Annex B. Implementation Limits - C, Ocean Edition I

B.0 General

An implementation shall be able to translate and execute at least one program that meets every limit in clause B.1 simultaneously.

An implementation should impose no fixed limit at all, and should be bounded only by available storage. Where a fixed limit exists, the implementation shall document it, and shall produce a diagnostic naming the limit when a program exceeds it. Silently truncating, wrapping, or miscompiling a program that exceeds a limit is not conforming behavior.

The values below are minimums. They are raised from the ISO C99 values, because the ISO values describe machines that no longer translate anything.

B.1 Translation Limits

Limit	Minimum
Nesting levels of blocks	256
Nesting levels of conditional inclusion	256
Pointer, array, and function declarators modifying a type	32
Nesting levels of parenthesized declarators	128
Nesting levels of parenthesized expressions	256
Significant initial characters in an internal identifier or macro name	1024
Significant initial characters in an external identifier	1024
External identifiers in one translation unit	65535
Identifiers with block scope declared in one block	4095
Macro identifiers simultaneously defined in one translation unit	65535
Parameters in one function definition	256
Arguments in one function call	256
Parameters in one macro definition	256
Arguments in one macro invocation	256
Characters in one logical source line	65535
Characters in a string literal, after concatenation	65535
Bytes in one object	1048576
Nesting levels for `#include` files	128
`case` labels in one `switch`, excluding nested switches	8191
Members in one structure or union	8191
Enumeration constants in one enumeration	8191
Nesting levels of structure or union definitions in one declaration list	128
Nesting levels of `typedef` chains	64

Significant characters in identifiers.

An implementation shall treat at least 1024 initial characters as significant for both internal and external identifiers, and shall treat case as significant in both.

Difference: ISO C99 requires only 31 significant characters for an external identifier and permits case to be ignored there. That allowance described linkers that no longer exist, and it made two distinct identifiers silently the same entity. Ocean Edition I removes it.

B.2 Numerical Limits

B.2.1 Integer types

<limits.h> shall define the macros listed in clause 7.2.8. The following values are fixed in every implementation:

CHAR_BIT    8
CHAR_MIN    0
CHAR_MAX    255
UCHAR_MAX   255
SCHAR_MIN   -128
SCHAR_MAX   127

For every other signed integer type of width N, the minimum and maximum are exactly:

minimum   -2^(N-1)
maximum    2^(N-1) - 1

For every unsigned integer type of width N, the maximum is exactly 2^N - 1.

Minimum widths, which the selected data model may exceed:

Type	Minimum width
`char`, `signed char`, `unsigned char`	8
`short`, `unsigned short`	16
`int`, `unsigned int`	16
`long`, `unsigned long`	32
`long long`, `unsigned long long`	64

An implementation targeting a hosted environment should provide an int of at least 32 bits. The 16-bit minimum is retained only for freestanding targets where a wider int would be a real cost rather than a preference.

Difference: ISO C99 permits INT_MIN to be -32767 rather than -32768, so that sign-magnitude and ones-complement machines could conform. Two's complement is required here, so the extra value always exists and shall be reported.

B.2.2 Exact-width types

Where the target has an integer type of width 8, 16, 32, or 64, the implementation shall provide the corresponding exact-width types of clause 7.2.20. Because there are no padding bits, an exact-width type exists whenever a type of that width exists.

B.2.3 Floating types

<float.h> shall define the macros listed in clause 7.2.5. Minimum requirements:

Macro	Minimum magnitude
`FLT_RADIX`	2
`FLT_MANT_DIG`	24
`FLT_DIG`	6
`FLT_MIN_10_EXP`	-37
`FLT_MAX_10_EXP`	+37
`FLT_MAX`	1E+37
`FLT_MIN`	1E-37
`FLT_EPSILON`	1E-5
`DBL_MANT_DIG`	53
`DBL_DIG`	10
`DBL_MIN_10_EXP`	-37
`DBL_MAX_10_EXP`	+37
`DBL_MAX`	1E+37
`DBL_MIN`	1E-37
`DBL_EPSILON`	1E-9
`LDBL_MANT_DIG`	53
`LDBL_DIG`	10

An implementation should map float to IEC 60559 binary32 and double to binary64 wherever the target supports them.

B.3 Library Limits

Limit	Minimum
Functions registerable with `atexit`	32
Simultaneously open files, `FOPEN_MAX`	8
Characters in a file name, `FILENAME_MAX`	260
Distinct temporary file names, `TMP_MAX`	25
Bytes in a multibyte character, `MB_LEN_MAX`	4
Signals distinguished by `<signal.h>`	6
Characters produced by one `printf` conversion	4095

MB_LEN_MAX is at least 4 because UTF-8 is the default execution character set and a UTF-8 sequence occupies up to four bytes.

B.4 Recommended Practice

An implementation should replace every fixed limit with a limit derived from available storage, and should report exhaustion as a diagnostic that names the resource. A program that fails to translate should never fail silently, and should never translate into something other than what it says.