On 21-07-2013, at 12:02 PM, Nicolas Cellier <nicolas.cellier.aka.nice at gmail.com> wrote:

> Clearly, character scanning inner loop is now written all in Smalltalk, and that seems efficient enough on our computers.

Well, when is efficient enough, enough? The first problem is that for every send to basicScanCharacters… we are wasting time on calling the prim and failing it after pointless work. Second is that we have no measurement for whether things would be faster with the prim working or not. I could easily believe that on a Cog vm it would be about the same either way, but only after seeing decent tests results.

> Is it a problem on Pi?

Performance wise, almost everything to do with running under Morphic is a problem. Run in an MVC project on a Pi and see it fly!

> IMHO, the thing to do would be to document the primitive as backward compatibility support only (interpreter VM)
> In case of COG, since there is no such backward compatibility anyway, drop the primitives.
> Or rewrite them to support i18n…

I'd love to se some reasonable tests done to measure the effect of
a) remove prim entirely
b) fix prim (probably by actually faking up a test scanner class that doesn't break the prim rules)
c)  machine differences -pc, mac, plain interp, cog, Pi, Pi stackvm


tim
--
tim Rowledge; tim at rowledge.org; http://www.rowledge.org/tim
Try not to let implementation details sneak into design documents.