man ftnchek (1): Fortran 77 program checker

SYNOPSIS

ftnchek [ -arguments[=list] ] [ -array[=list] ]
      [ -[no]brief ] [ -calltree[=list] ] [ -[no]check ]
      [ -columns[=num] ] [ -common[=list] ]
      [ -[no]crossref[=list] ] [ -[no]declare ]
      [ -[no]division ] [ -errors[=num] ] [ -[no]extern ]
      [ -[no]f77[=list] ] [ -[no]f90[=list] ]
      [ -[no]f95[=list] ] [ -[no]help ]
      [ -[no]identifier-chars[=list] ] [ -include=str ]
      [ -intrinsic[=list] ] [ -[no]library ] [ -[no]list ]
      [ -makedcls[=list] ] [ -mkhtml[=list] ]
      [ -[no]novice ] [ -output=str ]
      [ -pointersize[=num] ] [ -[no]portability[=list] ]
      [ -[no]pretty[=list] ] [ -project[=list] ]
      [ -[no]pure ] [ -[no]quiet ] [ -[no]reference ]
      [ -[no]resources ] [ -[no]sixchar ] [ -[no]sort ]
      [ -source[=list] ] [ -style[=list] ] [ -[no]symtab ]
      [ -[no]truncation[=list] ] [ -usage[=list] ]
      [ -[no]vcg ] [ -[no]version ] [ -[no]volatile ]
      [ -wordsize[=num] ] [ -wrap[=num] ] [ files ... ]

DESCRIPTION

ftnchek (short for Fortran checker) is designed to detect certain errors in a Fortran program that a compiler usually does not. ftnchek is not primarily intended to detect syntax errors. Its purpose is to assist the user in finding semantic errors. Semantic errors are legal in the Fortran language but are wasteful or may cause incorrect operation. For example, variables which are never used may indicate some omission in the program; uninitialized variables contain garbage which may cause incorrect results to be calculated; and variables which are not declared may not have the intended type. ftnchek is intended to assist users in the debugging of their Fortran program. It is not intended to catch all syntax errors. This is the function of the compiler. Prior to using ftnchek, the user should verify that the program compiles correctly.

This document first summarizes how to invoke ftnchek. That section should be read before beginning to use ftnchek. Later sections describe ftnchek's options in more detail, give an example of its use, and explain how to interpret the output. The final sections mention the limitations and known bugs in ftnchek.

INVOKING FTNCHEK

ftnchek is invoked through a command of the form:

$ ftnchek [-option -option ...] filename [filename ...]

The brackets indicate something which is optional. The brackets themselves are not actually typed. Here options are command-line switches or settings, which control the operation of the program and the amount of information that will be printed out. If no option is specified, the default action is to print error messages, warnings, and informational messages, but not the program listing or symbol tables.

Each option begins with the '-' character. (On VAX/VMS or MS-DOS systems you may use either '/' or '-'.) For the sake of conformity with an increasingly common convention, options can also begin with '--'. The options are described at greater length in the next section.

ftnchek options fall into two categories: switches, which are either true or false, and settings, which have a numeric or string value. The name of a switch is prefixed by 'no' or 'no-' to turn it off: e.g. -nopure would turn off the warnings about impure functions. The 'no' prefix can also be used with numeric settings, having the effect of turning off the corresponding warnings. Settings that control lists of warnings have a special syntax discussed below. Only the first 3 characters of an option name (not counting the '-') need be provided. A colon may be used in place of an equals sign for numeric or string setting assignments; however, we show only the equals sign form below.

The switches and settings which ftnchek currently recognizes are listed below. For each option, the default is the value used if the option is not explicitly specified, while the turn-on is the value used if the option is given without assigning it a value.

-arguments=list: Control warnings about subprogram type and argument mismatches. Default = turn-on = all.
-array=list: Control warnings in checking array arguments of subprograms. Default = turn-on = all.
-brief: Use shorter format for some error messages. Default = no.
-calltree=list: Produce subprogram call hierarchy in one of 3 formats: text call-tree, who-calls-who and VCG. Default = none, turn-on = tree,prune,sort.
If the -mkhtml option is invoked and tree is the applied calltree option, a file named CallTree.html, will be produced depicting the tree in HTML format.
-check: Perform checking. Default = yes.
-columns=num: Set maximum line length to num columns. (Beyond this is ignored.) Turn-on = max = 132. Default = 72.
-common=list: Set degree of strictness in checking COMMON blocks. Default = turn-on = all.
-crossref=list: Print cross-reference list of subprogram calls, label usage, and/or COMMON block use. Default = none.
-declare: Print a list of all identifiers whose datatype is not explicitly declared. Default = no.
-division: Warn wherever division is done (except division by a constant). Default = no.
-errors=num: Set the maximum number of error messages per cascade. Default = turn-on = 3.
-extern: Warn if external subprograms which are invoked are never defined. Default = yes.
-f77=list: Control specific warnings about supported extensions to the Fortran 77 Standard. Default = none, turn-on = all.
-f90=list: Control specific warnings about supported extensions to the Fortran 77 Standard that were not adopted as part of the Fortran 90 Standard. Default = none, turn-on = all.
-f95=list: Control specific warnings about standard Fortran 77 features that were deleted from the Fortran 95 Standard. Default = none, turn-on = all.
-help: Print command summary. Default = no.
-identifier-chars=list: Define non-alphanumeric characters that may be used in identifiers. Default = turn-on = dollar sign and underscore.
-include=path: Define a directory to search for INCLUDE files before searching in the system-wide directory. Cumulative. Default = turn-on = none.
-intrinsic=list: Control treatment of nonstandard intrinsic functions. Default = all except vms for Unix version, all except unix for VMS version, all except unix and vms for other versions. Turn-on = all.
-library: Begin library mode: do not warn about subprograms in file that are defined but never used. Default = no.
-list: Print source listing of program. Default = no.
-makedcls=list: Prepare a file of declarations. The list specifies options for the format of this file. Default = none, turn-on = declarations.
-mkhtml=list: Create individual HTML document files from ftnchek analysis and code comments. Usually you will also want to specify -call=tree to create the root HTML file CallTree.html. Default = none, turn-on = documents.
-novice: Give output suitable for novice users. Default = yes.
-output=filename: Send output to the given file. Default and turn-on sends output to the screen. (Default filename extension is .lis).
-pointersize=num: Set the size of ``Cray pointer'' variables to num bytes. Min = 1, max = 16. Default = turn-on = 4
-portability=list: Warn about non-portable usages. Default = none, turn-on = all.
-pretty=list: Give warnings for possibly misleading appearance of source code. Default = turn-on = all.
-project=list: Create project file (see explanation below). Default = no.
-pure: Assume functions are pure, i.e. have no side effects. Default = yes.
-quiet: Produce less verbose output. Default = no.
-reference: Print table of subprograms referenced by each subprogram. Default = no.
-resources: Print amount of resources used in analyzing the program. Default = no.
-sixchar: List any variable names which clash at 6 characters length. Default = no.
-sort: Print list of subprograms sorted in prerequisite order. Default = no.
-source=list: Select source formatting options: fixed or free form, DEC Fortran tab-formatted lines, VMS-style INCLUDE statement, UNIX-style backslash escape sequences, and implicit typing of parameters. Default = none, turn-on = all.
-style=list: Produce extra-picky warnings about obsolescent or old-fashioned programming constructions. Default = none, turn-on = all.
-symtab: Print symbol table and label table for each subprogram. Default = no.
-truncation=list: Check for possible loss of accuracy by truncation. Default = turn-on = all.
-usage=list: Control warnings about unused or uninitialized variables, common blocks, etc. Default = turn-on = all.
-vcg: Produce VCG format of call graph.
-version: Print version number. Default = no.
-volatile: Assume COMMON blocks lose definition between activations. Default = no. (Obsolete. Use -common=volatile instead.)
-wordsize=num: Set the default word size for numeric quantities to num bytes. Default = turn-on = 4 bytes.
-wrap=num: Set output column at which to wrap long error messages and warnings to the next line. If set to 0, turn off wrapping. Default = turn-on = 79.

When more than one option is used, they should be separated by a blank space, except on systems such as VMS where options begin with slash ( / ). No blank spaces may be placed around the equals sign ( = ) in a setting. ftnchek "?" will produce a command summary listing all options and settings.

For settings that take a list of keywords, namely -arguments, -array, -calltree, -common, -crossref, -f77, -f90, -f95, -intrinsic, -makedcls, -mkhtml, -portability, -pretty, -project, -source, -style, -truncation, and -usage, the list consists of keywords separated by commas or colons. If the list of keywords is omitted, the effect is to set the option to its turn-on value (same as ``all'' in most cases). Also, if the list is omitted, the setting name can be prefixed with no or no- to turn off all the options it controls. For example, -f77 turns on all warnings about nonstandard constructions, while -nof77 turns them all off. Three special keywords are:

help: Print out all the option keywords controlled by the setting, with a brief explanation of their meanings. This keyword cannot be given in a list with other keywords.
all: Set all options. This turns on all options controlled by the setting.
none: Clear all options. This turns off all options controlled by the setting.

