Re: [Chicken-hackers] CHICKEN in production

[r]

08.10.2014 7:52, Alex Shinn пишет:

On Wed, Oct 8, 2014 at 9:30 AM, r <address@hidden> wrote:

The main problem is that pure Scheme implementation double memory usage (half of memory reserved for coping GC), a full search index tree contains about 2M nodes (approx 40*2MB).

This should be easy to workaround using the mmap interface in the posix egg.

Not sure i get it, you propose to put Сhicken heap into memory mapped file?

Also library should run on ios/android/win (offline catalog on user device or PC) packaging libchicken with eggs very tricky.

Packaging utilities are easy to improve on, and for distribution to devices automatic static compilation may be the best option.

I think this is not so easy as it sound, at least for me :]

Performance tests show that reference implementation 2-3 times slower than C.

For the low-level nature of your application this may be difficult to improve on.

Yes, after all tunings our C implementation outperform Clucene/Lucene engines.

Passing strings through FFI is pretty slow, for example c-string type was used to represent color value

Why not an RGB value?

That was an early prototype dunno how this happen, currently we use orthodoxy argb8888 integers.