Specifies the default word size to be num bytes. This is the size of logical and single-precision numeric variables that are not given explicit precisions. (Explicit precisions for non-character variables are an extension to the Fortran 77 Standard, and are given by type declarations such as REAL*8 X.) Double-precision and complex variables will be twice this value, and double complex variables four times. Quad-precision constants and intrinsic function results will be four times this value. Note that variables declared as REAL*16 will be regarded as quad precision only if the word size is 4 bytes. Default = turn-on = 4 bytes.

The word size value does not matter for checking standard-conforming programs that do not declare explicit precisions for non-character variables or store Hollerith data in variables. This setting also does not affect the default size of character variables, which is always 1 byte. Hollerith constants also are assumed to occupy 1 byte per character.

The word size is used to determine whether truncation occurs in assignment statements, and to catch precision mismatches in subprogram argument lists and common block lists. The exact warnings that are issued will depend on the status of other flags. Under both the -portability=mixed-size and the -nowordsize flag, any mixing of explicit with default precision objects (character expressions not included) is warned about. This applies to arithmetic expressions containing both types of objects, and to subprogram arguments and COMMON variables. Under control of the -truncation=demotion and promotion options, a warning is given for assignment of an expression to a shorter variable of the same type, or for promotion of a lower precision value to higher precision in an arithmetic expression or an assignment statement.

Giving a word size of 0, or equivalently, using -nowordsize means that no default value will be assumed. This is equivalent to specifying -portability=mixed-size. Use it to find cases of mixing default and explicit precision, for example to flag places where REAL*8 is treated as equivalent to DOUBLE PRECISION.

See also: -pointersize, -portability, -truncation.

Next: -wrap