Uninitialised variable in print_exponential_number

[mattbucknall]

Hi - I get the following warning when compiling printf.c (master [8b831c1]):

printf.c:899:47: warning: 'abs_exp10_covered_by_powers_table' may be used uninitialized in this function [-Wmaybe-uninitialized]

I've not done any in-depth analysis of how print_exponential_number works, so I'm not sure if this really matters. Still, it would be nice if abs_exp10_covered_by_powers_table could be initialised with a default value to get rid of this warning.

[eyalroz]

This is a compiler "bug". Look at the logic:

  // ... snip ...
  if (abs_number == 0.0) {
    floored_exp10 = 0;
  }
  else  {
    // ... snip ...
    abs_exp10_covered_by_powers_table = false;
    // ... snip ...
  }
  // ... snip ...
  normalization.multiply = (floored_exp10 < 0 && abs_exp10_covered_by_powers_table);
  // ... snip ...

You see? It can't be used uninitialized. See it GodBolt also.

What compiler are you using?

[203Null]

I have the same issue with using this library with ESP-IDF

[eyalroz]

I have the same issue with using this library with ESP-IDF

I don't know what ESP-IDF means. Please file a bug against either the IDE or the compiler it uses to generate warnings.

[203Null]

I don't know what ESP-IDF means. Please file a bug against either the IDE or the compiler it uses to generate warnings.

It will not compile with -Wmaybe-uninitialized unless I default abs_exp10_covered_by_powers_table to false.

[203Null]

[eyalroz]

First of all, you're treating warnings as errors. That is your prerogative, but the library's build mechanism does not do that on its own.

Secondly, and again - a warning here is a bug in your compiler. There is no sense in complaining about it here. The code is correct. Please file a bug against the compiler.

[203Null]

First of all, you're treating warnings as errors. That is your prerogative, but the library's build mechanism does not do that on its own.

Secondly, and again - a warning here is a bug in your compiler. There is no sense in complaining about it here. The code is correct. Please file a bug against the compiler.

The only place that abs_exp10_covered_by_powers_table is defined is at

printf/src/printf/printf.c

Line 873 in 19a487c

abs_exp10_covered_by_powers_table = PRINTF_ABS(floored_exp10) < PRINTF_MAX_PRECOMPUTED_POWER_OF_10;

which is under an if-else statement. I don't see how this is a compiler error.

[203Null]

I see, the code won't check for abs_exp10_covered_by_powers_table if floored_exp10 == 0

[eyalroz]

@203Null : Exactly.

[203Null]

The compiler I used is Xtensa ESP32 gcc 8.4.0 (2021r2)
On GodBolt with the same compiler, it works fine https://godbolt.org/z/8W65Ga38n

I will make an issue in their Github repo

[203Null]

I saw you did some commits regarding this issue.

printf/src/printf/printf.c

Line 631 in f8ed5a9

#pragma GCC diagnostic ignored "-Wmaybe-uninitialized"

Those codes are actually fine with my compiler (not sure about others)

These lines in print_exponential_number are actually the ones causing issues for me. Porting your changes over solves it.

printf/src/printf/printf.c

Line 908 in f8ed5a9

normalization.multiply = (floored_exp10 < 0 && abs_exp10_covered_by_powers_table);