These three special keywords must be given in full. For all other keywords, only as many letters of the keyword as are necessary to identify it unambiguously need be given, or a wildcard pattern may be used. Including a keyword in the list turns the corresponding option on. For example, -f77=intrinsic would turn on only the warnings about use of nonstandard intrinsic functions. Prefixing a keyword by no- turns its option off. For example, -pretty=no-long-line turns off warnings about lines exceeding 72 columns in length while leaving all other warnings about misleading appearance in effect. If a setting has default none, you can turn on all options except one or two by using all first. For example, -f77=all,no-include enables warnings about all nonstandard extensions except INCLUDE statements. If a setting has default all, you can turn off all warnings except one or two by using none first. For example, -truncation=none,demotion would turn off all precision related warnings except about demotions. Wildcard patterns contain an asterisk to stand for any string of characters. If a wildcard pattern is used, all the warnings that match it are affected. If no- is prefixed to the pattern, all the matching warnings are turned off, otherwise they are all turned on. The minimum unambiguous length rule does not apply to wildcard matching. For example, use -usage=no-*var* to turn off all warnings relating to variable usage (both local and common). (Unix users may need to quote any options containing wildcards in order to prevent the shell from attempting to expand them.) Wildcards are recognized only in lists of warning keywords, not in the top-level options themselves.

When ftnchek starts up, it looks for environment variables and also for a preferences file. Any options defined in the environment or in the preferences file are used as defaults in place of the built-in defaults. They are over-ridden by any command line options. See the section on changing the defaults for details about the environment options and the preferences file.

When giving a name of an input file, the extension is optional. If no extension is given, ftnchek will first look for a project file with extension .prj, and will use that if it exists. If not, then ftnchek will look for a Fortran source file with the extension .for for VMS systems, .f for UNIX systems. More than one file name can be given to ftnchek, and it will process the modules in all files as if they were in a single file.

Wildcards are allowed in the specification of filenames on the command line for the VMS and MS-DOS versions, as also of course under UNIX and any other system that performs wildcard expansion in the command processor.

If no filename is given, ftnchek will read input from the standard input.

OPTIONS

This section provides a more detailed discussion of ftnchek command-line options. Options and filenames may be interspersed on a command line. Most options are positional: each option remains in effect from the point it is encountered until it is overridden by a later change. Thus for example, the listing may be suppressed for some files and not for others. Exceptions are: the -intrinsic, -pointersize, and -wordsize settings, which cannot be changed once processing of input files has started; the -arguments, -array, -calltree, -common, -crossref, -extern, -reference, -resources, -sort, -vcg, and -volatile options, where the action depends only on the value of the option after the processing of input files is finished; and the -include setting, which is cumulative.

The option names in the following list are in alphabetical order.

-arguments=list

Controls warnings about mismatches between actual and dummy subprogram arguments, and also about mismatches between expected and actual subprogram type. (An actual argument is an argument passed to the subprogram by the caller; a dummy argument is an argument received by the subprogram.) By default, all warnings are turned on.

The list consists of keywords separated by commas or colons. Since all these warnings are on by default, include a keyword prefixed by no- to turn off a particular warning. There are three special keywords: all to turn on all the warnings about arguments, none to turn them all off, and help to print the list of all the keywords with a brief explanation of each. If list is omitted, -arguments is equivalent to -arguments=all, and -noarguments is equivalent to -arguments=none. The warning keywords with their meanings are as follows:

arrayness:: warn about inconsistent use of arguments that are arrays. These warnings can be further controlled by the -array option.
type:: warn about dummy arguments of a different data type from the actual arguments.
function-type:: warn if the invocation assumes the function's return value is a different type than it actually is. Also warns if a function is called as a subroutine, or vice-versa.
number:: warn about invoking a subprogram with a different number of arguments than the subprogram expects.

For compatibility with previous versions of ftnchek, a numeric form of this setting is also accepted: the list is replaced by a number from 0 to 3. A value of 0 turns all the warnings off, 1 turns on only number, 2 turns on all except number, and 3 turns all the warnings on.

This setting does not apply to checking invocations of intrinsic functions or statement functions, which can only be turned off by the -nocheck option.

See also: -array, -library, -usage.

-array=list

Controls the degree of strictness in checking agreement between actual and dummy subprogram arguments that are arrays. The warnings controlled by this setting are for constructions that might legitimately be used by a knowledgeable programmer, but that often indicate programming errors. By default, all warnings are turned on.

The list consists of keywords separated by commas or colons. Since all these warnings are on by default, include a keyword prefixed by no- to turn off a particular warning. There are three special keywords: all to turn on all the warnings about array arguments, none to turn them all off, and help to print the list of all the keywords with a brief explanation of each. If list is omitted, -array is equivalent to -array=all, and -noarray is equivalent to -array=none. The warning keywords with their meanings are as follows:

dimensions:: warn if the arguments differ in their number of dimensions, or if the actual argument is an array element while the dummy argument is a whole array.
size:: warn if both arguments are arrays, but they differ in number of elements.

For compatibility with previous versions of ftnchek, a numeric form of this setting is also accepted: the list is replaced by a number from 0 to 3. A value of 0 turns all the warnings off, 1 turns on only dimensions, 2 turns on only size, and 3 turns all the warnings on.

Note: A warning is always given regardless of this setting if the actual argument is an array while the dummy argument is a scalar variable, or if the actual argument is a scalar variable or expression while the dummy argument is an array. These cases are seldom intentional. (To turn off even these warnings, use -arguments=no-arrayness.) No warning is ever given if the actual argument is an array element while the dummy argument is a scalar variable. Variable-dimensioned arrays and arrays dimensioned with 1 or asterisk match any number of array elements. There is no check of whether multi-dimensional arrays agree in the size of each dimension separately.

See also: -arguments, -library, -usage.

-brief

: Selects a shorter format for some warning messages. At present, the only warnings controlled by this flag are those that are printed at the end of processing each subprogram. These include warnings about variables that are set but not used or used before set, variable names that do not conform to the Fortran 77 standard, etc. (These warnings may be suppressed entirely depending on other flags, such as the -usage or -f77 flags.) In the default format each variable is listed on a separate line, along with the line number where the variable is declared, set or used, according to the nature of the warning. The briefer format simply lists all variables to which the warning applies, with up to 4 variables per line.
See also: -quiet.

-calltree=list

Causes ftnchek to print out the call structure of the complete program.

The list consists of keywords separated by commas or colons. There are two special keywords: none to turn off all the options, and help to print the list of all the keywords with a brief explanation of each. (The keyword all turns on all the options, but should not normally be used since only one format should be specified.) If list is omitted, -calltree is equivalent to -calltree=tree, and -nocalltree is equivalent to -calltree=none. By default no call graph is printed.

If the -mkhtml option is invoked and tree is the applied calltree option, a file named CallTree.html, will also be produced depicting the tree in HTML format. This file is useful as a starting point for browsing the HTML files describing each component of the program.

The keywords which control which format is used are as follows:

tree:: produce the call graph in tree format.
reference:: produce the call graph in who-calls-who format (same as -reference switch).
vcg:: produce the call graph in VCG format (same as -vcg switch).

Only one of the formats tree, reference, or vcg may be specified.

The following keywords control options affecting the output:

prune:: prune repeated subtrees (applicable only with tree). This the default.
sort:: sort children of each routine into alphabetical order. This is the default.

See the discussion of the -reference and -vcg flags for details about these formats.

For tree format, The call graph is printed out starting from the main program, which is listed on the first line at the left margin. Then on the following lines, each routine called by the main program is listed, indented a few spaces, followed by the subtree starting at that routine.

In the default mode, if a routine is called by more than one other routine, its call subtree is printed only the first time it is encountered Later calls give only the routine name and the notice ``(see above)''. To have the subtree printed for each occurrence of the routine, use option no-prune.

Note that the call tree will be incomplete if any of the input files are project files containing more than one module that were created in -library mode. See the discussion of project files below.

Technical points: Each list of routines called by a given routine is printed in alphabetical order unless the no-sort option is given. If multiple main programs are found, the call tree of each is printed separately. If no main program is found, a report to that effect is printed out, and the call trees of any top-level non-library routines are printed. This flag only controls the printing of the call tree: ftnchek constructs the call tree in any case because it is used to determine which library modules will be cross-checked. See the discussion of the -library flag.

For compatibility with previous versions of ftnchek, a numeric form of this setting is also accepted: the list is replaced by a number from 0 to 15. This number is formed from 1 for tree format, 2 for reference format, or 3 for vcg format, plus 4 for no-prune, and 8 for no-sort.

See also: -crossref, -library, -reference, -sort, -symtab, -vcg.

-check

This switch is provided so that errors and warning messages can be turned off when ftnchek is used for purposes other than finding bugs, such as making declarations or printing the call tree. It is positional, so after turning all checks off, selected checks can be turned back on. The effect of -nocheck is to put all switches, numeric settings, and settings controlling lists of warnings to their turn-off values, as if they had all been specified with the -no prefix. Switches and settings that specify options and modes of operation, rather than controlling warnings, are unaffected. These are -columns, -crossref, -include, -intrinsic, -library, -list, -makedcls, -novice, -output, -pointersize, -project, -quiet, -reference, -resources, -sort, -source, -symtab, -vcg, -version, -wordsize, and -wrap. Default = yes.

Parse errors (syntax errors due to unrecognized or malformed statements) are not suppressed by this switch, since the results may be incorrect if ftnchek has not parsed the program correctly.

