[erlang-questions] surprising integer (external) encoding

[Andy Sloane]

On 7/5/07, Valentin Micic <valentin@REDACTED> wrote:
> To make things even more confusing, the second complement encoding is using
> always 32, i.e.
>
> (C)   term_to_binary( -2 ) produces: <<131,98,255,255,255,254>>
>
>     - 131 is a *magic* number
>     - 98 represents external tag for four-byte integer (actually, 2^27)
>     -  <<255,255,255,254>> -- second complement encoding for -2.

According to the code in erts/emulator/beam/external.c:

/*
 * For backward compatibility reasons, only encode integers that
 * fit in 28 bits (signed) using INTEGER_EXT.
 */
#define IS_SSMALL28(x) (((Uint) (((x) >> (28-1)) + 1)) < 2)

Since (x) here isn't cast to an unsigned, this check doesn't work for
negative numbers.  So it's apparently the combination of historical
reasons and a bug.

I have no idea what the historical reasons are, but I found your
observation interesting.

-Andy