Post by @aesthr@wandering.shop

Anyone who doesn't know about this really shouldn't be speaking on what computers are supposedly good at

(This is not a bug. This is documented and expected behavior in pretty much every widely used programming language today. See: IEEE 754)

R console screenshot:

> 0.1 + 0.2 == 0.3
[1] FALSE — R console screenshot: > 0.1 + 0.2 == 0.3 [1] FALSE

0 Boosts

0 Likes

Ryan

@fullywoolly@mastodon.social 14 min. ago

@aesthr I always loved in python when it'd print out 0.3000... skip a bit ...001

Cassandra has had an adventure

@xgranade@wandering.shop 1 hr. ago

@aesthr I have acquired a healthy and respectful fear of using the equality operator to compare floats. It's just almost never what one wants, due to finite precision and the weirdness of how NaN works.

This is such a great and classic example of that.

Ferret

@ferret@hub.workersofthe.world 3 hr. ago

@Æ. bash seems cool with it

if [[ 0.1+0.2==0.3 ]]; then echo "true"; fi
true

Æ.

@aesthr@wandering.shop 3 hr. ago

@ferret yeah, but that's not great either, right? because it diverges from the behavior of standard floating point numbers that's established nearly everywhere else

Æ.

@aesthr@wandering.shop 3 hr. ago

@ferret actually, it's much worse

bash doesn't do floating point at all. I didn't know until today

the bad thing here is that it fails silently without a warning or error

esther@Jupiter:~$ if [[ 0.1+0.2==42 ]]; then echo "true"; fi
true — esther@Jupiter:~$ if [[ 0.1+0.2==42 ]]; then echo "true"; fi true

Miguel Grinberg

@miguelgrinberg@mstdn.social 29 min. ago

@aesthr @ferret bash can only do integer arithmetic, but zsh does floating-point. Both require arithmetic expressions to be wrapped in $(( ... )).

❯ if [[ "$((0.1 + 0.2))" == "0.3" ]]; then echo true; else echo false; fi
false
❯ echo "$((0.1 + 0.2))"
0.30000000000000004 — ❯ if [[ "$((0.1 + 0.2))" == "0.3" ]]; then echo true; else echo false; fi false ❯ echo "$((0.1 + 0.2))" 0.30000000000000004

Zack Weinberg

@zwol@masto.hackers.town 2 hr. ago

@aesthr @ferret it's even worse than that; this isn't doing arithmetic at all! there are no spaces in "0.1+0.2==42" and so `[[ 0.1+0.2==42 ]]` is interpreted as equivalent to `[[ -n "0.1+0.2==42" ]]`, i.e. test whether the *string* "0.1+0.2==42" is *not empty*

2+ more replies

Æ.

@aesthr@wandering.shop 3 hr. ago

@ferret it still fails silently even with `set -e` in place. awesome

Æ.

@aesthr@wandering.shop 5 hr. ago

And if you're a programmer and just learning about this now ... I'm so sorry, but someone had to tell you the truth eventually.

Katha

@kathakatze@chaos.social 2 hr. ago

@aesthr comparisons for equality without explicitly specifying the precision or maximum allowable error should have been made illegal in all programming languages instead of creating footguns.

this can actually work okay for nearly every case when doing it the right way, which is usually DIY for most environments and very easy to fuck up. may I (un)proudly present the C++ solution as an example: https://en.cppreference.com/cpp/types/numeric_limits/epsilon

std::numeric_limits<T>::epsilon - cppreference.com

etaoin shrdlu

@therealkuu@chaos.social 4 hr. ago

@aesthr
i mean, it _can_ work if you stay nicely aligned on the grid of powers of 2 or if you find a cursed bcd/decimal float

tibi

@tibi2@mstdn.social 4 hr. ago

@aesthr OK yeah screw programming I'm gonna stick to art