C#ifdef name C#ifndef name
statements; see below for details.
fpp is a preprocessor for Fortran and SFTRAN3, modelled on the ANSI C preprocessor, but tailored for Fortran use, and for reversibility. It may serve as a prototype for a future SFTRAN3 preprocessor conditional and macro facility.
These conventions ensure that the preprocessing is reversible; the output always contains the complete input, except that some code sections may have become comments, or vice versa. This is more useful than the non-reversible approach used by the C preprocessor.
Reversible preprocessing is convenient when a master source file must be maintained to generate multiple versions, such as for different operating-system, compiler, and architecture variations. The source code for any of these can serve as the master file to create any of the others.
The current implementation is written in nawk(1). That language is available on several different operating systems, in both commercial and free implementations, and serves as a convenient prototyping language for a program such as fpp.
C#name C#name args
Since all fpp directives are encoded as comments, both input and output files should be compilable without any preprocessing by fpp.
Blanks may optionally surround the initial # to permit indentation for better visibility, or to reflect conditional statement nesting.
The first column may contain any valid Fortran comment starter: C, c, or *.
Unrecognized C # word sequences are silently copied to the output, so as to permit the rare case of a # in the initial text of a Fortran comment.
Preprocessor names in conditionals and definitions, or set on the command line, consist of letters, digits, and underscores; the first character may not be a digit.
Symbols beginning with two underscores, or an underscore and an uppercase letter, are reserved for the local implementation; see the PREDEFINED SYMBOLS section below for details.
Following standard Fortran practice, letter case is not significant in directives, or in constants and operators in expressions. However, if nawk(1) is used to implement fpp, letter case is significant in names that are defined. With gawk(1), this limitation is removed; letter case is never significant.
For portability, it is recommended that lower-case letters be used for all directives, and upper-case letters for all defined names; this conforms to two decades of widespread practice in the C programming language.
fpp _OS_UNIX=1 _SUN386=1
or in the input file text itself:
C #define _OS_UNIX 1 C #define _SUN386 1
The cc(1)-like forms with define and undefine switches are supported; these two invocations are equivalent:
fpp +D_OS_UNIX +DWORDSIZE=32 +U_OS_VAXVMS fpp _OS_UNIX=1 WORDSIZE=32 _OS_VAXVMS=0
If the value is omitted, as in
fpp _OS_UNIX= _SUN386=
or
C #define _OS_UNIX C #define _SUN386
a value of 1 is assumed.
Names can be undefined by
C #undef name C #undefine name
If the name was not already defined, the request is silently ignored.
C#if constant-expression C#ifdef name C#ifndef name C#elseif constant-expression C#elif constant-expression C#else [optional comment] C#endif [optional comment]
Each C#ifxxx statement must have a matching C#endif following it. The two may be separated by any number of C#elseif statements, which may be followed by a single C#else statement.
A branch of a conditional is selected when the expression evaluates to a non-zero value; see the EXPRESSIONS section below for details.
Code between these statements is preserved, but in the unselected branches of a conditional statement, a non-comment statement will be altered to a comment by prefixing it with an initial C##, shifting the statement right by three columns. In the selected branch, any initial C## in columns 1 through 3 is stripped; lines without this prefix are copied verbatim. Because of Fortran's 72-character line length limitation, this means that code lines may not exceed 69 characters in length inside an fpp conditional. For example, the input
C#if _OS_UNIX C##C UNIX code C## CALL GETENV(...) C#elseif _OS_VAXVMS C VMS code CALL LIB$TRNLNM(...) C#endif
when _OS_UNIX=1 produces
C#if _OS_UNIX C UNIX code CALL GETENV(...) C#elseif _OS_VAXVMS C##C VMS code C## CALL LIB$TRNLNM(...) C#endif
When neither _OS_UNIX nor _OS_VAXVMS are defined, the output is
C#if _OS_UNIX C##C UNIX code C## CALL GETENV(...) C#elseif _OS_VAXVMS C##C VMS code C## CALL LIB$TRNLNM(...) C#endif
When only _OS_VAXVMS is defined, the original input is sent to the output. If by chance both _OS_VAXVMS and _OS_UNIX were defined, only the UNIX code would be selected, because only the first branch of the conditional would be executed.
