -1 0 -1 0 -1 0 as a matrix applied to the bitmap, this has issues with jpeg compression though as you get square artefacts from the lossy-ness of the compression, soi it may be an idea to detect the compression window for the jpeg image and use that to scale the matrix. I believe jpeg compression is more or less just an inverse fast Fourier transform.

1 0

Here's the image I've drawn
by NASA Jeff 15 Nov '20

15 Nov '20

https://svgshare.com/i/RUo.svg Some curves I've hand fitted to each other and some I've just relied on z-order with a cruder edge for the lower shape. I'm not sure how well relying on using z-order to account for matching detail of the edges of paths would work if we wanted to animate the shapes. e.g. make the dress and hair move as if the wind was blowing.. Incidentally, this svg renders funny in chrome, I've added some blurred, semi-transparent detail on a higher z-order than the base image but it's not showing... Here's a png exported from inkscape.. https://imgur.com/j8mDI6b

1 1

slightly more advanced
by NASA Jeff 15 Nov '20

15 Nov '20

Something slightly more advanced would be being able to take the pattern from the traced portion of an underlying image. So lets say I trace a sofa and I want to take the pattern from the material of the sofa and use it to fill my vector shape. Even more advanced is that the possibility to extract the shading from an underlying image and turn it into fill gradients may become possible once a uniform texture is extracted from the material.

1 0

snapping to lines &co.
by NASA Jeff 15 Nov '20

15 Nov '20

I've just been hand tracing a jpg image someone asked me to turn into a drawing. Here are my notes on the process: When moving points they need to be able to snap to points on the curve of lines of other shapes, maybe controlled by being in the same group. Smoothing worked quite well, it would be nice if there were some option to do it to the contours of an image either, Fitting when smoothing, Snapping curves to to an underlying image when editing a curve over an image, could be controlled by layers and opacity etc.. Another feature that would be nice is to snap one part of a path to another part of a path, breaking segments where necessasasary. So say you have two hand drawn paths, where some segments of one are close to or partly overlap the some of the segments of the other path. An option to fit the two segment sets of the path that are close to each other would be nice. I expect I can prototype some of this.

1 0

oh, i didn't send this and answered my own question.
by NASA Jeff 14 Nov '20

14 Nov '20

Does anyone know OTT how filling is performed? Is it just the standard, sort the vertices and walk the inside edges down either side of the vertex a line at a time? I can probably tell by making some shapes that don't fill in a uniform manner when you do that and see who inkscape fills them. I've got a good collection of path segment functions together now for prototyping, and found the first SVG feature implemented on the GPU, or at least in OpenCL, filters. So, that's a positive. I'll think I'll keep on working on the extension before doing any experimentation but I think I've to go a good outline of things to test to see how they fair in terms of rendering performance and enough SVG features down that I should be able to do a bit more than just a point line test. I looked up Bézier length calculations again, and they are basically an approximation using lookup tables. Wikipedia said you could calculate them definitely using a partial integral when I looked a while back but it seems it was incorrect and I'm back where I started in high school calculating the sum of the magnitude of the segments.

3 2

so that would mean.
by NASA Jeff 13 Nov '20

13 Nov '20

When I do a bucket fill I store the result as a collection of lines going from one inside edge to the other than enumerate the list to redraw the lines. Something like gimp would typically trace the extremities to produce a selection outline and fill a bit mask in the same way as I would bucket fill. To create a selection. I don't think a GPU would handle the outline trace at-all well, but most fill patterns and styles are quite possible using shaders, or tilled textures. (I just did a search and came across an OpenCL implementation of filters) https://bugs.webkit.org/show_bug.cgi?id=70099 I don't know what licence the code uses but it's a good example of what's achievable. Oh, I see now, there's fill rule evenodd and fill rule nonzero, so i was almost right.. That's a lot of experimentation, hopeful we should be able to identify what SVG elements or parts of elements can be calculated or rendered at high performance and which ones require a more traditional approach. Even then performance could possibly be improved manually vectorizing if GCC doesn't for a good job. (I expect you have to structure your code in a way that's friendly to GCC's algorithms) and switching to scaled integers over floats which are higher performance and give you more control over the (I've forgotten the word), basically the number of bits you've got to play with to maintain a desired rounding error. Integers having higher performance is certainly the case for older generation CPUs where that bit of performance may be really noticeable to the end user who may not have an acceleration viable GPU.

1 0