TIL Kicad's builtin font is defined as a set of Kicad symbol libraries, which (mis) use symbol features like pins to mark out important locations that matter for fonts. Then a python script goes brrr to convert all those "component symbols" into a C++ file containing hardcoded glyphs.

Here's the latin capital letter A, for instance.

It's... quite cursed? I went into this to try and see if I could find an easy way to extract an approximate size of individual glyphs, and the answer appears to be "lol, lmao".

@danderson this is deeply cursed

Oh also the font is defined as _combinations_ of these glyphs. That's why there are all those position anchors on the capital letter A, because then there's a charlist.txt that maps the supported unicode codepoints to some combination of these glyphs, welded together at those control points. For example A with an acute accent is defined as:

+ A_CAP ACUTE ABOVE=X

I don't know exactly how to parse this custom format yet, but you can kinda see that it's saying "take the A_CAP glyph, then paste the ACUTE glyph at the position marked "ABOVE". Sort of.

oh I see, "X" is a point defined (again using a component pin) in the ACUTE glyph. So, this is saying that the point ACUTE.X should be positioned at A_CAP.ABOVE.

I mean it makes sense, but also...

oh wow the python code that renders this font into a C file parses S-expressions with regular expressions. That... is certainly something you could do. I mean empirically you can, since it works, but...

Unfortunately this is all quite bespoke and tailored specifically for nothing but creating the C++ source file that Kicad itself uses to wrangle the font, afaict if I want to obtain the glyphs in a different form that I can use, I'm going to have to reimplement a lot of the bespoke parsing and wrangling.

Which I guess is fine since I was already doing that in reverse to write Kicad symbol files, but parsing them back deeply enough to be able to do geometry transforms on them is still quite a task.

Ooor I can go with my original plan A, which was to just brute-force generate a symbol library where each symbol is a single character, then use kicad-cli to render that out to svg, then parse that back out to determine the bounding box on each glyph.

It's not perfect though, I lose the spacing markers in the original source font, which would allow me to get 100% precise layout estimates... Hmm.

okay okay, even more cursed, what does the generated C code look like? Would that end up being easier to crunch into the data I need?...

Hmm not sure using the generated C++ font data is going to be easier... What on earth is going on in there...

// In stroke font, coordinates values are coded as <value> + 'R', where
// <value> is an ASCII char.

wat

Oh wait it gets better, each point is stored as a uint8 offset by "R", but that represents a float value between -1.0 and +1.0. So to get the coordinate, you take your uint8, subtract 0x52 (ascii "R"), then divide by 21 to get the coordinate.

Oh and the Y coordinate is actually offset by "Z" not "R", for legacy reasons.

Oh and the first two letters are the X/Y size of the glyph, and _those_ are both relative to "R", not "R" and "Z".

So now you parse those characters in pairs and that gives you coordinates. Oh and "<space>R" is a magic value for "raise the pen", i.e. don't draw a line between the two points it separates.

I mean, you know... it's quite cursed as a format, but once you know how to parse it... It's not that hard to convert back to geometry.

I have to assume this is some kind of cursed industrial control format, a distant cousin of gerber or something, and that the original Hershey fonts were in that format, and so as it evolved into Kicad's own font all these layers remained.

@danderson I've definitely been assuming this all bottoms out in Hershey Font arcana eventually, but meanwhile have been enjoying your sausage factory tour in all its grisly detail....

Okay one nice thing about this cursed font encoding is that, with a little practice, you can kinda read the shape of glyphs from the raw data.

For example, "F^K[KFYFY[K[" is the little square that represents a character that's not supported by the font.

Remove the first two chars since those define the dimensions, then pair up the rest: K[ KF YF Y[ K[

Even without knowing the numeric equivalence, you can see how the pen is moving in a rectangular pattern.

@danderson

@danderson

That is the original Hershey font format :-)