Preprocessor conditionals may be nested:
C # if _OS_UNIX C # if _SUN C # if _SUN4 C # elseif _SUN3 C # elseif _SUN386 C # endif C # endif C # elseif _OS_VMS C # endif
Any text following #else or #endif is ignored; it can be used to document the conditional, usually with the test from the preceding #if:
C # if _OS_UNIX C # else NOT _OS_UNIX C # endif _OS_UNIX
fpp directives are not executed if they are in a branch of a conditional that is not currently selected. However, conditional statements are processed to keep track of the current nesting.
On UNIX, the -Dname option for diff(1) can be used to get output from the comparison of two files that is almost correct input for fpp. A simple command pipe diff -Dxxx file1 file2 | sed -e 's/^#/C#/' >file3 will produce an output file file3 from which fpp +Dxxx file3 will recover file2 and fpp +Uxxx file3 will recover file1.
In expressions, primaries are Fortran integer, floating-point, logical, and character constants, and preprocessor names.
Undefined names silently evaluate to zero in arithmetic expressions, and to empty strings in string expressions.
Character strings appearing in arithmetic expressions are converted to numbers, which are zero if the string does not look like a number. Character strings appearing by themselves evaluate to themselves.
Arithmetic expressions are evaluated in floating-point arithmetic; for Boolean (Fortran logical) tests, zero is false, and non-zero is true.
The usual Fortran arithmetic operators + - * / ** are recognized, along with the C modulus operator %; x % y is Fortran's mod(x,y). This operator is rigorously defined for all arguments to be x % y = x - int(x/y)*y.
The Fortran logical and relational operators are supported, with convenient modern C-like synonyms: .and. (& and &&), .or. (| and ||), .not. (!), .eq. (==), .ne. (!=), .lt. (<), .le. (<=), .gt. (>), and .ge. (>=). Letter case in the dotted operators is not significant. Finally, the Fortran character string concatenation operator, //, is handled.
One special name, defined, is recognized, in any letter case; it may be used either in functional form, defined(name), or in prefix operator form, defined name. It evaluates to 1 if the name is defined (even if the value of the name is zero), and otherwise, to 0. Several defined operators can be used in a single expression; that is much more convenient than a series of nested conditionals using C#ifdef and C#ifndef.
The #(...) form is only recognized in a comment line, and the next line is converted to a comment (see the section MACRO EXPANSION below); the parentheses hold an expression involving Fortran constants and preprocessor names.
Examples of expressions are
C#if defined(_OS_UNIX) || defined _OS_VAXVMS || (WORDSIZE == 32)" C REAL A(#(MAXA**2)), B(#(MAXA % 32)) C INTEGER BITS(#(WORDSIZE))
The first line of the pair is always exactly preserved in the output, while the second is replaced by the expansion of the comment, with the first character deleted, to change the comment into a non-comment. The original contents of the second line are preserved as a comment with the prefix C-fpp- in a third output line.
This peculiar input line pairing is necessary to ensure that the expansion is reversible.
The parentheses around the expression serve to distinguish between macros and fpp preprocessor directives in comments, and serendipitously permit the extension from simple names to arbitrary constant expressions that can be evaluated by fpp.
Care must be taken in writing the input to ensure that any expected expansion does not make the line longer than 72 characters; fpp has almost no knowledge of Fortran, and therefore cannot provide correct line wrapping for it.
Similarly, macro expansion in a multi-line continued statement should avoided, since it introduces comment lines between continuation lines. While such comments are legal in full Fortran 77, they are illegal in subset Fortran 77, and in older Fortrans, and may cause problems for other tools that process Fortran code.
Here is a small example. Given command-line definitions
FPTYPE='DOUBLE MAXA=19 MAXB=25
then the input
C #(FPTYPE) A(#(MAXA)), B(#(MAXB),#(MAXA**2)) REAL A(100), B(255,10000)
is converted to the output
C #(FPTYPE) A(#(MAXA)), B(#(MAXB),#(MAXA**2)) DOUBLE PRECISION A(19), B(25,361) C-fpp- REAL A(100), B(255,10000)
Fortran 77 PARAMETER statements can be used to achieve similar effects, but in more restricted circumstances. In particular, fpp permits such expansions to happen in strings:
C10000 FORMAT ('Host operating system = #(OS)')
10000 FORMAT ('Host operating system = UNIX')
This may be awkward to achieve in standard Fortran.
