The format specifier %a for printf() in C

Question

I am reading a C language book, it said %f, %e, %g, %a were printf conversion specifications used for float and double data types. Currently I can understand %f, %e, %g completely.

When do I need use %a to print float and double type data ?

Can you please show me an example.

Privative

Answer 1

The %a formatting specifier is new in C99. It prints the floating-point number in hexadecimal form. This is not something you would use to present numbers to users, but it's very handy for under-the-hood/technical use cases.

As an example, this code:

printf("pi=%a\n", 3.14);

prints:

pi=0x1.91eb86p+1

The excellent article linked in the comments explains that this should be read "1.91EB86₁₆ * 2¹" (that is, the p is for power-of-two the floating-point number is raised to). In this case, "1.91EB86₁₆" is "1.5700000524520874₁₀". Multiply this by the "2¹", and you get "3.140000104904175₁₀".

Note that this also has the useful property of preserving all bits of precision, and presenting them in a robust way. For instance you could use this to serialize floating point numbers as text, and not have to worry about repeating/infinite decimals.

Also note that strtod() can convert floating point numbers in hexadecimal form back to actual numbers. Not 100% sure about sscanf() and friends, the documentation was not very clear and I've never used that.

Rubescent

You can read more about hexadecimal floating-point constants here: exploringbinary.com/hexadecimal-floating-point-constants . – Rhiana

Answer 2

As far as an example of why you would want to use the hex representation, you may want to use %a to precisely represent a floating point value being sent to another machine for processing.

We're using this currently for unit testing an embedded controller by sending data from a simulated plant model that is emulating sensors and actuators to the embedded processor via a UART where the embedded processor does it's control processing and returns feedback (again, float represented as %a) back to the plant model to close the loop.

Kipp

Just a heads up on embedded system, at least with the Atmel SAME70 %a and %A don't work – Trebuchet

Answer 3

When do I need use %a to print float and double type data ?

In addition to guaranteed exact representation, printing with %a is also much faster. When you are printing only a few numbers, that is not much of an issue, but if you are forced to format-print many floating-point numbers, this may well speed your work up. Switching to base 10 and handling rounding each take a not-insignificant amount of work.

Caveat: The I/O may take even more time than the formatting of characters, which may eat away at these benefits; and you might, instead, get away with just printing the raw bytes with no formatting.

Supplicatory