Actually this is VERY related to MC-53850. Also confirmed for 1.8.6. I'll post what I posted in that bug report (which the comment is here: https://bugs.mojang.com/browse/MC-53850?focusedCommentId=228736&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-228736)
Played with this a bit, pretty positive I know what's going on.
When an item is found to be in fire, it takes an initial hit of 1 damage every game tick, and also lights it on fire which does 2 damage every second. The client is ignoring the Invulnerable tag for the initial fire damage hit and "killing" the item in 5 game ticks when it gets to Health:0. So while the server sees the Invulnerable tag and keeps it alive at 5 health, the client sees it in fire, goes Health:4..3..2..1...and kills it at 0, making it invisible.
So for this bug report, even though the fire is quickly removed when lighting a glowstone block, it is still taking damage and getting "killed" by the client (not the server) when it reaches Health:0.
Played with this a bit, pretty positive I know what's going on.
When an item is found to be in fire, it takes an initial hit of 1 damage every game tick, and also lights it on fire which does 2 damage every second. The client is ignoring the Invulnerable tag for the initial fire damage hit and "killing" the item in 5 game ticks when it gets to Health:0. So while the server sees the Invulnerable tag and keeps it alive at 5 health, the client sees it in fire, goes Health:4..3..2..1...and kills it at 0, making it invisible.
Confirmed for 1.8.6
Also, related to MC-4438 which is another bug to do with fire turning items invisible.
So according to MC-56035 this is actually happening for player on players too! So the game is still detecting player collision and updating this stat, even though you can't push other players anymore since they removed player collision foreeevvveeer ago.
Actually it may have been, I just didn't feel like responding to this old issue so it got auto-closed after no response... It mostly affected my dealing with falling sand spawners, but those aren't too used anymore so I didn't care much about it. But ya this is still probably an issue.
It's mostly so that everyone logging into a fresh world won't be all spawning directly on top of each other and are spread out a bit. It's a good feature. It's always been around for multiplayer servers. It's actually MORE consistent this way since there aren't different behaviors between multiplayer and single player now. It's not a big deal to respawn in a random area around worldspawn in single player.
ya clientside it seems to be acting properly
Actually, this bug's description/title is wrong. The real bug is that tnt entity is actually in the dirt block but displayed above 1 block on client-side.
Ya this is true. A TNT entity that is within a dirt block is of course going to blow up that dirt block, and be able to chain the explosion to other explode-able blocks around it. A TNT entity is not an item entity, so it's not going to get pushed to the side or anything, it's just going to sit inside that block until it explodes. The bug here is the visual TNT is getting displayed on top of the block instead of just visually staying inside of it.
This is what it looks like in-game: http://i.imgur.com/7aGyIry.png
But quickly logging out and opening MCEdit, you can see the TNT entity (red box) is sitting on top of the dispenser, within the obsidian block: http://i.imgur.com/i5OdLVP.png
Dupe of MC-8565.
Nope. That one is about dispensing TNT directly into a block.
This is new with 1.8, it has to do with the "hop" when you trigger a tnt block not factoring in ceiling collision.
Could be. But I don't think so. Pressing the button and quickly logging out and checking the entity in mcedit shows it popping up into the block http://i.imgur.com/rYyKBOS.png
Though I've seen weird issues where things like this may only happen in a single player world and may work properly on a server. But I haven't tested that.
Confirmed, and I'm pretty sure I know why this happened. When they ported the block models over to the resource pack system, the models for repeaters and comparators accidentally ended up reversed. So, remember this issue for one snapshot where repeaters and comparators were getting placed in the wrong direction? https://bugs.mojang.com/browse/MC-56951 Well this is when it happened.
So instead of correctly turning the models around, comparators and repeaters were switched to function with this new backwards orientated block model. But in doing this, the blockstate names stayed the wrong direction, which is the reason for this bug.
Updated Gif to http://gfycat.com/LimitedAppropriateAmurminnow
Confirmed in 1.8.1-pre5
Does anyone know the exact steps to reproduce this?
Confirmed in 1.8.1-pre4
Confirmed for 1.21.1
Creating a custom map above cloud height and encountered this. I set the min_y back to 0 and it started working again.