[Freeciv-Dev] Re: [FreeCiv-Cvs] jdorje: Avoid a math error and failed assertion when...

[Jason Short <jdorje@xxxxxxxxxxxxxxxxxxxxx>]

freeciv@xxxxxxxxxxxxxxxxxxx wrote:
> This is an automated notification of a change to freeciv cvs, 
> on Mon Sep 30 08:27:46 PDT 2002 = Mon Sep 30 15:27:46 2002 (GMT)
> by Jason Short <jdorje@xxxxxxxxxxxxxxxxxxxxx>
>
> ---- Files affected:
>
> freeciv/common dataio.c
>
> ---- Log message:
>
> Avoid a math error and failed assertion when dio_put_bit_string is called
> with an empty string.  Change the assertion to check the correct value
> (the number of bits, not the number of bytes) in any case.
>
> Bug reported by Thomas Strub <ue80@xxxxxxxxxxxxxxxxxxx>.  Patch by me.
I used my fix, not Raimar's.

The fix involves two changes - either of which would prevent the 
incorrect assertion.
I changed

  size_t bytes = (bits - 1) / 8 + 1;

to

  size_t bytes = (bits + 7) / 8;

to prevent a math error with the unsigned value bits [ (0-1)/8+1 => 
really large unsigned value].  Note that this changes the intended value 
of bytes from 1 to 0 if bits is 0 - this is the "correct" value (0 bytes 
are used, not 1), and the change should not break network compatability.
I also changed

  assert(bytes < 255);

to

  assert(bits < 255);

because it is the latter assertion that is correct.  Note, though, that 
this assertion is more strict and could fail on large bit fields.  As I 
said before, allowing the number of bits to be an unsigned int instead 
of an unsigned char would pretty much remove this restriction, at the 
cost of a one-time break in network compatability.  Should this be done 
before the release (network compatability is broken anyway, right?)?
jason