A lightweight printf implementation. With some reasonable limitations, and non-standard behavior suited to microcontrollers.
If you need a full featured printf(), I strongly recommend checking out eyalroz's fork of mpaland's printf(), which can be found here. After some involvement in that project, I realized it would never quite be exactly what I wanted. So I wrote this. Some of the ideas used in this are from the authors of that project. Particularly the function pointer output from mpaland, and the output wrapping used in this was suggested by eyalroz.
-
Thread and re-enterant safe.
-
Low stack & ram usage, zero heap usage.
-
Full support for AVR's PROGMEM requirements, with almost no cost to non-AVR targets.
-
Compatible enough to make use of GCC's format and argument checking (even for AVR).
-
Optional support for long long types.
-
Optional support for float or double.
-
Full test suite using Greatest.
-
NO exponential form, %e provides SI units (y z a f p n u m - k M G T P E Z Y).
-
NO Octal, %o outputs binary instead, (who wants octal?)
-
NO adaptive %g %G
If you're a linux user, for a demonstration you can type 'make' then './demo' in the /demo folder.
Standard placeholder syntax:
%[flags][width][.precision][length]type
Supported [flags]:
- left align the output
+ prepends a + for positive numeric types
(space) prepends a space for positive numeric type
0 prepends 0's instead of spaces to numeric types to satisfy [width]
Unsupported [flags]:
# If you want 0x it needs to be in your format string.
' (apostrophe) No 1000's separator is available
[width]
The minimum number of characters to output.
If width is specified as * A dynamic width must be provided as an int argument preceding the argument to be formatted.
A non-standard method of centering text is also provided (see below).
[.precision]
For float %f, this is the number of fractional digits after the '.', valid range is .0 - .8
For integers, this will prepend 0's (if needed) until the total number of digits equals .precision
For strings %s %S, This is the maximum number of characters to read from the source.
If precision is specified as .* A dynamic precision must be provided as int argument preceding the argument to be formatted.
Centering strings: If [width] is specified and precision is .0 %s arguments will be centered.
Example to center within a 16 character LCD would be "%16.0s"
**Caution - If you are generating formatting strings at runtime, and generate a %[width].0s, you will NOT get 0 characters.
A dynamic precision of 0 provided to .* will not centre the string.
Supported [length]:
Used to specify the size of the argument. The following is supported:
hh Expect an int-sized argument which was promoted from a char.
h Expect an int-sized argument which was promoted from a short.
l Expect a long-sized argument.
ll Expect a long long sized argument. ** DISABLED BY DEFAULT, YOU MUST #define PRNF_SUPPORT_LONG_LONG in prnf_conf.h
z Expect a size_t sized argument.
t Expect a ptrdiff_t sized argument.
Unsupported [length]:
j intmax_t not supported
Supported types:
d,i Signed decimal integer
u Unsigned decimal integer
x,X Hexadecimal. Always uppercase, .precision defaults to argument size [length]
o NOT Octal. Actually binary, .precision defaults to argument size [length]
s null-terminated string in ram, or NULL. Outputs nothing for NULL.
S For AVR targets, read string from PROGMEM, otherwise same as %s
c character
f,F Floating point (optional), enable PRNF_SUPPORT_FLOAT or PRNF_SUPPORT_DOUBLE in prnf_conf.h
NAN & INF are always uppercase.
Default precision is 3 (not 6).
Digits printed must be able to be represented by an unsigned long (or unsigned long long if supported).
ie. with a precision of 3 using 32bit long values, maximum range is +/- 4294967.296
ie. with a precision of 3 using 64bit long or long long values, maximum range is +/- 18446744073709551.615
Values outside this range will produce "OVER".
A value of 0.0 is always positive.
e NOT exponential. Floating point with engineering notation (y z a f p n u m - k M G T P E Z Y).
Number is postpended with the SI prefix. Default precision is 0.
Unsupported types:
g,G Adaptive floats not available
a,A Double in hex notation not available
p Pointer not available
n classic %n is not available, but %n may be repurposed for extensions if enabled (see below)
The symbol PLATFORM_AVR must be defined, either before including prnf.h, or preferably by adding -DPLATFORM_AVR to compiler options. prnf.c will #include itself for a second pass, to build the _P versions of the functions. If you wish to write code which is cross compatible with AVR and non-AVR, use the _SL (String Literal) macros provided, these will place string literals in PROGMEM for AVR targets and produce normal string literals for 'normal' von-newman targets.
On AVR, both the the format string, and string arguments, may be in either ram or program memory.
Unlike avr-libc, prnf DOES provide argument type checking using some macro tricks. This works by putting a do-nothing function which takes format checking into a while(0) loop. The compiler will perform the format checking, then remove the do-nothing function, providing some level of optimization is enabled.
Create a function which handles a single character, in the following form:
void my_character_handler(void* x, char c)
{
(void)x; // As x is not used, this line may be needed to avoid a compiler warning.
// Do stuff here to output character c, ie. write to uart ect.
}
Then at the start of your program assign the default prnf() output to the above handler with:
prnf_out_fptr = my_character_handler;
A non-standard function is available which can be used for stream-like printing to different destinations.
int fptrprnf(void(*out_fptr)(void*, char), void* out_vars, const char* fmtstr, ...);
So to print to the above example of my_character_handler:
fptrprnf(mycharacter_handler, NULL, "Hello Fred is %i years old\n", freds_age);
If you have specific information to pass to your character handler, you can pass the address of it instead of NULL, and it will be passed to the void* x in my_character_handler.
