Kevin Lawler

Good catch. I would've missed that. http://www.delorie.com/gnu/docs/readline/rlman%5F41.html Libedit non-blocking support looks less advanced. I am guessing linenoise has none. I would like to do this without managing an input thread....

It has been a long time since this issue was revisited. Much has changed about the C landscape in the intervening time, particularly on Darwin & BSD. Is libedit now...

Not in favor of this formulation, in favor of the general idea. I have always thought `vprintf` support, potentially as reserved verb `_printf`, was a good idea, provided it is...

"Here's a profile, run on 32-bit Debian Linux. Probably everything above reverse (except divide) in the flat profile is interpreter overhead that could be reduced. It appears to be spending...

" wd is called only 3 times, but valence is called 6+ million times. (according to the call graph)"

In dv_ex() dyadic-over reuses a container K. The other adverbs don't yet have this optimization. The line in question makes heavy use of eachright. Eachright is missing the optimization. I...

I tried reusing the container in eachright. That didn't seem to affect the runtime. Next to try would be each. Probably shouldn't spend too much time doing this. Adverb dispatch...

``` +/1+2*&{&/{y~|y:x _vs y}'[2 10;x]}/:1+2*!500000 / 40s ``` T. Nordin's patches have removed about 20% of the overhead so far. There is more to go. Doing valences implicitly seems to...

> Can you clarify what you mean by "doing valences implicitly"? Not storing/caching the valence near the function pointer or object pointer. > Is recalculating mempool cell sizes really a...

I now think the problem is that _vs is implemented in K instead of C. (Based on an OS X Instruments profile.)