String representation in erlang

Shawn Pearce spearce@REDACTED
Tue Sep 13 16:17:09 CEST 2005

Interesting - but how is this better than a binary?

If I recall the source code correctly any binary using less than
255 words is stored on the process heap; larger binaries are
allocated in a shared heap (to reduce message passing costs).
This of course means that any "string" stored in a binary would
require 8 + NumberOf8bitChars bytes of memory (rounded up to the
next full word).  If NumberOf8BitChars is < (255 * 4 = 1020) then
it will be allocated on the private heap.

Further binaries can be easily pattern matched in function headers
and are already handled by the io library; this packed string
representation is more difficult to pattern match against and isn't
directly handled by the io library functions.

Thinus Pollard <thinus@REDACTED> wrote:
> Hi there
> According to the Erlang efficiency guide a string is internally represented as 
> a list of integers, thus consuming 2 words (8 bytes on a 32bit platform) of 
> memory *per* character.
> The attached code is an attempt at reducing the memory footprint of strings in 
> erlang (passing between functions etc etc).
> The basic idea is to pack a string into n byte sized integers and unpacking 
> them on the other side. The text file called compare.txt also shows the 
> memory needed to represent strings in normal erlang strings and this string 
> packing.
> Normal erlang strings are 2 words/character. The packed representation uses 1 
> word of memory per list element plus n bytes/wordsize per integer element, 
> where every integer element contain n characters.
> Deficiencies:
> If the string length is not divisible by n, space is wasted (the string gets 
> padded with zeros). 
> Usage:
> Pick your the integer representation length.
> packstring/1 takes a string returns a list of n byte integers
> unpackstring/1 takes an integer representation and returns a string.
> There is a simple test suite in test/0.
> If anyone can improve upon this code, please do. If this was an exercise in 
> futility, please let my know, I've only been programming erlang for 2 weeks 
> and still need to learn all the gotchas ;)
> -- 
> Thinus Pollard


More information about the erlang-questions mailing list