What's impossible about it? I've seen devices in the past that can combine multiple HDMI signals into a single image (like picture-in-picture, or making one high-res display out of several lower-res video connections). And if you can do that, surely you can do instant switching just by toggling between "full screen input 1" and "full screen input 2" layouts.
Were any of these devices monitors? Were they cheap? Like it or not, price is a major factor.
On top of that, just look around - there is no perfect display,no matter how much one is willing to pay. Every single one has major drawbacks in certain areas.
What are the hardware limitations? Surely this is possible if there was investment and interest in making this work? Or do the protocols really not allow this?
Most of the delay is I/O bound. Handshaking happens over the Display Data Channel (DDC) which is a protocol over I²C (a two-wire comunication bus). HDMI/DP/DVI have 2 dedicated wires inside the cable for this, USB-C has to switch over to use differential signaling on pins A8 and B8, and then switch back after communication is done.
The bandwidth of that bus is small indeed, but most of the latency comes from artificial delays and error recovery in code: monitor is sending a message to host then sleeping 30ms to make sure the reply is ready, host doing the same thing, and this happens multiple times to negotiate pixel clock, signal timing, colors, resolution, additional features like suspend/reset/HDR/overscan etc.)
You can see how in the MCCS specification [1] both host and display entrypoints for communication start with a delay: https://shots.panaitiu.com/e8D7tQ
You can also see how that looks in Lunar's DDC code [2] [3].
For now, it's incredible that we actually have such a standard as DDC and most monitors work with most devices. A coordinated effort between monitor vendors and laptop/PC/GPU manufacturers into creating another faster standard with modern technology seems out of the question. But maybe with the advent of Thunderbolt this may change in the future.
Naively it seems like you could optimise this at least for the second and subsequent switches: you could detect if the cable corresponding to the input you've previously switched to has been reconnected since last switch. If it hasn't, then you could re-use the previous EDID values for the device.
You'd obviously want this to be optional for special cases, but would that work, if implemented in a monitor or KVM?
It's not so much the delay that's annoying, but the renegotiation with something which is so obviously the same device.
The problem that OP is facing is switching to an inactive input that went into power saving mode. In that case, the link between the inactive device and the monitor may have been broken and it's impossible to know if the device is the same as the one before after reconnection (with the same exact video settings etc.).
This doesn't have to be the case though. Before I found out how to use I²C on Apple Silicon Macs, I was using a Raspberry Pi connected to an HDMI input of the monitor that could listen to DDC commands through an HTTP server: https://alinpanaitiu.com/blog/journey-to-ddc-on-m1-macs/#the...
And because I didn't want that Pi Zero to get hot, I managed to only keep the DDC lines open and stopped the video signal from being sent using `vcgencmd display_power 0`.
With those settings, the monitor still switched between inputs in 1 second, with the added small latency of having to run `vcgencmd display_power 1` (which was fast enough once it was automated to run on input switching)
In theory you could have a monitor that handles both signals all the time, buffering the output. Then it could switch immediately.
It would however be more expensive, since you would need to duplicate at least some of the electronics within the monitor, a few of which also incur licensing/certification/patent/whatever fees.
Why would you buffer? Monitors are capable of playing full-screen video, which implies the ability to update every pixel once per... 30hz? frame, doesn't it?
Because when you want instant switching you need to already have the data you want to display available the moment someone flips the switch.
You can't just tell the computer that you're now interested in the display data of the other display - that would incur latency.
You'll have to be receiving the information for the second display the entire time, storing it in what is commonly called a framebuffer, aka a bitmap in memory somewhere.
Most types of display have such a thing backing them anyways - one way or another - since they need to remember what they're supposed to display.
Okay, I'm willing to wait 1/30th of a second while the screen gets its pixels; is that not an option? Or is there no way to request all pixels and it doesn't send them all every frame?
With basic analog you could set it up in a way where you would just wait for the next frame (or don't even wait that long), but with the new digital stuff there's bootstrapping/handshaking - whatever you want to call it. It's not so simple anymore.
Quite a few monitors support a Picture-in-Picture mode, so they are able to receive and decode two signals at the same time. But if that mode isn't enabled, they only receive one signal.