When merging three belts whose throughput are respectively 1/4, 1/4 & 1/2, a random bug makes that the outbound belt is not always 1. We measured a lost of 6.25% (1/16). This behavior is not systematic, it can occur after a moment. More details here : #1383789390790398085 message
We have tested it with a miner (whose blueprint is available in the linked thread), but seems to occur everywhere this pattern is used. 3 ways splitting seems also to be an issue (it naturally splits at throughput 1/3, 1/3, 1/3, but issue can occur when the consumption of each belt is 1/4, 1/4, 1/2)
#[0.1.0-pre8-rc1] simulation quirk with 3 ways merger
wise iris
wise iris
I will test it, but I saw nothing about it changelog, so I assume yes
Well you did the measure actually 🙂
tawdry salmon
No. The 6.25% was something completely different iirc. This would only cause tiny gaps occasionally in the output. (Which I haven't experimented with enough to give any numbers (but it's a tiny loss which compounds over repeated mergers into noticeable gaps))
compact isle
Well you did the measure actually 🙂
What exactly are you referring to? I couldn't yet replicate or evaluate this bug, you discovered it when comparing different miners and concluded the 3-way mergers and splitters are problematic. They get clogged up which loses efficiency for no evident reason.
Me tossing around a 6.25% delay is in relation to a different and less important topic, it's not about this bug. Unless you coincidentally measured the same value, we shouldn't get them mixed up 😅
wise iris
Ah ok I misunderstood
tawdry salmon
Finding a way to reproduce this (without just leaving that miner running overnight) is turning out to be really hard. 🫤
wise iris
Best way I found was wait stops, but it just highlights the difference of production between floors. If quirks appear equally across floors, I'm not sure how to do.
tawdry salmon
The closest thing I have found so far involves getting a space belt behind the merger to become saturated when that should be impossible with 100% throughput everywhere. (But I can't get it to work reliably yet)
Also it may have something to do with onscreen/offscreen differences (like being offscreen may exaggerate it).