Hacker News Reader

Comment by pierrec | Hacker News Reader

This tool seems to unify a bunch of text-to-diagram tools into one, which is a great idea. It's funny to see them all united, since I recently went on a binge trying out all those text-to-diagram tools for a project. My use case: writing articles on DSP including block diagrams [1].

I wanted to write the articles in Markdown and include the diagrams in there, so text-to-diagram tools seemed ideal. Turns out most of them give ridiculous layouts as soon as you have loops, with limited options to correct this. Other are based on ASCII art, so you control the overall layout, but the result usually has massive, awkward padding between everything. And then there's the holy grail, LaTeX with TikZ/PSTricks, which can give perfect results, but the time investment was a little discouraging. Result: I ended up making the diagrams with a good old SVG editor.

[1]: https://www.osar.fr/notes/waveguides/

View on HN

Thanks so much for your comment. I've just read (and played around with) your waveguide synthesis article, which led me down the rabbit hole of Sporth, ChucK, Soundpipe and now I find myself really excited looking over the contents of the "Physical Audio Signal Processing" book. I am going to learn a lot of new stuff today, thanks!

That's great to hear and exactly why the article exists. I'd say the DAFX papers are the best rabbit hole in this field: http://ant-s4.unibw-hamburg.de/dafx/paper-archive/search.php...

(And this is not entirely derailing the thread, since many of these papers are full of block diagrams, and a tool to generate nice DSP diagrams from text would be pretty useful)

I can tear through backs of napkin, but typing a diagram usually stalls me out, even though I'm pretty fast on a keyboard and the specifications seem pretty simple.

Is it just my lack of practice, or is diagramming on a keyboard always awkward and slow? Or... do people generally surpass their freehand speed with enough practice, like with typing?

If not, is the ideal solution here a digital pen or wacom tablet with diagram detection / normalization software? Scanned napkins to mechanical turk? Something else?

Note: No disrespect though, the link is an amazing project for those that must type charts.

I think you meant this as a top level comment, not a reply to mine. But yes, nothing beats a napkin in terms of speed.

> Turns out most of them give ridiculous layouts as soon as you have loops

Could you share an example? GraphViz should handle loops easily.

Once again, my use case is DSP block diagrams. I guess "ridiculous" is a bit strong, but the results certainly didn't look like what you generally find in this context. You can look at the diagrams in the link I gave for an idea of the kind of layout I was trying to achieve.

Yeah when you have requirements around layout graphviz gets a bit hairy. It's still doable with rank directives and invisible nodes, but that's the point where I start to think if I shouldn't do it another way (e.g. in this case I'd probably have done an ascii diagram rendered).

    digraph G {
      rankdir=LR
      node [shape=rectangle]

      // nodes declared in order of horizontal position ("rank")
      exciter
      filter

      subgraph H {
        // nodes at the same horizontal rank
        rank=same
        plus [label="+"]
        // invisible node for layout purposes
        p1 [width=0, height=0, margin=0, label=""]
      }
      subgraph I {
        rank=same
        delay
        nl
      }
      subgraph J {
        rank=same
        filter_ [label="filter"]
        p2 [width=0, height=0, margin=0, label=""]
      }

      out

      exciter -> filter -> plus -> delay -> filter_ -> out
      filter_ -> p2 [arrowhead=none]
      p2 -> nl -> p1 [arrowhead=none]
      p1 -> plus  
    }

Here's the result: http://prntscr.com/n35czl

Nice! Thanks for sharing your graphviz-fu, that's actually not as hairy as I thought it would be, and the result is neat enough. I do remember playing with subgraphs and ranks when testing graphviz, but I gave up before getting this far. I'll try applying those techniques to more complex diagrams, and if I find it workable, I might just switch to that.

Reply on news.ycombinator.com