text transformation

[raszpl]

fixing https://github.com/Klowner/inkscape-applytransforms/pull/59 by applying same way of accessing attributes as used by transformRectangle Works on both text and tspan in tests/data/svg/applytransform.svg

[raszpl]

I was soo overthinking this, now applies scaling to tspan. The only thing missing is scaling font size, dont know how to approach it since scaling is X Y independent but we can only control font size using one parameter. Example:

transform="matrix(1.3333333,0,0,1.6,229.22667,101.70267)"><tspan
         id="tspan70"
         style="font-variant:normal;font-weight:normal;font-size:4.8px

X 1.3, Y 1.6. What do? Scale 4.8px by average? :|

[raszpl]

Instead of overthinking it I adopted scaleStrokeWidth to scale any attribute passed as parameter. self.scaleStyleAttrib(node, transf, 'font-size') works surprisingly well on Text :o My imported Schematics dont differ all that much after applytransforms from originals. Ill call it a win.

[vmario89]

this PR is good for text without rotations, but it will mess around with this example:

[raszpl]

You are correct :( My example diagram sheet had no rotated text, checked on another one and this is losing rotation. This patch also doesnt scale DX DY.

Looking into both.

[vmario89]

basically i really like the idea to transform text, but i think the rotation is not possible (technically, due to specification of SVG i think). but it could be done to filter if text has rotation or not: if the text has rotation, skip it, else apply transform. So thats a lot more than nothing

[raszpl]

Im having trouble coercing Inkskape to set exactly what I want transform="matrix(0,1.3333333,-1.5555556,0,290.30667,236.53333)" should be turned into

       x="290.307"
       y="236.533"
       transform="rotate(90.000 290.307 236.533)"

but for some reason node.attrib['transform'] = "rotate(90.000 290.307 236.533)" automagically mangles it back into a weird matrix :|

       x="290.307"
       y="236.533"
       transform="matrix(6.12323e-17 1 -1 6.12323e-17 526.84 -53.774)"

Mathematically its the same thing, but I really want clean rotate(angle x y) in the output :( I node.attrib['transform1'] = "rotate(90.000 290.307 236.533)" it doesnt touch it, its some deliberate automation.

I feel like Im really close to the solution. I want SVG with flattened applied matrix equations for easier automatic scale detection when importing to Kicad. Source still slightly mangled, but almost there output right now

[raszpl]

Added scaling dx dy. Looks almost perfect now, or as perfect as Inkscape can do with converted pdf. original: result:

Two problems:

• Still dont know how to force it to output "rotate(a x y)". Keeps translating this to matrix in the background :( Whole point of me doing this patch was getting svg with nice rotate transforms. There is nothing about it in https://gitlab.com/inkscape/extensions/-/blob/master/inkex/elements/_base.py#L179 tr = str(f"rotate({angle:.3f} {new_x:.3f} {new_y:.3f})") tr is rotate(-90.000 404.499 96.693) node.set('transform',tr) node.get('transform') returns matrix(6.12323e-17 -1 1 6.12323e-17 307.806 501.192) Doesnt do it to node.set('transform1',tr) or any other random attribute name, doesnt do it to pure rotate(-90.000) with no pivot point. :| For my own use I resigned to outputting node.set('transform1',tr) and renaming in later pass.

• Useless clip-path="url(#clipPathxxx" that Inkscape inserted in random places to individual letters (][ and g) when converting from PDF. Those become invisible when matrix is replaced with xy coordinates. Regex clip-path=.*" takes care of that garbage.

[raszpl]

First problem fixed :] node.attrib['transform'] = str(f"rotate({angle:.3f} {new_x:.3f} {new_y:.3f})") bypasses magic transmutations.

Second problem can be fixed by

if 'clip-path' in node.attrib:
			del node.attrib['clip-path']

but thats not universal, someone might want to keep 'clip-path's

[vmario89]

thats cool! i messed around and the rotation seems to work now! i found another thing which might be problematic - skewing / non-uniform transformation does the following

before:

after applying transform:

source file:

[raszpl]

Yes, it only cleanly handles uniform scaling. In my Example rotated pin numbers have slightly wrong offsets. ~~I suspect they move from their original location because we can only scale Fonts with one number while example text has 1.3:1.6 scaled ratio. Probably could be computed better, but this is good enough for my original purpose.~~

For things like skewing there definitely should be an option of detecting and skipping.

[raszpl]

I copied

		sx = math.sqrt(a**2 + c**2)
		sy = math.sqrt(b**2 + d**2)

straight from transformRectangle and didnt even think about it, but its Wrong. It doesnt extract real X Y scaling! Cant even notyice because test rectangle is 1 white, and browsers usually renders SVG on white background so cant see it 2 is uniformly scaled by -1,1! correct calculation is

		sx = math.sqrt(a**2 + b**2)
		sy = math.sqrt(c**2 + d**2)

and now both transformRectangle and my rotated text works!

[vmario89]

have some new simple test file where scaling messes up

gets

[raszpl]

your

https://private-user-images.githubusercontent.com/5631071/481832060-c42bde55-d57c-4059-9f45-9f487aa90ba1.svg

  <text
     id="text350"
     xml:space="preserve"
     x="28.402536"
     y="2.6447694"><tspan
       id="tspan350"
       x="21.988474"
       y="2.6447694">Kontext Kommunikation</tspan></text>

doesnt have any scaling?

[vmario89]

very strange. now when i open again i cannot reproduce ... :O