C#message text C#error text
The difference between them is that #error sets an exit code of 1, and also sends the text to stdout. This can be used to ensure that a preprocessing error forces a compilation error if an attempt is made later to compile the output source program.
The output of both directives is prefixed with the file name and input line number to identify the origin of the message.
When #error is executed, processing is not terminated; instead, fpp tries to process the remaining input so as to uncover additional errors in the same run.
C-fpp- =================================================================
C-fpp- fpp version 1.0 [10-Dec-1990]
C-fpp- Date: Sat Dec 8 23:06:30 MST 1990
C-fpp- Directory: /u/sy/beebe
C-fpp- User: [email protected]
C-fpp- Macro: _OS_VAXVMS=1
C-fpp- Macro: FPTYPE=DOUBLE PRECISION
C-fpp- =================================================================
These comments provide a record of the processing, including what symbol definitions and macro values have been selected.
Input comments beginning
C-fpp-
are flushed. Thus, any existing comment header is always replaced by a new header.
Each command-line name=value or +Dname setting, and each input definition directive
C#define name value
produce an output comment of the form
C-fpp- Macro: name=value
Thus, all output lines beginning
C-fpp- Macro:
document which names have been defined.
A C#undefine statement results in output like
C-fpp- Macro: name=--UNDEFINED--
Following ANSI C, predefined symbols always begin with two underscores, or an underscore and an uppercase letter; such names are reserved for the local implementation. Predefined symbols that do not follow this convention are forbidden. This requirement makes it possible to distinguish separate name spaces for the user and for the implementation, preventing surprises from unexpected substitutions that happen when code is moved to a new environment.
The complete set of definitions is always recorded in the output header; they can easily be displayed on stdout by giving fpp an empty input file:
fpp /dev/null
Following the conventions of the ANSI C preprocessor, predefined symbols always begin with two underscores, or an underscore and an uppercase letter.
fpp always predefines exactly one major operating-system symbol:
_OS_PCDOS _OS_TOPS20 _OS_UNIX _OS_VAXVMS
For _OS_UNIX, exactly one minor operating-system variant may also be defined:
_AIX _AIX370 _BSD _GOULD _HPUX _MACH _MIPS _POSIX _STARDENT _SUN
Additional architectural variants may be defined on some systems:
_IBM_3090 _IBM_PS_2 _IBM_RS_6000 _IBM_RT _STARDENT_1500 _STARDENT_3000 _SUN3 _SUN386 _SUN4Host byte addressing order is defined by one of these:
_BIG_ENDIAN _LITTLE_ENDIAN
Big-endian addressing is used by IBM, Motorola, and most RISC systems. Little-endian addressing is used by the Intel 80xxx and DEC VAX architectures.
Host floating-point architecture must be defined by
_IEEE_754on those machines that have IEEE 754 floating-point arithmetic.
If the Fortran implementation supports NAMELIST I/O, the symbol
_NAMELISTmust be defined.
To ensure standardization, all such names must be registered with the author of fpp, and will be listed in these manual pages.
Following ANSI C, four standard permanent symbols are always defined; these each have two leading and two trailing underscores. Permanent symbols always begin with two underscores, and they may not be redefined by the user.
C WRITE (*,*) 'Compiled on #(__DATE__) at #(__TIME__) #(__TIMEZONE__)' WRITE (*,*) 'Compiled on ??? ?? ???? at ??:??:?? ???'
might produce
C WRITE (*,*) 'Compiled on #(__DATE__) at #(__TIME__) #(__TIMEZONE__)' WRITE (*,*) 'Compiled on Dec 10 1990 at 09:10:07 MST' C-fpp- WRITE (*,*) 'Compiled on ??? ?? ???? at ??:??:?? ???'
C#error text
directives that are executed send their text argument to stderr and to stdout, and on UNIX, cause a later exit code of 1.
C#message text
directives that are executed send their text argument to stderr.
Center for Scientific Computing
Department of Mathematics
South Physics Building
University of Utah
Salt Lake City, UT 84112
Tel: (801) 581-5254
FAX: (801) 581-4148
Email: <[email protected]>