There are more warnings that fail occasionally, especially if one sometimes for a good cause departs from "guidelines" (and, as you said, still might be correct) - that's why compilers have pragmas to disable warnings *locally* (OK this does not work particularly well on SDCC I found out in the mentioned experiment). I see no problem in this.

-------------


I don't understand.

Let's talk about this scheme:

• I declare a variable critical.
  
• if this is a single-byte variable, this is ignored by compiler... (OK this was simple :-) )
  
• when this variable is used,
  
  
  • if this is in a section of code () tagged already as critical, it is used normally
    
  • otherwise interrupts are disabled (interrupt flag stored) before variable usage and restored after it
    
    
    • extension: alternatively, user defined "disable"/"enable" function/inline function/macro can be used
      

I see no problem with this except it might not be too optimal - but who told C programs shall be optimal? There is no problem to fall back to "standard" "manual" disable/enable - simply don't flag the variable as critical.

Can you please point me out what's fundamentally wrong on this scheme (there still might be problematic *details* of course - I am not a compiler maker).

JW