I found the odetocode blog after idly googling demon fly out nose: 


As the author of the original "demons may fly out of your nose", I'd like 

to point out that the real point of the "undefined behavior" discussions 

was not to assert that bizarre behavior must necessarily result from 

incorrect or undefined code; it was to defeat the frequently seen argument

that "compiler X is junk because it doesn't generate correct code for my 

program", where "my program" inevitably included an instance of undefined 

behavior (dereferencing NULL, sizeof(struct undefined), or whatever). 

Merely replying "the standard says you shouldn't" never convinces these 

people; they always come back with "well, compilers Y and Z do what I 

expect, so obviously I'm right", followed closely by "how can the standard 

possibly permit a compiler to output broken code even if a construct isn't 

officially defined?" Thus there was an ever escalating insistence that 

"undefined" does not mean "whatever the user thinks would make his broken 

program work", it means "undefined by the standard and beyond the pale of 

the language".



Compilers aren't obligated to mistreat programmers who step outside the 

language definition, but they certainly aren't obligated to treat them 

kindly, and a compiler which "works" one release with an incorrect construct 

may not "work" the next release -- even though the compiler is, in fact, 

perfectly conforming in both cases. (Although there have been some 

programmers who I *wished* had compilers which formatted their hard 

drives when undefined language constructs were used...)



(I just idly decided to go googling for "nose demons compiler" to see how 

many references to the idea were still floating around.)