22

Is there a way to print the value of a constexpr or #defined value at compile time? I want the equivalent of std::cout <<, or some way to do something like

constexpr int PI_INT = 4;
static_assert(PI_INT == 3,
              const_str_join("PI_INT must be 3, not ", const_int_to_str(PI_INT)));

Edit: I can do some basic compile-time printing with constexprs, at least on gcc by doing something like

template <int v>
struct display_non_zero_int_value;

template <>
struct display_non_zero_int_value<0> { static constexpr bool foo = true; };

static constexpr int v = 1;

static_assert(v == 0 && display_non_zero_int_value<v>::foo, "v == 0");

which gives me error: incomplete type ‘display_non_zero_int_value<1>’ used in nested name specifier static_assert(v == 0 && display_non_zero_int_value<v>::foo, "v == 0");. (icpc, on the other hand, is less helpful, and just says error: incomplete type is not allowed) Is there a way to write a macro that can generalize this so that I can do something like

constexpr int PI_INT = 4;
PRINT_VALUE(PI_INT)

and get an error message that involves 4, somehow?

Jason Gross
  • 5,928
  • 1
  • 26
  • 53
  • 2
    Not posting this as an answer because I don't have evidence to hand, but I remember trying to do this in the past and I think the standard says that static_assert has to take a string literal, and as a consequence you can't use a constexpr expression instead. sorry. – je4d Nov 20 '12 at 01:56
  • Note that your fix isn't really using `static_assert` at all. It's just reinventing the basic idea of a construct that does nothing but pass or fail. The thing that does the "printing" also has to perform the test, so you're stuck solving the entire problem with SFINAE. – Potatoswatter Nov 21 '12 at 09:12

2 Answers2

16

Quoting the grammar given for declarations in §7/1 [dcl.dcl]:

static_assert-declaration:

static_assert ( constant-expression , string-literal ) ;

The standard says it has to be a string literal, so you're out of luck; you can't use a constexpr function to construct your error message.

You can, however, use whatever preprocessor magic you like to generate a string literal to go in there. If PI_INT is a #define instead of a constexpr int, you could use something like this:

#define PI_INT 4
#define pi_err_str_(x) #x
#define pi_err_str(x) pi_err_str_(x)
#define pi_int_err "PI_INT must be 3, not " pi_err_str(PI_INT)

static_assert(PI_INT == 3, pi_int_err);

output:

error: static assertion failed: "PI_INT must be 3, not 4"

Edit in response to comment by OP and updated question

Is there a way to write a macro that can generalize this so that I can do something like ... and get an error message that involves 4, somehow?

Sure, a bit of preprocessor magic can generalize that, assuming you're happy to be reliant on compiler-specific error message behaviour:

#define strcat_(x, y) x ## y
#define strcat(x, y) strcat_(x, y)
#define PRINT_VALUE(x) template <int> struct strcat(strcat(value_of_, x), _is); static_assert(strcat(strcat(value_of_, x), _is)<x>::x, "");

constexpr int PI_INT = 4;
PRINT_VALUE(PI_INT)

stackoverflow/13465334.cpp:20:1: error: incomplete type ‘value_of_PI_INT_is<4>’ used in nested name specifier

As for other compilers, I don't know what you can do offhand, but you may want to look at a copy of boost's static_assert.hpp to see if any of the tricks employed there can be used to get an evaluated template arg printed.

Community
  • 1
  • 1
je4d
  • 7,628
  • 32
  • 46
  • 1
    That answers the last two-thirds of my question. The first third, about printing `constexpr` values at compile time, isn't as obvious. See the edit I just made. – Jason Gross Nov 21 '12 at 05:40
  • Any way for this to work at block scope (for printing inside templated functions)? – mxmlnkn Feb 06 '18 at 14:38
  • 1
    @mxmlnkn for GCC, there exists a tool: https://github.com/Viatorus/compile-time-printer – Viatorus Feb 22 '21 at 08:59
0

You could evaluate constant expression as a parameter to a template, where would be some kind of warning printed with templates instantiation name with its arguments.

Trzeci
  • 11
  • Your answer could be improved with additional supporting information. Please [edit] to add further details, such as citations or documentation, so that others can confirm that your answer is correct. You can find more information on how to write good answers [in the help center](/help/how-to-answer). – Community Jan 11 '22 at 16:52
  • This is already present in both my question and the existing (accepted) answer. Did you have something in mind beyond what's already there? – Jason Gross Jan 12 '22 at 18:27