Thanks! I have something else to suggest. Nothing critical, just small technical things for rare cases. Take a try to load this map: htic_map15.zip. It is a replacement for E1M1 of very well-known map we was using for benchmarking a lot. There will be two notable problems:

• If you are using not very powerful CPU, loading and re-loading map will take few seconds, while in Doom it takes just a couple of milliseconds. I have tracked down the reason: Heretic's (or is it comes from Doom 1.2?) P_GroupLines is extremely ineffective. Simple replacement with actual Doom implementation should fix it, and I believe it is safe for demos.
• Try to walk around and shoot using Wand. Almost immediately crash will happen, this is because of BLOCKMAP issue. Can be easily fixed by adding your implementation of blockmap recreating.

Should I take care about these?

I believe it is safe for demos.

We must be absolutely sure that it is safe for demos. But since DSDA-Doom also uses the same (optimized) P_GroupLines() for both Doom and Heretic I think we can be confident that it is.

Can be easily fixed by adding your implementation of blockmap recreating.

Well, that's Lee Killough's code, I just adopted it for Crispy. Sure, if we detect a corrupted BLOCKMAP, we should attempt to fix it.

Thank you for clarification. I was thinking about demos in a first place and did some tests for all five episodes using DSDA's demos for Heretic, everything was fine. This is also will be useful for loading/reloading of extreme maps like that map15, on original code it takes about 6-7 seconds on my quiet nettop's Intel 11th Gen N5095. On new one it's just 300-350 milliseconds. Absolutely astronomical difference.