There are some miscellaneous errors and warning messages that are not controlled by any other switch, and so can be turned off only by this switch. Note that using -check following -nocheck only has the effect of turning these special warnings back on, and does not restore all the checks it turned off. These warnings are:

Module contains no executable statements.
In free source form, missing space where space is required (e.g. between a keyword and an identifier) or space present where none is allowed (e.g. within an identifier).
Zero or negative length specification in a data type declaration of the form type*len.
Invalid operand(s) in an expression.
Array assigned to scalar.
Type mismatch between DO index and bounds.
Undefined common block declared in SAVE statement.
Intrinsic function explicitly declared with an incompatible type.
Unknown intrinsic function explicitly declared in an INTRINSIC statement.
Intrinsic function passed as a subprogram argument is not declared in an INTRINSIC statement.
Intrinsic function or statement function invoked incorrectly.
Function does not set return value prior to RETURN statement.
Parameter constant value not evaluated (this is ftnchek's fault, and it is just informing you of the fact).
Entry point of a subprogram is later used as a different subprogram's name.
Unknown keyword used in an I/O statement.
Illegal label reference (e.g. GOTO refers to a non-executable statement; I/O statement refers to a non-format statement).

CHANGING THE DEFAULTS

ftnchek includes two mechanisms for changing the default values of all options: by defining environment variables or by creating a preferences file. When ftnchek starts up, it looks in its environment for any variables whose names are composed by prefixing the string FTNCHEK_ onto the uppercased version of the option name. If such a variable is found, its value is used to specify the default for the corresponding switch or setting. In the case of settings (for example, the -common strictness setting) the value of the environment variable is read as the default setting value. In the case of switches, the default switch will be taken as true or yes unless the environment variable has the value 0 or NO.

Note that the environment variable name must be constructed with the full-length option name, which must be in uppercase. For example, to make ftnchek print a source listing by default, set the environment variable FTNCHEK_LIST to 1 or YES or anything other than 0 or NO. The names FTNCHEK_LIS (not the full option name) or ftnchek_list (lower case) would not be recognized.

Here are some examples of how to set environment variables on various systems. For simplicity, all the examples set the default -list switch to YES.

>1. UNIX, Bourne shell: >$ FTNCHEK_LIST=YES
> >$ export FTNCHEK_LIST

>2. UNIX, C shell: >% setenv FTNCHEK_LIST YES

>3. VAX/VMS: >$ DEFINE FTNCHEK_LIST YES

>4. MSDOS: >$ SET FTNCHEK_LIST=YES

After processing any environment variables, ftnchek looks for a preferences file containing options and settings. It will search in the following order, using only the first file found: (1) .ftnchekrc in the current directory, (2) ftnchek.ini in the current directory, (3) .ftnchekrc in the user's home directory, (4) ftnchek.ini in the home directory. If such a file is found, the options defined in it are used as defaults in place of the built-in defaults and overriding any defaults set in the environment..

Each option or setting in the preferences file must be on a separate line. They are given in the same form as on the command line, except without the initial dash. The preferences file can contain blank lines and comments. Comments are introduced at any point in a line by a space character (blank or tab) or the '#' character, and are terminated by the end of the line.

Command-line options override the defaults set in the environment or in the preferences file, in the same way as they override the built-in defaults.

USING PROJECT FILES

This section contains detailed information on how to use project files most effectively, and how to avoid some pitfalls.

One can divide the checks ftnchek does into two categories, local and global. Local checking is restricted to within a single routine, and catches things like uninitialized variables, unintended loss of precision in arithmetic expressions, etc. This sort of checking can be done on each subprogram independently. Furthermore, local checking of a subprogram does not need to be repeated when some other subprogram is changed. Global checking catches things like calling a subroutine with the wrong argument types, or disagreeing in common block declarations. It requires looking at the whole set of subprograms interacting with each other.

The purpose of project files is to allow the local checking and global checking steps to be separated. Assuming that each subprogram is in its own source file, you can run ftnchek once on each one to do local checking while suppressing global checking. Then ftnchek can be run once on all the project files together to do the global checking. The sample makefile below shows how to automate this task. The ``.f.prj'' target updates a project file for a particular file any time the source file changes. The information needed for global checking is saved in the project file. The ``check'' target does the combined global checking. Typically ``make check'' would repeat the ``ftnchek -project'' step only on changed source files, then do the global check. This is obviously a big advantage for large programs, when many subprograms seldom if ever change.

It is best when using project files to place each subprogram in a separate source file. If each source file may contain more than one subprogram, it complicates the definition of ``local'' and ``global'' checking because there is some inter-module checking that is contained within a file. ftnchek tries to do the right thing in this case, but there are some complications (described below) due to the trade-off between avoiding re-doing cross-checks and preserving information about the program's structure.

Ordinarily, to do the least amount of re-checking, project files should be created with the -library flag in effect and trimming turned on. In this mode, the information saved in the project file consists of all subprogram declarations, all subprogram invocations not resolved by declarations in the same file, and one instance of each COMMON block declaration. This is the minimum amount of information needed to check agreement between files.

If the source file contains more than one routine, there are some possible problems that can arise from creating the project file in library mode, because the calling hierarchy among routines defined within the file is lost. Also, if the routines in the file make use of COMMON blocks that are shared with routines in other files, there will not be enough information saved for the correct checking of set and used status of COMMON blocks and COMMON variables according to the -usage setting. Therefore if you plan to use project files when -usage checking is turned on (which is the default situation), and if multiple routines in one project file share COMMON blocks with routines in other files, the project files should be created with the -library flag turned off. In this mode, ftnchek saves, besides the information listed above, one invocation of each subprogram by any other subprogram in the same file, and all COMMON block declarations. This means that the project file will be larger than necessary, and that when it is read in, ftnchek may repeat some inter-module checks that it already did when the project file was created. If each project file contains only one module, there is no loss of information in creating the project files in library mode.

Because of the possible loss of information entailed by creating a project file with the -library flag in effect, whenever that project file is read in later, it will be treated as a library file regardless of the current setting of the -library flag. On the other hand, a project file created with library mode turned off can be read in later in either mode.

The foregoing discussion assumes that the trimming options of the -project setting are turned on when the project file is created. This is the normal situation. The no-trim options of the -project setting are provided in case one wants to use the project files for purposes other than checking the program with ftnchek. For instance, one could write a Perl script to analyze the project files for information about how the different subprograms are called. You should not use the no-trim options to deal with the issues of information loss discussed above, since they cause more information than necessary to be stored. This makes the project files bigger and causes ftnchek to do more work later when it reads them to check your complete program. Ordinarily, you should use the -library option to control how much information to store for later use by ftnchek in checking your program.

Here is an example of how to use the UNIX make utility to automatically create a new project file each time the corresponding source file is altered, and to check the set of files for consistency. Add these lines to your makefile. The example assumes that a macro OBJS has been defined which lists all the names of object files to be linked together to form the complete executable program. (In this makefile, the indented lines should each begin with a tab, not blanks.) If any source file contains multiple routines that share common blocks among themselves, then the no-com-\* option should be removed from NOGLOBAL, and/or drop the -library flag.

    # tell make what a project file suffix is
    .SUFFIXES: .prj
    # these options suppress global checks.
    NOGLOBAL=-usage=no-ext-undefined,no-com-\*
    # tell make how to create a .prj file from a .f file
    .f.prj:
            ftnchek -project $(NOGLOBAL) -library $<
    # set up macro PRJS containing project filenames
    PRJS= $(OBJS:.o=.prj)
    # "make check" will check everything that has been changed.
    check: $(PRJS)
            ftnchek $(PRJS)

When a program uses many routines defined in a large number of different source files in different directories, it can be cumbersome to specify all the different project files needed to check the program properly. To deal with such cases, ftnchek allows project files to be concatenated into a single large file. This single file can then be given to ftnchek to provide the information for checking the use of any or all of the routines defined in the combined project files. When using such a ``library'' project file, you may want ftnchek's error reports to document precisely the name of the file where the specific function is defined. If the various source files are in several directories, an error report that gives only the file name may be ambiguous, and rather should include the path to the file. The solution is to create each of the individual project files by giving the complete path to the source file. Then this complete path will appear in the error reports. For example, suppose that all of the library subprogram source files are in subdirectories of a directory named /util/lib. Then the individual project files could first be created by a command such as

find /util/lib -name '*.f' -exec ftnchek -project '{}' ';'

(Possibly other options would be provided to ftnchek as discussed above. Also, this step could be handled instead by a revised makefile rule that would provide the complete source file path instead of just the local name when invoking ftnchek.) Next, concatenate all of these project files manually.

find /util/lib -name '*.prj' -exec cat '{}' ';' > ourlib.prj

Then a program source file can be checked by using the command

ftnchek prog.f ... -lib ourlib.prj

and an error message related to any library routine will include the full path to the routine's source file.

At present, there is no archive utility like ar to manage the contents of a concatenated project file like the one in the illustration above. If changes are made to one of the library routines, the only way to update the combined project file is to concatenate all the individual project files once again. Such a utility would be quite easy to write. Someone should do so and contribute it to the ftnchek effort.

