[BUG] double/float comparison not working

[Talon1024]

GZDoom version

GZDoom g4.8.1

Which game are you running with GZDoom?

Doom 2

What Operating System are you using?

Linux x86_64

Please describe your specific OS version

Kubuntu 20.04 LTS

Relevant hardware info

Intel Core i3-3110M CPU, Intel HD Graphics 4000 GPU

Have you checked that no other similar issue already exists?

• I have searched and not found similar issues.

A clear and concise description of what the bug is.

When doubles are compared to floats, the result is incorrect.

This has caused an issue in Wolfenstein: Blade of Agony where the player's disguise item makes them unable to move through doors or gates while crouching. I did a quick workaround, but that doesn't solve the issue.

The comparison works properly on GZDoom 4.7.1 because it converts the scale to doubles, but it does not work properly on GZDoom 4.8.1.

Here's a small mod demonstrating the issue.

Steps to reproduce the behaviour.

Explain how to reproduce

1. Launch GZDoom with the test mod provided above, and start a new game.
2. Enter give ScaleCompareTest into the console.
3. If it works properly, you should see "Owner's scale is NOT 0.650." only once.

Your configuration

No response

Provide a Log

GZDoom 4.7.1 (correct):

MAP01 - Entryway

]give scalecomparetest
Owner's scale is NOT 0.650.
]give scalecomparetest
Owner's scale is already 0.650. Doing nothing
]give scalecomparetest
Owner's scale is already 0.650. Doing nothing

GZDoom 4.8.1 (incorrect):

MAP01 - Entryway

]give scalecomparetest
Owner's scale is NOT 0.650.
]give scalecomparetest
Owner's scale is NOT 0.650.
]give scalecomparetest
Owner's scale is NOT 0.650.
]give scalecomparetest
Owner's scale is NOT 0.650.

[Blue-Shadow]

I believe this is already fixed in 4.8.2. So try that.

[Talon1024]

No dice. I just tried this in GZDoom 4.8.2, and float/double comparisons don't work properly in 4.8.2 either.

[MrRaveYard]

I think this a float vs double precision issue.

float:
0.64999997615814208984375000000000000000000000000000000
double:
0.65000000000000002220446049250313080847263336181640625

[Talon1024]

Was scale changed internally from double to float between GZDoom 4.7.1 and 4.8.0?

[MrRaveYard]

Yes f9bf6f9

[OrdinaryMagician]

Does it work if ~== is used instead of just ==?

[MrRaveYard]

Does it work if ~== is used instead of just ==?

Nope

[coelckers]

Looks like the scale vector needs to be reverted to double. Well, here goes the chance of saving some space in AActor... :(

[SanyaWaffles]

That's a shame, but it seems this unfortunately breaks stuff too much with no obvious workaround.

[Talon1024]

Looks like the scale vector needs to be reverted to double. Well, here goes the chance of saving some space in AActor... :(

Why not convert/coerce both values to float if either value being compared is a float?

[coelckers]

Currently there are no float opcodes whatsoever, this would require a lot of support code.
I'm sorry, but as I see it, single precision floats in ZScript are not viable for anything that requires precise comparisons.

[coelckers]

fixed in aae1d14

[SanyaWaffles]

Still a few floats in actor.zs causing zscript compilation to fail.

EDIT: #1666