When using Magnesium with CTM, All of CTM Blocks broken and disconnected.
[Magnesium] 1.5
https://www.curseforge.com/minecraft/mc-mods/sodium-reforged

I am experiencing the same issue. Connected textures randomly change.

Can confirm.
Also CTM emissive textures don't seem to work either.

Re-rendering chunks via F3+A fixes the issue.

I am inclined to think this is a Magnesium issue, but really I have no idea. The way CTM checks for neighbor blocks is very simple, so breaking it would require some very strange behavior changes. I leave it up to Magnesium to explain if/how I could fix it on my end.

It is definitely a Magnesium issue. Rubidium seems to have a similar problem, so I expect it comes from Sodium upstream (all share the same codebase if I'm not mistaken ..?)
Supposedly Rubidium mc1.16.5-0.2.6 fixes this according to its changelog on curseforge. Unfortunately, I haven't been able to get that version to run on my end in order to verify.

Here is the relevant Rubidium issue - Asek3/Rubidium#69. Looks like there's a commit for the change that fixes it. I'd be tempted to try the same fix in Magnesium, would fork it if I thought my Java skills were adequate. I'll go post the same link on that Magnesium issue thread from above

I’m having the same issue with Rubidium. As previously mentioned, F3 + A to reload chunks is a temporary fix. I hope this is fixed.

Chasing down #180, is the remark from Rubidium, here, relevant?

This appears to be caused by an issue in WorldSlice. Creating a new WorldSlice in ChunkRenderCacheLocal#init fixes this issue, though this would probably result in a fairly significant performance impact and an increase in memory usage. My guess is that the connected texture states aren't being saved or looked up correctly, or something along those lines.

This at least explains why it's fixed when reloading chunks, because new WorldSlice instances are created. I'll try to look into this more to see if it can be resolved, or if I can at least add more onto this issue.

Strangely enough, Connected Glass Mod works perfectly fine with Rubidium, so my only assumption is that something weird is happening with the way CTM handles block states, or at least the way it's stored in WorldSlice; it's hard to tell so far.

Just commenting to say that I'm also having this issue and would greatly appreciate an update if there is any! It seems like greysondn dug up something helpful?

For me this is occurring on -
Minecraft 1.19.2
Forge 43.2.11
Rubidium 0.6.2b
CTM 1.1.6+8

Can confirm still an issue on 1.19.2 (with Rubidium).
From what I gather,

• Looks like the old mixin fix from Asek3/Rubidium#22 (comment) (also referred to in Asek3/Rubidium#510) results in severe performance loss (per Asek3/Rubidium#22 (comment), sounds like rubidium becomes near-useless)
• The aforementioned possible fix from Asek3/Rubidium#160 (comment) is a rubidium-side workaround for what now seems like a CTM issue, again with an impact on performance, but I'm curious if better/worse than the mixin fix
• Mods that have connected textures and don't use CTM are fine, plus the quote from rubidium makes it seem like root issue is on CTM's side?

something weird is happening with the way CTM handles block states, or at least the way it's stored in WorldSlice

Gotta look for the source of this in CTM, it would seem.

has anyone made any progress on identifying the underlying issue with this one?

I believe the underlying issue here is CTM's RegionCache relying on an identity comparison to determine when a new chunk is being rendered:

This assumption is valid in vanilla (as it allocates fresh objects) but Rubidium/Sodium use object pooling to mitigate a high allocation rate. After making a simple modification in Rubidium to ensure that a unique BlockGetter is provided each time a subchunk is built, the issue appears to be resolved from what I can see. The fix should be linked above if anyone who has the skills is interested in applying it to their MC version.

This is the change I was hoping for. I tested disabling that cache for non-vanilla classes, but I was worried the reflective check would negate any performance gains. I believe it is pretty much assumed that the readonly chunk views won't be modified after they are passed.