AN EXAMPLE

The following simple Fortran program illustrates the messages given by ftnchek. The program is intended to accept an array of test scores and then compute the average for the series.

C       AUTHORS: MIKE MYERS AND LUCIA SPAGNUOLO
C       DATE:    MAY 8, 1989
C       Variables:
C               SCORE -> an array of test scores
C               SUM ->   sum of the test scores
C               COUNT -> counter of scores read in
C               I ->     loop counter
        REAL FUNCTION COMPAV(SCORE,COUNT)
            INTEGER SUM,COUNT,J,SCORE(5)
            DO 30 I = 1,COUNT
                SUM = SUM + SCORE(I)
30          CONTINUE
            COMPAV = SUM/COUNT
        END
        PROGRAM AVENUM
C
C                       MAIN PROGRAM
C
C       AUTHOR:   LOIS BIGBIE
C       DATE:     MAY 15, 1990
C
C       Variables:
C               MAXNOS -> maximum number of input values
C               NUMS    -> an array of numbers
C               COUNT   -> exact number of input values
C               AVG     -> average returned by COMPAV
C               I       -> loop counter
C
            PARAMETER(MAXNOS=5)
            INTEGER I, COUNT
            REAL NUMS(MAXNOS), AVG
            COUNT = 0
            DO 80 I = 1,MAXNOS
                READ (5,*,END=100) NUMS(I)
                COUNT = COUNT + 1
80          CONTINUE
100         AVG = COMPAV(NUMS, COUNT)
        END

The compiler gives no error messages when this program is compiled. Yet here is what happens when it is run:

$ run average
70
90
85
<EOF>
$

What happened? Why didn't the program do anything? The following is the output from ftnchek when it is used to debug the above program:

$ ftnchek -list -symtab average
FTNCHEK Version 3.3 November 2004
File average.f:
      1 C       AUTHORS: MIKE MYERS AND LUCIA SPAGNUOLO
      2 C       DATE:    MAY 8, 1989
      3 
      4 C       Variables:
      5 C               SCORE -> an array of test scores
      6 C               SUM ->   sum of the test scores
      7 C               COUNT -> counter of scores read in
      8 C               I ->     loop counter
      9 
     10         REAL FUNCTION COMPAV(SCORE,COUNT)
     11             INTEGER SUM,COUNT,J,SCORE(5)
     12 
     13             DO 30 I = 1,COUNT
     14                 SUM = SUM + SCORE(I)
     15 30          CONTINUE
     16             COMPAV = SUM/COUNT
                           ^
Warning near line 16 col 20: integer quotient expr SUM/COUNT  converted to
 real
     17         END
Module COMPAV: func: real
Variables:
 
      Name Type Dims     Name Type Dims     Name Type Dims     Name Type Dims
    COMPAV real         COUNT intg             I intg*            J intg   
     SCORE intg  1        SUM intg   
* Variable not declared. Type has been implicitly defined.
Warning in module COMPAV: Variables declared but never referenced:
    J declared at line 11
Warning in module COMPAV: Variables may be used before set:
    SUM used at line 14
    SUM set at line 14
Statement labels defined:
    Label   Line  StmtType
     <30>     15      exec
     18 
     19 
     20         PROGRAM AVENUM
     21 C
     22 C                       MAIN PROGRAM
     23 C
     24 C       AUTHOR:   LOIS BIGBIE
     25 C       DATE:     MAY 15, 1990
     26 C
     27 C       Variables:
     28 C               MAXNOS -> maximum number of input values
     29 C               NUMS    -> an array of numbers
     30 C               COUNT   -> exact number of input values
     31 C               AVG     -> average returned by COMPAV
     32 C               I       -> loop counter
     33 C
     34 
     35             PARAMETER(MAXNOS=5)
     36             INTEGER I, COUNT
     37             REAL NUMS(MAXNOS), AVG
     38             COUNT = 0
     39             DO 80 I = 1,MAXNOS
     40                 READ (5,*,END=100) NUMS(I)
     41                 COUNT = COUNT + 1
     42 80          CONTINUE
     43 100         AVG = COMPAV(NUMS, COUNT)
     44         END
Module AVENUM: prog
External subprograms referenced:
    COMPAV: real*  
Variables:
 
      Name Type Dims     Name Type Dims     Name Type Dims     Name Type Dims
       AVG real         COUNT intg             I intg        MAXNOS intg*  
      NUMS real  1
* Variable not declared. Type has been implicitly defined.
Warning in module AVENUM: Variables set but never used:
    AVG set at line 43
I/O Operations:
     Unit ID Unit No. Access Form Operation   Line
             5          SEQ  FMTD READ         40 
Statement labels defined:
    Label   Line  StmtType    Label   Line  StmtType
     <80>     42      exec    <100>     43      exec
 0 syntax errors detected in file average.f
 6 warnings issued in file average.f
Warning: Subprogram COMPAV argument data type mismatch at position 1:
    Dummy arg SCORE in module COMPAV line 10 file average.f is type intg
    Actual arg NUMS in module AVENUM line 43 file average.f is type real

According to ftnchek, the program contains variables which may be used before they are assigned an initial value, and variables which are not needed. ftnchek also warns the user that an integer quotient has been converted to a real. This may assist the user in catching an unintended roundoff error. Since the -symtab flag was given, ftnchek prints out a table containing identifiers from the local module and their corresponding datatype and number of dimensions. Finally, ftnchek warns that the function COMPAV is not used with the proper type of arguments.

With ftnchek's help, we can debug the program. We can see that there were the following errors:

1.: SUM and COUNT should have been converted to real before doing the division.
2.: SUM should have been initialized to 0 before entering the loop.
3.: AVG was never printed out after being calculated.
4.: NUMS should have been declared INTEGER instead of REAL.

We also see that I, not J, should have been declared INTEGER in function COMPAV. Also, MAXNOS was not declared as INTEGER, nor COMPAV as REAL, in program AVENUM. These are not errors, but they may indicate carelessness. As it happened, the default type of these variables coincided with the intended type.

Here is the corrected program, and its output when run:

C       AUTHORS: MIKE MYERS AND LUCIA SPAGNUOLO
C       DATE:    MAY 8, 1989
C
C       Variables:
C               SCORE -> an array of test scores
C               SUM ->   sum of the test scores
C               COUNT -> counter of scores read in
C               I ->     loop counter
C
       REAL FUNCTION COMPAV(SCORE,COUNT)
            INTEGER SUM,COUNT,I,SCORE(5)
C
            SUM = 0
            DO 30 I = 1,COUNT
                SUM = SUM + SCORE(I)
30          CONTINUE
            COMPAV = FLOAT(SUM)/FLOAT(COUNT)
        END
C
C
        PROGRAM AVENUM
C
C                       MAIN PROGRAM
C
C       AUTHOR:   LOIS BIGBIE
C       DATE:     MAY 15, 1990
C
C       Variables:
C               MAXNOS -> maximum number of input values
C               NUMS    -> an array of numbers
C               COUNT   -> exact number of input values
C               AVG     -> average returned by COMPAV
C               I       -> loop counter
C
C
            INTEGER MAXNOS
            PARAMETER(MAXNOS=5)
            INTEGER I, NUMS(MAXNOS), COUNT
            REAL AVG,COMPAV
            COUNT = 0
            DO 80 I = 1,MAXNOS
                READ (5,*,END=100) NUMS(I)
                COUNT = COUNT + 1
80          CONTINUE
100         AVG = COMPAV(NUMS, COUNT)
            WRITE(6,*) 'AVERAGE =',AVG
        END
$ run average
70
90
85
<EOF>
AVERAGE =   81.66666
$

With ftnchek's help, our program is a success!

INTERPRETING THE OUTPUT

The messages given by ftnchek include not only syntax errors but also warnings and informational messages about things that are legal Fortran but that may indicate errors or carelessness. Most of these messages can be turned off by command-line options. Which option controls each message depends on the nature of the condition being warned about. See the descriptions of the command-line flags in the previous sections, and of individual messages below. Each message is prefixed with a word or phrase indicating the nature of the condition and its severity.

``Error'' means a syntax error. The simplest kind of syntax errors are typographical errors, for example unbalanced parentheses or misspelling of a keyword. This type of error is caught by the parser and appears with the description ``parse error'' or ``syntax error'' (depending on the version of the parser generator and whether it is GNU bison or UNIX yacc). This type of error message cannot be suppressed. Be aware that this type of error often means that ftnchek has not properly interpreted the statement where the error occurs, so that its subsequent checking operations will be compromised. You should eliminate all syntax errors before proceeding to interpret the other messages ftnchek gives.

``Warning: Nonstandard syntax'' indicates an extension to Fortran that ftnchek supports but that is not according to the Fortran 77 Standard. The extensions that ftnchek accepts are described in the section on Extensions below. One example is the DO ... ENDDO construction. If a program uses these extensions, warnings will be given according to specifications under the -f77 setting. The default behavior is to give no warnings.

``Warning'' in other cases means a condition that is suspicious but that may or may not be a programming error. Frequently these conditions are legal under the standard. Some are illegal but do not fall under the heading of syntax errors. Usage errors are one example. These refer to the possibility that a variable may be used before it has been assigned a value (generally an error), or that a variable is declared but never used (harmless but may indicate carelessness). The amount of checking for usage errors is controlled by the -usage flag, which specifies the maximum amount of checking by default.

