Confusion caused by use of -fpermissive in AVR and megaAVR compilation recipes

[allnick]

Hi. There is the problem.

=============================================================

void setup() {
  Serial.begin(9600);
  Serial.println("Start");
}

void loop() {
  char ch = Serial.read();
  switch(ch){
   case '1':
    Serial.println("Block 1 starts executing");
    // Some block 1
    break;
   case '2':
    Serial.println("Block 2 starts executing");
    byte b;
    b = 0; // It's OK.
    // Some block 2
    break;
   case '3':
    Serial.println("Block 3 starts executing");
    // Some block 3
    break;
   case '4':
    Serial.println("Block 4 starts executing");
    // Some block 4
    break;
  }
}

Sketch uses 1668 bytes (5%) of program storage space… Global variables use 294 bytes (14%) of dynamic memory…

=============================================================

void setup() {
  Serial.begin(9600);
  Serial.println("Start");
}

void loop() {
  char ch = Serial.read();
  switch(ch){
   case '1':
    Serial.println("Block 1 starts executing");
    // Some block 1
    break;
   case '2':
    Serial.println("Block 2 starts executing");
    byte b = 0; // This causes the problem! All blocks below the 2nd are never executed !!!
             // However, the value of the local variable in the 2nd block remains correct!
    // Some block 2
    break;
   case '3':
    Serial.println("Block 3 starts executing");
    // Some block 3
    break;
   case '4':
    Serial.println("Block 4 starts executing");
    // Some block 4
    break;
  }
}

Sketch uses 1596 bytes (4%) of program storage space… Global variables use 244 bytes (11%) of dynamic memory…

=============================================================

Note the noticeably smaller code size in the second case. Moreover, this does not depend on the number of blocks below the second one! It looks like the linker just doesn't include them in the final code! Bye.

[per1234]

Related: https://github.com/arduino/ArduinoCore-avr/issues/268

[MHotchin]

This is not valid C++ code, and should result in a compiler error. The standard says this program is ill-formed, not just undefined behaviour.

This exact scenario is called out the in the language standard: https://en.cppreference.com/w/cpp/language/switch

[per1234]

The reason it compiles is due to the use of the -fpermissive compiler option (which is why I referenced https://github.com/arduino/ArduinoCore-avr/issues/268 above). If you try to compile this code for one of the platforms that doesn't use -fpermissive (e.g., Arduino SAMD Boards), it does fail.

I suppose we could close this as covered by https://github.com/arduino/ArduinoCore-avr/issues/268. That is somewhat specific to the AVR platform, due to being in its repository, but I added a comment there noting that the same applies to the Arduino megaAVR Boards platform.

[MHotchin]

Is it possible to selectively re-enable errors that -fpermissive is hiding? Since it's producing results that are not helpful, allowing projects with this defect to compile is counter-productive - silently broken is much worse than in-your-face broken.

[allnick]

This is not valid C++ code, and should result in a compiler error. The standard says this program is ill-formed, not just undefined behaviour.

I agree. The following code solves the problem:

void setup() {
 Serial.begin(250000);
 Serial.println("Start");
}

void loop() {
 char ch = Serial.read();
 switch(ch){
  case '1':
   Serial.println("Block 1 starts executing");
   // Some block 1
   break;
  case '2': {
   Serial.println("Block 2 starts executing");
   byte b = 0; // Now it's OK.
   // Some block 2
   break;
  }
  case '3':
   Serial.println("Block 3 starts executing");
   // Some block 3
   break;
  case '4':
   Serial.println("Block 4 starts executing");
   // Some block 4
   break;
 }
}

But in the previous example, there is no compilation error, there is an incorrectly assembled final code.