AVR core 1.6.22 (IDE 1.8.6) crashes avr-gcc compiler (5.4.0) when F() string is assigned to a static variable

[bxparks]

The following program will crash the 5.4.0 avr-gcc compiler used in AVR core 1.6.22 (IDE 1.8.6). The problem does not happen in AVR core 1.6.21 (IDE 1.8.5):

void setup() {}

void loop() {
  static const __FlashStringHelper* F_STRING = F("fstring");
  Serial.println(F_STRING);
}

The error message is:

lto1: internal compiler error: in lto_output_varpool_node, at lto-cgraph.c:624
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.
lto-wrapper: fatal error: /home/brian/dev/arduino-1.8.6/hardware/tools/avr/bin/avr-gcc returned 1 exit status
compilation terminated.
/home/brian/dev/arduino-1.8.6/hardware/tools/avr/bin/../lib/gcc/avr/5.4.0/../../../../avr/bin/ld: error: lto-wrapper failed
collect2: error: ld returned 1 exit status
exit status 1
Error compiling for board Arduino Nano.

Interestingly, the following alternatives do not cause problems:

void setup() {}

static const char P_STRING[] PROGMEM = "pstring";

void loop() {
  Serial.println((const __FlashStringHelper*) P_STRING); // ok
  Serial.println(F("inline fstring")); // ok
}

My environment is Ubuntu Linux 17.10, but another person on the Arduino Forum thread reports the same error on Windows 10.

[bxparks]

A workaround that works for me is to declare the F_STRING as a const, not a non-const function static, like this:

void loop() {
  const __FlashStringHelper* const F_STRING = F("fstring");
  Serial.println(F_STRING);
}

That is what I had actually intended in my real code, but I seem to have gotten confused by the "static const" part in front of a pointer type.

[bxparks]

From the above mentioned Arduino forum thread, here's another example that produces the same compiler error, but without using a static keyword:

void loop() {
  const __FlashStringHelper* strings[] = { F("a"), F("b") };
  for (int i = 0; i < 2; i++) {
    Serial.println(strings[i]);
  }
}

My guess is that the compiler is performing an internal optimization that causes strings to become static. If the code is slightly rewritten, the compiler is happy:

void loop() {
  const __FlashStringHelper* const a = F("a"); 
  const __FlashStringHelper* const b = F("b");
  const __FlashStringHelper* strings[] = { a, b };
  for (int i = 0; i < 2; i++) {
    Serial.println(strings[i]);
  }
}

[matthijskooijman]

Note that the PSTR() macro (and thus also F()) implicitly creates a static variable too.

I wonder what happens if we replace the PSTR() macro with the one shown here: https://github.com/mmurdoch/arduinounit/blob/5d16d578008f665f4ec0a9cc0818423cf45ee1c9/src/ArduinoUnitUtility/Flash.h#L11-L27

That version of PSTR() is intended to fix deduplication of flash strings (arduino/Arduino#4863), but it also uses a completely different way of generating the string variables, so I suspect it might not trigger this internal compiler error. Aynyone care to test? This would involve copying the linked code into, I think, WString.h and update the F() macro to use it.

[Floessie]

Great findings and analysis, @bxparks.

I wonder, if anyone has upstreamed this ICE (and also Arduino issue 7949)... If so, I'd like to see a link to the upstream issue to follow it up there.

AFAIK, those problems were tackled in GCC 7 something, but the official Atmel toolchain lags behind (one can replace it with an up-to-date crosstool-ng generated, though).

Best,
Flössie

[Floessie]

Ping. Any news or plans?

[facchinm]

@Floessie can you take a look at arduino/Arduino#7949 (comment) ? Thanks!