Firstly, I would advise NOT doing this. Another reader of your code may see printf() and expect it to behave exactly like the standard printf(). If you return to your code to make changes in 5 years, that other programmer may be you.
If you still intend to override printf() this can be done with macros. Be aware that after you define 'printf' you will break anything which tries to add the format checking attribute attribute((format(printf, 1, 2)))
Another approach is to use GCC's --wrap feature in the compiler flags, which is probably better.
The usual functions are available (with print shortened to prn), and return a character count (disregarding any truncation).
int sprnf(char* buffer, const char* format, ...)
int snprnf(char* buffer, size_t buffer_size, const char* format, ...)
The following is useful for debug/diagnostic and cross platform friendly with AVR:
#define DBG(_fmtarg, ...) prnf_SL("%S:%.4i - "_fmtarg , PRNF_ARG_SL(__FILE__), __LINE__ ,##__VA_ARGS__)
Example usage:
DBG("value is %i\n", 51);
The above will output something like "main.c:0113 - value is 51"
Note that even if you use this on many lines, the FILE string literal will only occur once in the string pool. So it won't eat up all your program memory space.
The PRNF_ARG_SL() macro simply puts string literals into program memory on AVR targets or RAM on normal targets.
This can easily be achieved by writing your own character handler, and variadic function for the module. Then the variadic version of fptrprnf (vfptrprnf) can be used to print to your character handler.
An example for lcd_prnf() may look something like:
#include <stdarg.h>
// LCD prnf character handler
static void prnf_write_char(void* nothing, char x)
{
lcd_data(x); // (function or code to write a single character to the LCD)
}
// formatted print to LCD
void lcd_prnf(const char* fmt, ...)
{
va_list va;
va_start(va, fmt);
vfptrprnf(prnf_write_char, NULL, fmt, va);
va_end(va);
}
Safely APPEND to a string in a buffer of known size.
int snappf(char* dst, size_t dst_size, const char* fmtstr, ...);
In addition to the regular snprnf(), which is used for safely printing to a buffer of n bytes. snappf() can safely append to the buffer. This is useful in situations where you may need to iterate through a number of fields in a loop. The return value is the number of characters appended (ignoring truncation).
To enable this feature put -DPRNF_COL_ALIGNMENT in your compiler flags
It is possible to advance output to a specific column, with respect to the start of the output, or the last line ending. To achieve this prnf hijacks the \v (vertical tab) character.
The required format is:
\v<col><pad character>
\v should be followed by a decimal number indicating the column on which the following text will start on (with 0 being the first column). The character used for padding is the first non-numeric character after this number. Note that due to this, digits cannot be used as the padding character. This feature is useful if you have output which contains fields of uncontrolled length, and then wish to align further output. If the current column is already at or past the column specified, then no padding will be applied. If \v occurs in your string without being followed by digits, then a regular \v character will be output.
Example:
prnf("%s:%.4i(%s)\v30 %s\n", __FILE__, __LINE__, __func__, "This text starts on column 30");
Will yield something like:
myfile.c:0041(main) This text starts on column 30
It can also be used as an easy way to create banners, for example:
prnf("\v30*\n Main Menu\n\v30*\n");
Will yield:
******************************
Main Menu
******************************
THE FOLLOWING FEATURE IS DISABLED BY DEFAULT DUE TO HEAP INTERACTION
Most applications need to print things beyond what is offered by the standard placeholder types. This feature provides an easy way of doing that, but it requires the heap (or some other dynamic memory allocator) to work.
When enabled, prnf() will accept a %n as a C string (like %s), only it will free() the address after printing. The argument type is int*, even though it points to a string. This allows the user to create functions which produce whatever strings they need (coordinates, timestamps, etc..), and then use these functions as arguments to prnf(). Example:
static int* prext_bananas(int bananas)
{
int txt_size = 25;
char* txt = malloc(txt_size);
if(bananas > 10)
snprnf(txt, txt_size, "LOTS OF BANANAS!");
else if(bananas > 1)
snprnf(txt, txt_size, "A few bananas");
else if (bananas)
snprnf(txt, txt_size, "Only one banana");
else
snprnf(txt, txt_size, "NO BANANAS! :-(");
return (int*)txt;
}
prnf("Gorilla Fred has %n and Uncle Bob has %n\n", prext_bananas(freds_bananas), prext_bananas(bobs_bananas));
As these functions need to return an int* to a string on the heap, their name should indicate that their sole purpose is to provide arguments to prnf() for %n placeholders. Note that if you mistakenly mix up %s and %n, you will get a compilation warning (a good thing), as %s expects a char* and %n expects an int*
Q. Why cannibalize %n when there are so many letters left in the alphabet?
- GCC (and probably most compilers) will check the format string and issue a warning "unknown specifier %"
- Unsupported specifiers offer no argument type checking.
- The standard behavior of %n does not appear to be useful, (if you do use it, please share why).
Q. Why not just make functions to output the custom strings instead of passing them to prnf?
A. Consider you use vprnf to create yourself a uart_prnf() lcd_prnf(), and also use snprnf(), snappf(), and fptrprnf(). You can pass your custom strings to any of these functions.
- Create a configuration file similar to prnf_cfg.h in this repository.
- Add -DPRNF_CFG_FILE=prnf_cfg.h to compiler flags, avoiding spaces around the =
- Add your own extensions file similar to prext.c and prext.h in this repository.
Please raise an issue if you find a bug (unlikely), or even if you find any of the above instructions confusing. I'm willing to help.