.TH bfort 1 "6/25/2019" " " ""
.SH NAME
bfort \-  program to extract short definitions for a Fortran to C interface 
.SH INPUT
.PD 0
.TP
.B filenames 
- Names the files from which lint definitions are to be extracted
.PD 1
.PD 0
.TP
.B -nomsgs   
- Do not generate messages for routines that can not be converted
to Fortran.
.PD 1
.PD 0
.TP
.B -nofort   
- Generate messages for all routines/macros without a Fortran
counterpart.
.PD 1
.PD 0
.TP
.B -dir name 
- Directory for output file
.PD 1
.PD 0
.TP
.B -I name   
- file that contains common includes
.PD 1
.PD 0
.TP
.B -mapptr   
- translate pointers to integer indices
.PD 1
.PD 0
.TP
.B -ptrprefix 
- prefix for names of functions to convert to/from pointers
(default is __).  The macro that selects the form based on the
pointer size can be changed with -ptr64.
.PD 1
.PD 0
.TP
.B -anyname   
- Generate a single Fortran wrapper that works for almost all
systems, by adding C preprocessor names (see below).  These
names can be changed with -fcaps, -fuscore, and -fduscore.
.PD 1
.PD 0
.TP
.B -ferr     
- Fortran versions return the value of the routine as the last
argument (an integer).  This is used in MPI and is a not
uncommon approach for handling error returns.
.PD 1
.PD 0
.TP
.B -shortargname 
- Use short (single character) argument names instead of the
name in the C definition.
.PD 1
.PD 0
.TP
.B -fstring  
- Enable handling of Fortran character parameters, using the most
common call interface.  This is the default unless the
environment variable BFORT_STRINGHANDLING is set to IGNORE.
.PD 1
.PD 0
.TP
.B -fnostring 
- Disable handling of Fortran character parameters.  Routines
with 
.I char
arguments in C will not have interface routines
generated for them.
.PD 1
.PD 0
.TP
.B -fstring
- asis - No special processing for string arguments.  Provided
for backward compatiblity to older versions of bfort that
did not handle strings
.PD 1
.PD 0
.TP
.B -mpi      
- Handle MPI types (some things are pointers by definition)
.PD 1
.PD 0
.TP
.B -no_pmpi  
- Do not generate PMPI names
.PD 1
.PD 0
.TP
.B -pmpi name 
- Change macro used to select MPI profiling version
.PD 1
.PD 0
.TP
.B -noprofile 
- Turn off the generation of the profiling version
.PD 1
.PD 0
.TP
.B -mnative  
- Multiple indirects are native datatypes (no coercion)
.PD 1
.PD 0
.TP
.B -voidisptr 
- Consider "void *" as a pointer to a structure.
.PD 1
.PD 0
.TP
.B -nodebug  
- Do not add
.PD 1
.nf
#ifndef DEBUG_ALL
#define DEBUG_ALL
#endif
.fi

to the wrapper file.
.PD 0
.TP
.B -anyname   
- Generate a single wrapper that can handle the three most common
cases: trailing underscore, no underscore, and all caps.  The
choice is based on whether
.PD 1
.nf
FORTRANCAPS:       Names are uppercase, no trailing underscore
FORTRANUNDERSCORE: Names are lowercase, trailing underscore
are defined.
FORTRANDOUBLEUNDERSCORE: Names are lowercase, with TWO trailing
.fi

underscores.  This is needed when some versions of "f2c" are
used to generate C for Fortran routines.  Note that f2c uses two
underscores ONLY when the name already contains an underscore
(at least on the FreeBSD system that I use that uses f2c).
To handle this case, the generated code contains the second
underscore only when the name already contains one.

If -mapptr is also chosen, then
.nf
POINTER_64_BITS
.fi

will also be used to determine if pointers are to long to
fit in a 32-bit Fortran integer.  Routines that destroy
a pointer will need to manually insert a call to
__RmPointer.  The routines for managing the pointers are
in ptrcvt.c

In addition, if -mpi is used and -no_pmpi is not, the MPI
profiling names are also generated, surrounded by
MPI_BUILD_PROFILING.

.SH NOTE
We really need a way to specify a general type as a pointer, so that it
will be handled as a pointer.  The -mpi option is a kludge for a pressing
need.  Eventually should provide a "-ptr name" option and keep in a
search space when looking for known types.

.SH AUTHOR
Bill Gropp
.SH LOCATION
bfort.c