Truncation warnings cover situations in which accuracy may be lost unintentionally, for example when a double precision value is assigned to a real variable. These warnings are controlled by the -truncation setting, which is on by default.

``Nonportable usage'' warns about some feature that may not be accepted by some compilers even though it is not contrary to the Fortran 77 Standard, or that may cause the program to perform differently on different platforms. For example, equivalencing real and integer variables is usually a non-portable practice. The use of extensions to the standard language is, of course, another source of non-portability, but this is handled as a separate case. To check a program for true portability, both the -portability and the -f77 flags should be used. They are both turned off by default. The -wordsize setting is provided to check only those nonportable usages that depend on a particular machine wordsize.

``Possibly misleading appearance'' is used for legal constructions that may not mean what they appear to mean at first glance. For example, Fortran is insensitive to blank space, so extraneous space within variable names or the lack of space between a keyword and a variable can convey the wrong impression to the reader. These messages can be suppressed by turning off the -pretty flag, which is on by default.

Other messages that are given after all the files are processed, and having to do with agreement between modules, do not use the word ``warning'' but generally fall into that category. Examples include type mismatches between corresponding variables in different COMMON block declarations, or between dummy and actual arguments of a subprogram. These warnings are controlled by the -common and -arguments settings respectively. By default both are set for maximum strictness of checking.

Another group of warnings about conditions that are often harmless refer to cases where the array properties of a variable passed as a subprogram argument differ between the two routines. For instance, an array element might be passed to a subroutine that expects a whole array. This is a commonly-used technique for processing single rows or columns of two-dimensional arrays. However, it could also indicate a programming error. The -array setting allows the user to adjust the degree of strictness to be used in checking this kind of agreement between actual and dummy array arguments. By default the strictness is maximum.

``Oops'' indicates a technical problem, meaning either a bug in ftnchek or that its resources have been exceeded.

The syntax error messages and warnings include the filename along with the line number and column number. ftnchek has two different options for the appearance of these error messages. If -novice is in effect, which is the default, the messages are in a style approximating normal English. (In default style, the filename is not printed in messages within the body of the program if -list is in effect.) The other style of error messages is selected by the -nonovice option. In this style, the appearance of the messages is similar to that of the UNIX lint program.

ftnchek is still blind to some kinds of syntax errors. The two most important ones are detailed checking of FORMAT statements, and almost anything to do with control of execution flow by means of IF, DO, and GOTO statements: namely correct nesting of control structures, matching of opening statements such as IF ... THEN with closing statements such as ENDIF, and the proper use of statement labels (numbers). Most compilers will catch these errors. See the section on Limitations for a more detailed discussion.

If ftnchek gives you a syntax error message when the compiler does not, it may be because your program contains an extension to standard Fortran which is accepted by the compiler but not by ftnchek. (See the section on Extensions.) On a VAX/VMS system, you can use the compiler option /STANDARD to cause the compiler to accept only standard Fortran. On most UNIX or UNIX-like systems, this can be accomplished by setting the flag -ansi.

Many of the messages given by ftnchek are self-explanatory. Those that need some additional explanation are listed below in alphabetical order.

Common block NAME: data type mismatch at position n

The n-th variable in the COMMON block differs in data type in two different declarations of the COMMON block. By default (-common strictness level 3), ftnchek is very picky about COMMON blocks: the variables listed in them must match exactly by data type and array dimensions. That is, the legal pair of declarations in different modules:

     COMMON /COM1/ A,B
and
     COMMON /COM1/ A(2)

will cause ftnchek to give warnings at strictness level 3. These two declarations are legal in Fortran since they both declare two real variables. At strictness level 1 or 2, no warning would be given in this example, but the warning would be given if there were a data type mismatch, for instance, if B were declared INTEGER. Controlled by -common setting.

Common block NAME has long data type following short data type

Some compilers require alignment of multi-byte items so that each item begins at an address that is a multiple of the item size. Thus if a short (e.g. single-precision real) item is followed by a long (e.g. double precision real) item, the latter may not be aligned correctly. Controlled by -portability=common-alignment option.

Common block NAME has mixed character and non-character variables

The ANSI standard requires that if any variable in a COMMON block is of type CHARACTER, then all other variables in the same COMMON block must also be of type CHARACTER. Controlled by -f77=mixed-common option.

Common block NAME: varying length

For -common setting level 2, this message means that a COMMON block is declared to have different numbers of words in two different subprograms. A word is the amount of storage occupied by one integer or real variable. For -common setting level 3, it means that the two declarations have different numbers of variables, where an array of any size is considered one variable. This is not necessarily an error, but it may indicate that a variable is missing from one of the lists. Note that according to the Fortran 77 Standard, it is an error for named COMMON blocks (but not blank COMMON) to differ in number of words in declarations in different modules. Given for -common setting 2 or 3.

Error: Badly formed logical/relational operator or constant

Error: Badly formed real constant

The syntax analyzer has found the start of one of the special words that begin and end with a period (e.g. .EQ.), or the start of a numeric constant, but did not succeed in finding a complete item of that kind.

Error: cannot be adjustable size in module NAME

A character variable cannot be declared with a size that is an asterisk in parentheses unless it is a dummy argument, a parameter, or the name of the function defined in the module.

Error: cannot be declared in SAVE statement in module NAME

Only local variables and common blocks can be declared in a SAVE statement.

Error: No path to this statement

ftnchek will detect statements which are ignored or by-passed because there is no foreseeable route to the statement. For example, an unnumbered statement (a statement without a statement label), occurring immediately after a GOTO statement, cannot possibly be executed.

Error: Parse error

This means that the parser, which analyzes the Fortran program into expressions, statements, etc., has been unable to find a valid interpretation for some portion of a statement in the program. If your compiler does not report a syntax error at the same place, the most common explanations are: (1) use of an extension to ANSI standard Fortran that is not recognized by ftnchek, or (2) the statement requires more lookahead than ftnchek uses (see section on Bugs).

NOTE: This message means that the affected statement is not interpreted. Therefore, it is possible that ftnchek's subsequent processing will be in error, if it depends on any matters affected by this statement (type declarations, etc.).

Error: Syntax error

This is the same as ``Error: Parse error'' (see above). It is generated if your version of ftnchek was built using the UNIX yacc parser generator rather than GNU bison.

Identifiers which are not unique in first six chars

Warns that two identifiers which are longer than 6 characters do not differ in the first 6 characters. This is for portability: they may not be considered distinct by some compilers. Controlled by -sixchar option.

Nonportable usage: argument precision may not be correct for intrinsic function

The precision of an argument passed to an intrinsic function may be incorrect on some computers. Issued when a numeric variable declared with explicit precision (e.g. REAL*8 X) is passed to a specific intrinsic function (e.g. DSQRT(X)). Controlled by -portability=mixed-size and -wordsize.

Nonportable usage: character constant/variable length exceeds 255

Some compilers do not support character strings more than 255 characters in length. Controlled by -portability=long-string.

Nonportable usage: File contains tabs

ftnchek expands tabs to be equivalent to spaces up to the next column which is a multiple of 8. Some compilers treat tabs differently, and also it is possible that files sent by electronic mail will have the tabs converted to blanks in some way. Therefore files containing tabs may not be compiled correctly after being transferred. ftnchek does not give this message if tabs only occur within comments or character constants. Controlled by -portability=tab.

Nonportable usage: non-integer DO loop bounds

This warning is only given when the DO index and bounds are non-integer. Use of non-integer quantities in a DO statement may cause unexpected errors, or different results on different machines, due to roundoff effects. Controlled by -portability=real-do.

Possibly it is an array which was not declared

This message is appended to warnings related to a function invocation or to an argument type mismatch, for which the possibility exists that what appears to be a function is actually meant to be an array. If the programmer forgot to dimension an array, references to the array will be interpreted as function invocations. This message will be suppressed if the name in question appears in an EXTERNAL or INTRINSIC statement. Controlled by the -novice option.

Possibly misleading appearance: characters past 72 columns

The program is being processed with the statement field width at its standard value of 72, and some nonblank characters have been found past column 72. In this case, ftnchek is not processing the characters past column 72, and is notifying the user that the statement may not have the meaning that it appears to have. These characters might be intended by the programmer to be significant, but they will be ignored by the compiler. Controlled by -pretty=long-line.

Possibly misleading appearance: Common block declared in more than one statement

Such multiple declarations are legal and have the same effect as a continuation of the original declaration of the block. This warning is only given if the two declarations are separated by one or more intervening statements. Controlled by -pretty=multiple-common.

Possibly misleading appearance: Continuation follows comment or blank line

ftnchek issues this warning message to alert the user that a continuation of a statement is interspersed with comments, making it easy to overlook. Controlled by -pretty=continuation.

Possibly misleading appearance: Extraneous parentheses

Warns about parentheses surrounding a variable by itself in an expression. When a parenthesized variable is passed as an argument to a subprogram, it is treated as an expression, not as a variable whose value can be modified by the called routine. Controlled by -pretty=parentheses.

