I don't think the cost of dynamic lookup will be very high, at least in news.arc there isn't that much text. I intend to use a function that looks into a table:
(prn (i18n 117))
I think numbers over pseudo-text-symbols will fit with arc's keep-it-brief style.
I don't like the numeric setup - I find it makes the code hard to read. Most of the other i18n libs I've seen use English in the code and then a mapping between English and other languages in a separate file. I think that approach makes more sense. When I'm skimming code, I'd rather read (i18n "Add Comment") than (i18n 143).
Thanks for the hint, I've now got a simple version where it takes the who-called-who from recording invocations. I'm currently working on sample call data, but profiling would also be interesting.
Thank you for making this reduced graph. I'm all for visualization, but the full call graph seemed to show nothing except how useless a full call graph is.
So I did the first round using Arc now, worked quite well apart from me not being used to the language that well. Might switch to CL for the rounds where time is crucial, as that has better Emacs integration.
