your luddite statement "debugger not useful" is now followed by
I meant JTAG debuggers like on Silabs 8051, Atmel AVR, NXP ARM and so on.
There is abolute no way to inspect something on the fly.
Whenever you want to see something, the CPU stops completely.

that one thing (inspect something on the fly) does not qualify that "debugger not useful". Your car can not brew beer, is it thus not useful?

I absolutotally LOVE the JTAG debuggers like on Silabs, can they "inspect something on the fly" yes, in a way they can and in a much less intrusive way that your "sending debug informtaion over the UART" can.

First of all an emulator is in most cases able to show you what you need to see, without "inserting code". I remeber the "insert code days" insert code, see something, insert code elsewhere learn a bit more, insert code elsewhere learn a bit more, insert code elsewhere learn a bit more, insert code elsewhere learn a bit more, .... what a cumbersome method.

Second if you do the "send the code if it is 47" a much less intrusive way is "store the code elsewhere if it is 47" and, after a while stop the emulator and watch "elsewhere".

Are there rare cases where inserting code is the only method possible?, of course, but stating that that makes other methods "not useful" is bullshit covered with horsefeathers.

The more tools you have in your toolchest, the more likely you will finish in a timely manner.

I just realize a thing about yout 'method' related to the original point re optimization: if you "insert code to inspect something on the fly" the optimizer, most likely will totally change the flow.

Erik