Subprogram NAME: argument data type mismatch at position n

The subprogram's n-th actual argument (in the CALL or the usage of a function) differs in datatype or precision from the n-th dummy argument (in the SUBROUTINE or FUNCTION declaration). For instance, if the user defines a subprogram by

     SUBROUTINE SUBA(X)
     REAL X

and elsewhere invokes SUBA by

     CALL SUBA(2)

ftnchek will detect the error. The reason here is that the number 2 is integer, not real. The user should have written

     CALL SUBA(2.0)

When checking an argument which is a subprogram, ftnchek must be able to determine whether it is a function or a subroutine. The rules used by ftnchek to do this are as follows: If the subprogram, besides being passed as an actual argument, is also invoked directly elsewhere in the same module, then its type is determined by that usage. If not, then if the name of the subprogram does not appear in an explicit type declaration, it is assumed to be a subroutine; if it is explicitly typed it is taken as a function. Therefore, subroutines passed as actual arguments need only be declared by an EXTERNAL statement in the calling module, whereas functions must also be explicitly typed in order to avoid generating this error message. Controlled by -arguments setting.

Subprogram NAME: argument arrayness mismatch at position n

Similar to the preceding situation, but the subprogram dummy argument differs from the corresponding actual argument in its number of dimensions or number of elements. Controlled by -array together with -arguments settings.

Subprogram NAME: argument mismatch at position n

A character dummy argument is larger than the corresponding actual argument, or a Hollerith dummy argument is larger than the corresponding actual argument. Controlled by -arguments setting.

Subprogram NAME: argument usage mismatch

ftnchek detects a possible conflict between the way a subprogram uses an argument and the way in which the argument is supplied to the subprogram. The conflict can be one of two types, as outlined below.

Dummy arg is modified, Actual arg is const or expr A dummy argument is an argument as named in a SUBROUTINE or FUNCTION statement and used within the subprogram. An actual argument is an argument as passed to a subroutine or function by the caller. ftnchek is saying that a dummy argument is modified by the subprogram, implying that its value is changed in the calling module. The corresponding actual argument should not be a constant or expression, but rather a variable or array element which can be legitimately assigned to. Controlled by the -usage=arg-const-modified option.

Dummy arg used before set, Actual arg not set Here a dummy argument may be used in the subprogram before having a value assigned to it by the subprogram. The corresponding actual argument should have a value assigned to it by the caller prior to invoking the subprogram. Controlled by the -usage=var-uninitialized option.

This warning is not affected by the -arguments setting.

Subprogram NAME invoked inconsistently

Here the mismatch is between the datatype of the subprogram itself as used and as defined. For instance, if the user declares

     INTEGER FUNCTION COUNT(A)

and invokes COUNT in another module as

     N = COUNT(A)

without declaring its datatype, it will default to real type, based on the first letter of its name. The calling module should have included the declaration

     INTEGER COUNT

Given for -arguments setting 2 or 3.

Subprogram NAME: varying length argument lists:

An inconsistency has been found between the number of dummy arguments (parameters) a subprogram has and the number of actual arguments given it in an invocation. ftnchek keeps track of all invocations of subprograms (CALL statements and expressions using functions) and compares them with the definitions of the subprograms elsewhere in the source code. The Fortran compiler normally does not catch this type of error. Given for -arguments setting 1 or 3.

Variable not declared. Type has been implicitly defined

When printing the symbol table for a module, ftnchek will flag with an asterisk all identifiers that are not explicitly typed and will show the datatype that was assigned through implicit typing. This provides support for users who wish to declare all variables as is required in Pascal or some other languages. This message appears only when the -symtab option is in effect. Alternatively, use the -declare flag if you want to get a list of all undeclared variables.

Variables declared but never referenced

Detects any identifiers that were declared in your program but were never used, either to be assigned a value or to have their value accessed. Variables in COMMON are excluded. Controlled by the -usage=var-unused option.

Variables set but never used

ftnchek will notify the user when a variable has been assigned a value, but the variable is not otherwise used in the program. Usually this results from an oversight. Controlled by the -usage=var-set-unused option.

Variables used before set

This message indicates that an identifier is used to compute a value prior to its initialization. Such usage may lead to an incorrect value being computed, since its initial value is not controlled. Controlled by the -usage=var-uninitialized option.

Variables may be used before set

Similar to used before set except that ftnchek is not able to determine its status with certainty. ftnchek assumes a variable may be used before set if the first usage of the variable occurs prior in the program text to its assignment. Controlled by the -usage=var-uninitialized option.

Warning: DO index is not integer

This warning is only given when the DO bounds are integer, but the DO index is not. It may indicate a failure to declare the index to be an integer. Controlled by -truncation=real-do option.

Warning: integer quotient expr ... converted to real

The quotient of two integers results in an integer type result, in which the fractional part is dropped. If such an integer expression involving division is later converted to a real datatype, it may be that a real type division had been intended. Controlled by -truncation=int-div-real option.

Warning: Integer quotient expr ... used in exponent

The quotient of two integers results in an integer type result, in which the fractional part is dropped. If such an integer expression is used as an exponent, it is quite likely that a real type division was intended. Controlled by -truncation=int-div-exponent option.

Warning: NAME not set when RETURN encountered

The way that functions in Fortran return a value is by assigning the value to the name of the function. This message indicates that the function was not assigned a value before the point where a RETURN statement was found. Therefore it is possible that the function could return an undefined value.

Warning: Nonstandard syntax: adjustable size cannot be concatenated here

The Fortran 77 Standard (sec. 6.2.2) forbids concatenating character variables whose size is an asterisk in parentheses, except in an assignment statement. Controlled by -f77=mixed-expr.

Warning: Nonstandard syntax : significant characters past 72 columns

This warning is given under the -f77=long-line setting if the -columns setting has been used to increase the statement field width, and a statement has meaningful program text beyond column 72. Standard Fortran ignores all text in those columns, but some compilers do not. Thus the program may be treated differently by different compilers.

Warning: Nonstandard syntax : Statement out of order.

ftnchek will detect statements that are out of the sequence specified for ANSI standard Fortran 77. Table 1 illustrates the allowed sequence of statements in the Fortran language. Statements which are out of order are nonetheless interpreted by ftnchek, to prevent ``cascades'' of error messages. The sequence counter is also rolled back to prevent repetition of the error message for a block of similar statements. Controlled by the -f77=statement-order option.

--------------------------------------------------------
                  |               | implicit
                  |  parameter    |---------------------
                  |               | other specification
        format    |---------------|---------------------
         and      |               | statement-function
        entry     |  data         |---------------------
                  |               | executable
--------------------------------------------------------
                      Table 1

Warning: Possible division by zero

This message is printed out wherever division is done (except division by a constant). Use it to help locate a runtime division by zero problem. Controlled by -division option.

Warning: real truncated to intg

ftnchek has detected an assignment statement which has a real expression on the right, but an integer variable on the left. The fractional part of the real value will be lost. If you explicitly convert the real expression to integer using the INT or NINT intrinsic function, no warning will be printed. A similar message is printed if a double precision expression is assigned to a single precision variable, etc. Controlled by -truncation=demotion option.

Warning: subscript is not integer

Since array subscripts are normally integer quantities, the use of a non-integer expression here may signal an error. Controlled by -truncation=real-subscript option.

Warning: Unknown intrinsic function

This message warns the user that a name declared in an INTRINSIC statement is unknown to ftnchek. Probably it is a nonstandard intrinsic function, and so the program will not be portable. The function will be treated by ftnchek as a user-defined function. This warning is not suppressed by any option, since it affects ftnchek's analysis of the program. However, if the intrinsic function is in one of the supported sets of nonstandard intrinsics, you can use the -intrinsic setting to cause ftnchek to recognize it.

LIMITATIONS AND EXTENSIONS

ftnchek accepts ANSI standard Fortran-77 programs with some minor limitations and numerous common extensions.

Limitations:

The dummy arguments in statement functions are treated like ordinary variables of the program. That is, their scope is the entire subprogram, not just the statement function definition.

The checking of FORMAT statements is lax, tolerating missing separators (comma, etc.) between format descriptors in places where the Standard requires them, and allowing .d fields on descriptors that should not have them. It does warn under -f77=format-edit-descr about nonstandard descriptor types (like O), and supported extensions.

There are some syntactic extensions and Fortran 90 elements that ftnchek accepts but does very little checking. For instance, pointer usage (whether the nonstandard Cray syntax or the Fortran 90 syntax) is not checked other than for set and used status. It is hoped that some day more thorough checking will be implemented, but for now the user should regard the acceptance of these syntactic features simply as a convenience to enable checking of other aspects of code that contains them. See the section Extensions for specifics about what features are accepted but not fully checked.

If a user-supplied subprogram has the same name as one of the nonstandard intrinsic functions recognized by ftnchek, it must be declared in an EXTERNAL statement in any routine that invokes it. Otherwise it will be subject to the checking normally given to the intrinsic function. Since the nonstandard intrinsics are not standard, this EXTERNAL statement is not required by the Fortran 77 Standard. Using the -intrinsic=none setting, recognition of most nonstandard intrinsics (excepting only those needed to support the double complex data type) can be turned off. See the lists of supported nonstandard intrinsic functions under the discussion of the -intrinsic setting above.

Extensions:

All of these extensions (except lower-case characters) will generate warnings if the relevant -f77 option is set. Some of the extensions listed below are part of the Fortran-90 Standard. These are indicated by the notation (F90).

