Re: mikachu/badarrays (Re: Valgrind testing, ideas)

[Sebastian Gniazdowski <psprint@xxxxxxxxxxx>]

On 1 czerwca 2017 at 22:35:55, Bart Schaefer (schaefer@xxxxxxxxxxxxxxxx) wrote:
> I said it mostly passes tests. There's still something screwy about
> its interaction with the newer optimizations to re-use array storage
> when growing/shrinking an array, which for purposes of the tests is
> invoked whenever the $match array is changed by the pattern code; so
> all the tests that require $match are failing.

Ah ok, mostly passes tests. I think the optimizations should agree with caching of arrlen(), the new array size is computed as before (the way it is computed didn't change), and cached array len can be stored.

I'm curious how the screen saver zmorpho would perform, as far as I remember it was ~360s with no optimizations, 40s with them, 25s with hashes used to store data under integer indices. I would test cached arrlen(), would it be 30, 20, 10 seconds, that's a puzzle, arrlen() was top function in callgrind run. Is there code available somewhere?

--
Sebastian Gniazdowski
psprint /at/ zdharma.org