Difference in float numbers.

Please post any questions about developing your plugin here. Please use the search function before posting!
User avatar
Senior Member
Posts: 114
Joined: Fri Mar 15, 2019 6:08 am

Difference in float numbers.

Postby InvisibleSoldiers » Wed Jan 22, 2020 5:54 am

I have this:

Syntax: Select all

def post_hook(stack_data, ret):

# CBasePlayer pointer.
pointer = stack_data[0].get_pointer(0x4)
index = index_from_pointer(pointer)
player = playerdict[index]

# CMoveData pointer.
pointer = stack_data[0].get_pointer(0x8)

new_velocity = make_object(Vector, pointer + 0x40)
old_velocity = player.velocity

if stack_data[2] > 0.00001:
for i in range(2):
d = new_velocity[i] - old_velocity[i]

if d != 0.0:
wishdir = make_object(Vector, stack_data[1])
gain = d / wishdir[i]

accelspeed = stack_data[3] * stack_data[2] * TICKINTERVAL

print(gain, accelspeed)

if gain == accelspeed:


Output with airaccelerate 1.0:

Syntax: Select all

2.5999991896173777 2.599999636709697
2.5999982694367216 2.599999636709697
2.5999983129257602 2.599999636709697
2.599997708347497 2.599999636709697
2.5999983641270976 2.5999999418854713
2.5999988814964086 2.599999636709697
2.5999978331397577 2.599999636709697

In theory they should be equal, because:

Syntax: Select all

// Determine acceleration speed after acceleration
accelspeed = accel * wishspeed * gpGlobals->frametime * player->m_surfaceFriction;

// Cap it
if (accelspeed > addspeed)
accelspeed = addspeed;
User avatar
Hymns For Disco
Posts: 32
Joined: Wed Nov 22, 2017 7:18 am

Re: Difference in float numbers.

Postby Hymns For Disco » Wed Jan 22, 2020 7:54 am

My first thought is that gpGlobals->frametime is not guarenteed to be constant. Perhaps it changes to reflect the real time passed on a given tick, and not the ideal exact same time every tick. I haven't combed through the source sdk code to verify.

Also note that the wishspeed is derived from the length of wishvel which is derived from wishdir, which is derived through trigenometry from the player's view angle. So any variation in the player's view angle will have a knock-on effect causing every value in the acceleration to vary slightly.
User avatar
Project Leader
Posts: 1345
Joined: Sat Jul 14, 2012 9:29 pm
Location: Québec

Re: Difference in float numbers.

Postby L'In20Cible » Wed Jan 22, 2020 4:54 pm

Seems like a common loss of precision to me. Though, not sure why you would want to reverse-engineer the calculation if you already have the precise result?
Senior Member
Posts: 100
Joined: Tue Jul 03, 2018 3:00 pm
Location: *DELETED*

Re: Difference in float numbers.

Postby Sam » Thu Jan 23, 2020 4:27 am

Loss of accuracy in Python float is inevitable. For this, there is module `decimal`
Last edited by Sam on Thu Jan 23, 2020 4:27 am, edited 1 time in total.
Reason: Original post version
User avatar
Project Leader
Posts: 2098
Joined: Sat Jul 07, 2012 8:17 am
Location: Germany

Re: Difference in float numbers.

Postby Ayuto » Sat Jan 25, 2020 9:59 am

Can't tell you if your calculation is correct, because the relevant information is missing in your post, but this also seems like a loss of accuracy to me. This might surprise you:

Syntax: Select all

>>> 0.3 - 0.1
Junior Member
Posts: 25
Joined: Wed Jan 08, 2014 10:32 am

Re: Difference in float numbers.

Postby DeaD_EyE » Mon Jan 27, 2020 4:32 am

Visit f"http://{0.1 + 0.2}.com"

If you need to compare the values of equality, use math.isclose(a, b)
Problem solved :-D

Return to “Plugin Development Support”

Who is online

Users browsing this forum: No registered users and 9 guests