Tabs are permitted, and translated into equivalent blanks which correspond to tab stops every 8 columns. The standard does not recognize tabs. Note that some compilers allow tabs, but treat them differently. The treatment defined for DEC FORTRAN can be achieved using the -source=dec-tab setting.

Strings may be delimited by either quote marks or apostrophes. A sequence of two delimiter characters is interpreted as a single embedded delimiter character. (F90)

Strings may contain UNIX-style backslash escape sequences. They will be interpreted as such if the -source=unix-backslash setting is given. Otherwise the backslash character will be treated as a normal printing character.

Source code can be in either Fortran 90 free format or traditional fixed format. (F90)

A semicolon is allowed as a statement separator. (F90)

Lower case characters are permitted, and are converted internally to uppercase except in character strings. The standard specifies upper case only, except in comments and strings. (F90)

Hollerith constants are permitted, in accordance with the Fortran 77 Standard, appendix C. They should not be used in expressions, or confused with datatype CHARACTER.

The letter 'D' (upper or lower case) in column 1 is treated as the beginning of a comment. There is no option to treat such lines as statements instead of comments.

Statements may be longer than 72 columns provided that the setting -columns was used to increase the limit. According to the standard, all text from columns 73 through 80 is ignored, and no line may be longer than 80 columns.

Variable names may be longer than six characters. The standard specifies six as the maximum. ftnchek permits names up to 31 characters long (F90).

Variable names may contain underscores and dollar signs (or other non-alphabetic characters as specified by the -identifier-chars option). These characters are are treated the same as alphabetic letters. The default type for variables beginning with these characters is REAL. In IMPLICIT type statements specifying a range of characters, the dollar sign follows Z and is followed by underscore. (Any other user-defined characters are treated the same as the dollar sign.) Fortran 90 permits underscores in variable names.

The UNIX version tolerates the presence of preprocessor directives, namely lines beginning with the pound sign (#). These are treated as comments, except for #line directives, which are interpreted, and are used to set the line number and source file name for warnings and error messages. Note that #include directives are not processed by ftnchek. Programs that use them for including source files should be passed through the preprocessor before being input to ftnchek. As noted below, ftnchek does process INCLUDE statements, which have a different syntax. An optional program, ftnpp(1L) (available separately) provides preprocessing that properly handles INCLUDE files.

The Fortran 90 DO ... ENDDO control structure is permitted. The CYCLE and EXIT statements are accepted. All of these may have an optional do-construct name, but construct names are not checked for consistency. (F90)

The Fortran 90 SELECT CASE construct is accepted. (F90)

Construct names are also accepted on IF, THEN, ELSE, ENDIF and SELECT CASE statements. (F90)

The ACCEPT and TYPE statements (for terminal I/O) are permitted, with the same syntax as PRINT.

The so-called ``Cray pointer'' syntax is tolerated. It is not the same as the Fortran 90 POINTER statement. There is no real checking of the statement other than basic syntax. The form of this statement is
POINTER (pointer, pointee) [,(pointer, pointee)]
The pointer variables are assigned a data type of INTEGER *4. Usage checking of the pointee variables is suppressed, since in practice they are accessed indirectly via the pointers.

The following Fortran 90 pointer related syntaxes are accepted: ALLOCATABLE, POINTER, and TARGET statements and the same attributes in type declarations; ALLOCATE, DEALLOCATE, and NULLIFY executable statements; pointer assignment using => operator; and the intrinsic functions ALLOCATED and ASSOCIATED. Little semantic checking of pointer variables and operations is done beyond basic set and used status. For instance, there is no checking for such errors as dangling pointers, or use of unallocated arrays.

Statements may have any number of continuation lines. The Fortran 77 and Fortran 90 standards allow a maximum of 19 in fixed source form. The Fortran 90 standard allows a maximum of 39 in free source form.

Relational (comparison) operators composed of punctuation, namely: < <= == /= > >= are allowed. (F90)

Inline comments, beginning with an exclamation mark, are permitted. (F90)

NAMELIST I/O is supported. The syntax is the same as in Fortran 90.

FORMAT statements can contain a dollar sign to indicate suppression of carriage-return. An integer expression enclosed in angle brackets can be used anywhere in a FORMAT statement where the Fortran 77 Standard allows an integer constant (except for the length of a Hollerith constant), to provide a run-time value for a repeat specification or field width.

Nonstandard keywords are allowed in I/O statements, corresponding to those in VMS Fortran.

The IMPLICIT NONE statement is supported. The meaning of this statement is that all variables must have their data types explicitly declared. Rather than flag the occurrences of such variables with syntax error messages, ftnchek waits till the end of the module, and then prints out a list of all undeclared variables, as it does for the -declare option. (F90)

Data types INTEGER, REAL, COMPLEX, and LOGICAL are allowed to have an optional precision specification in type declarations. For instance, REAL*8 means an 8-byte floating point data type. The REAL*8 datatype is not necessarily considered equivalent to DOUBLE PRECISION, depending on the -wordsize setting. The Fortran 77 Standard allows a length specification only for CHARACTER data.

ftnchek supports the DOUBLE COMPLEX type specification for a complex quantity whose real and imaginary parts are double precision. Mixed-mode arithmetic involving single-precision complex with double-precision real data, prohibited under the Standard, yields a double complex result.

Combined type declarations and data-statement-like initializers are accepted. These have the form of a standard Fortran 77 type declaration, followed by a slash-delimited list of constants like that used in a DATA statement. An example of the syntax is
INTEGER N / 100 /
This bastard form of initializing declaration was not adopted in Fortran 90. Such declarations should be written using the standard form described below, which is accepted by ftnchek.

There is limited support for Fortran 90 attribute-based type declarations. This style of declaration is distinguished by the use of a double colon (::) between the list of attributes and the list of declared variables. The features supported may be adequate for novice programmers, but are not yet sufficient for professional-quality Fortran 90 programs. I hope to add support for more features in future releases. I invite volunteers to assist in this task. See the ToDo file in the source code distribution for details. The attributes currently accepted, besides all the usual data types, are DIMENSION, EXTERNAL, INTRINSIC, PARAMETER, and SAVE. The new form of declaration also allows assignment of values to the variables declared. At present, the (LEN=value) form of specifying character lengths is also accepted. Kind specifications, using (KIND=value) are parsed but are not processed: all kinds are treated as default kind. Also, there is little checking of these declarations beyond basic syntax.

Many commonly found nonstandard intrinsic functions are provided. See the discussion of -intrinsic for a list of functions and how to control which ones are recognized.

Argument checking is not tight for those nonstandard intrinsics that take arrays or mixed argument types.

ftnchek permits the INCLUDE statement, which causes inclusion of the text of the given file. The syntax is
INCLUDE 'filename'
This is compatible with Fortran 90. If the -source=vms-include option is given, ftnchek follows VMS conventions with respect to this statement: it assumes a default extension of .for if no filename extension is given, and allows the qualifier /[NO]LIST following the filename, to control the listing of the included file. There is no support for including VMS text modules.

In diagnostic output relating to items contained in include files, the location of the error is specified by both its location in the include file and the location in the parent file where the file was included.

ftnchek accepts PARAMETER statements which lack parentheses. These will be warned about if the -f77=param-noparen flag is given.

ftnchek accepts PARAMETER definitions that involve intrinsic functions and exponentiation by a non-integer exponent. Both of these cases are prohibited by the Fortran 77 Standard, and will be warned about if the -f77=param-intrinsic flag is given. If an intrinsic function value is a compile-time integer constant, ftnchek will evaluate it. This allows better checking if the parameter is used in declaring array sizes. Fortran 90 allows intrinsic functions in PARAMETER definitions.

The intrinsic functions that are evaluated are:

ABS	IABS	DIM	IDIM	MAX
MAX0	MIN	MIN0	MOD	SIGN
ISIGN	LEN	ICHAR	INDEX

The functions of integer arguments are evaluated only if the arguments are integer constant expressions. (These may involve integer constants, parameters, and evaluated intrinsic functions.) The function LEN is evaluated if its argument is an expression involving only character constants and variables whose length is not adjustable. The functions ICHAR and INDEX are evaluated only if the arguments are character constants. ftnchek gives a warning if it needs the value of some intrinsic function that is not evaluated.

NEW FEATURES

Here are the changes from Version 3.2 to Version 3.3:

1.: Front-end has been rewritten for unlimited lookahead, eliminating the longstanding bug that caused incorrect interpretation of statements whose ambiguity was not resolved in the first line.
2.: The -mkhtml option is now available in the MS-DOS version.
3.: Added support for Fortran 90 pointer related syntax: ALLOCATE, DEALLOCATE, and NULLIFY statements; the ALLOCATABLE, POINTER and TARGET attributes in type declarations; the pointer assigment operator => and intrinsic functions ALLOCATED and ASSOCIATED; and deferred-shape array declarations. At present these new syntax features are accepted but not properly checked. This feature was added by Robert Landrito.
4.: The -f77 and -f90 pointer option controlling warnings about ``Cray pointers'' has been renamed to cray-pointer. The -f77=pointer option now instead controls warnings for code containing Fortran 90 pointer-related syntax.
5.: Re-implemented -mkhtml processing so it is now much faster on source files containing many routines.
6.: Changed the arrangement of the test directory so there is no longer any need to modify the distribution in order to run the test suite (check.bat) under MS-DOS.
7.: Fixed bug in reading numeric settings on command line when setting name abbreviated to 3 characters.
8.: Fixed bug causing spurious warning for a GOTO referring to a labeled END statement when the statement before END was a FORMAT.
9.: New flag -f77=character to control warnings about extensions to the Fortran 77 character data type. Accompanying this new flag is support for Fortran 90 rules for character variable declarations that evaluate to zero or negative length, allowing them and treating negative length values as zero.
10.: Fixed minor bug in printing of comments and blank lines following last END statement in -list mode.

BUGS

ftnchek still has much room for improvement. Your feedback is appreciated. We want to know about any bugs you notice. Bugs include not only cases in which ftnchek issues an error message where no error exists, but also if ftnchek fails to issue a warning when it ought to. Note, however, that ftnchek is not intended to catch all syntax errors (see section on Limitations). Also, it is not considered a bug for a variable to be reported as used before set, if the reason is that the usage of the variable occurs prior in the text to where the variable is set. For instance, this could occur when a GOTO causes execution to loop backward to some previously skipped statements. ftnchek does not analyze the program flow, but assumes that statements occurring earlier in the text are executed before the following ones.

We especially want to know if ftnchek crashes for any reason. It is not supposed to crash, even on programs with syntax errors. Suggestions are welcomed for additional features which you would find useful. Tell us if any of ftnchek's messages are incomprehensible. Comments on the readability and accuracy of this document are also welcome.

You may also suggest support for additional extensions to the Fortran language. These will be included only if it is felt that the extensions are sufficiently widely accepted by compilers.

If you find a bug in ftnchek, first consult the list of known bugs below to see if it has already been reported. Also check the section entitled ``Limitations and Extensions'' above for restrictions that could be causing the problem. If you do not find the problem documented in either place, then send a report including

1.: The operating system and CPU type on which ftnchek is running.
2.: The version of ftnchek and values of any environment options or settings defined in startup file. (Capturing the output of ftnchek -help is useful for this.)
3.: A brief description of the bug.
4.: If possible, a small sample program showing the bug.

The report should be sent to Dr. Robert Moniot (see contact information in section entitled ``Installation and Support'').

Highest priority will be given to bugs which cause ftnchek to crash.

Certain problems that arise when checking large programs can be fixed by increasing the sizes of the data areas in ftnchek. (These problems are generally signaled by error messages beginning with ``Oops''.) The simplest way to increase the table sizes is by recompiling ftnchek with the LARGE_MACHINE macro name defined. Consult the makefile and README file for the method of doing this.

The following is a list of known bugs.

1.

Bug: Used-before-set message is suppressed for any variable which is used as the loop index in an implied-do loop, even if it was in fact used before being set in some earlier statement. For example, consider J in the statement

      WRITE(5,*) (A(J), J=1,10)

Here ftnchek parses the I/O expression, A(J), where J is used, before it parses the implied loop where J is set. Normally this would cause ftnchek to report a spurious used-before-set warning for J. Since this report is usually in error and occurs fairly commonly, ftnchek suppresses the warning for J altogether.

Prognosis: A future version of ftnchek is planned which will handle implied-do loops correctly.

2.

Bug: Variables used (not as arguments) in statement-function subprograms do not have their usage status updated when the statement function is invoked.

: Prognosis: To be fixed in a future version of ftnchek.

3.

Bug: VAX version does not expand wildcards in filenames on the command line if they are followed without space by an option, e.g. ftnchek *.f/calltree would not expand the *.f. This is because VMS-style options without intervening space are not supported by the GNU shell_mung routine that is used to expand wildcards.

: Prognosis: unlikely to be fixed.

4.

Bug: checking for nonstandard format edit descriptors is done only in FORMAT statements, not in character strings used as formats.

: Prognosis: may be fixed someday.

ACKNOWLEDGEMENTS

ftnchek was designed by Dr. Robert Moniot, professor at Fordham University. During the academic year of 1988-1989, Michael Myers and Lucia Spagnuolo developed the program to perform the variable usage checks. During the following year it was augmented by Lois Bigbie to check subprogram arguments and COMMON block declarations. Brian Downing assisted with the implementation of the INCLUDE statement. John Quinn wrote the common block usage checks. Heba Elsayed wrote the label table printout and label usage checks. Nelson H. F. Beebe of the University of Utah added most of the new code to implement the -makedcls feature and wrote the dcl2inc script. The -mkhtml feature was contributed by Mark McVeigh of Framatome ANP, Inc. The -reference feature was contributed by Gerome Emmanuel, Ecole des mines, U. Nancy (slightly modified). The -vcg option was contributed by Dr. Philip Rubini of Cranfield University, UK. The support for Cray pointer syntax was provided by John Dannenhoffer of United Technologies Research Center. John C. Bollinger of Indiana University added the parser syntax for the SELECT CASE construct. Robert Landrito added the parser syntax for F90 pointer-related features. Additional features will be added as time permits. As of Version 2.5, the name was changed from forchek to ftnchek, to avoid confusion with a similar program named forcheck, developed earlier at Leiden University.

We would like to thank John Amor of the University of British Columbia, Reg Clemens of the Air Force Phillips Lab in Albuquerque, Markus Draxler of the University of Stuttgart, Victor Eijkhout of the University of Tennessee at Knoxville, Greg Flint of Purdue University, Daniel P. Giesy of NASA Langley Research Center, Fritz Keinert of Iowa State University, Judah Milgram of the University of Maryland College Park, Hugh Nicholas of the Pittsburgh Supercomputing Center, Dan Severance of Yale University, Phil Sterne of Lawrence Livermore National Laboratory, Larry Weissman of the University of Washington, Warren J. Wiscombe of NASA Goddard, and Nelson H. F. Beebe of the University of Utah, for pointing out bugs and suggesting some improvements. Stefan A. Deutscher, Gunnar Duus, Clive Page of the University of Leicester, Stephan Wefing of Heidelberg University, and Bob Wells of Oxford University were extremely helpful as alpha testers. We also thank Jack Dongarra for putting ftnchek into the netlib library of publicly available software.

INSTALLATION AND SUPPORT

The ftnchek program is free software. It can be obtained by anonymous ftp from many software servers, including ftp://netlib.org/fortran . Note that on Netlib the distribution is named ftnchek.tar.gz whereas on most other servers the file name includes the version number, e.g. ftnchek-3.3.0.tar.gz. If the file extension is .Z, uncompress with the Unix uncompress(1) utility. If the file extension is .gz, uncompress with the GNU gunzip(1L) program. Then use tar(1) to unpack the files into a subdirectory.

Installation requires a C compiler for your computer. See the INSTALL file provided with the distribution for instructions on installing ftnchek on your system. Executable binary for particular systems such as IBM PC or Macintosh, as available, can be obtained by anonymous ftp from ftp://ftp.dsm.fordham.edu/pub/ftnchek . Assistance in preparing such executable binary forms is welcome.

The nroff version of this document is named ftnchek.man. On UNIX systems, this file can be used as the man page, but actually it is a multi-purpose source file which is used to produce the other forms of the documentation. The cleaned-up man page document, created during installation of ftnchek, is named ftnchek.1. The distribution also includes a plain ASCII version named ftnchek.doc, a PostScript version named ftnchek.ps, an HTML version in directory html, and a VMS HELP version named ftnchek.hlp.

Information about the latest version and the status of the project can be obtained by visiting ftnchek's home page, http://www.dsm.fordham.edu/~ftnchek . For further information and to report bugs, you may contact Dr. Robert Moniot, whose contact information can be found by a Web search for his name and Fordham University. (E-mail address is not provided here because it attracts unsolicited commercial e-mail, but it is easily constructed by combining his last name with the name of the university and the edu domain.)

ACTION	PAD	READWRITE
ADVANCE	POSITION	SIZE
DELIM	READ	WRITE
EOR

BLOCKSIZE	EXTENDSIZE	READONLY
BUFFERCOUNT	INITIALSIZE	RECORDSIZE
CARRIAGECONTROL	MAXREC	RECORDTYPE
DEFAULTFILE	NAME (in OPEN)	SHARED
DISP	NOSPANBLOCK	TYPE
DISPOSE	ORGANIZATION

accept-type	double-complex	param-noparen
backslash	format-dollarsign	cray-pointer
byte	format-edit-descr	quad-constant
cpp	function-noparen	type-size
d-comment	name-dollarsign	variable-format
dec-tab	param-implicit-type	vms-io

CDABS	CDSQRT	DREAL	ZLOG
CDCOS	DCMPLX	IMAG	ZSIN
CDEXP	DCONJG	ZABS	ZSQRT
CDLOG	DIMAG	ZEXP	ZCOS
CDSIN

CQABS	QARCOS	QEXT	QNINT
CQCOS	QARSIN	QEXTD	QPROD
CQEXP	QATAN	QFLOAT	QREAL
CQLOG	QATAN2	QIMAG	QSIGN
CQSIN	QCMPLX	QINT	QSIN
CQSQRT	QCONJG	QLOG	QSINH
DBLEQ	QCOS	QLOG10	QSQRT
IQINT	QCOSH	QMAX1	QTAN
IQNINT	QDIM	QMIN1	QTANH
QABS	QEXP	QMOD	SNGLQ