The type printed will usually contain one of the words text (the file contains only ASCII characters and is probably safe to read on an ASCII terminal), executable (the file contains the result of compiling a program in a form understandable to some \s-1UNIX\s0 kernel or another), or data meaning anything else (data is usually `binary' or non-printable). Exceptions are well-known file formats (core files, tar archives) that are known to contain binary data. When modifying the file /etc/magic or the program itself, "preserve these keywords" . People depend on knowing that all the readable files in a directory have the word ``text'' printed. Don't do as Berkeley did - change ``shell commands text'' to ``shell script''.
The filesystem tests are based on examining the return from a stat (2) system call. The program checks to see if the file is empty, or if it's some sort of special file. Any known file types appropriate to the system you are running on (sockets, symbolic links, or named pipes (FIFOs) on those systems that implement them) are intuited if they are defined in the system header file sys/stat.h .
The magic number tests are used to check for files with data in particular fixed formats. The canonical example of this is a binary executable (compiled program) a.out file, whose format is defined in a.out.h and possibly exec.h in the standard include directory. These files have a `magic number' stored in a particular place near the beginning of the file that tells the \s-1UNIX\s0 operating system that the file is a binary executable, and which of several types thereof. The concept of `magic number' has been applied by extension to data files. Any file with some invariant identifier at a small fixed offset into the file can usually be described in this way. The information in these files is read from the magic file /etc/magic.
If an argument appears to be an ASCII file, file attempts to guess its language. The language tests look for particular strings (cf names.h ) that can appear anywhere in the first few blocks of a file. For example, the keyword .br indicates that the file is most likely a troff (1) input file, just as the keyword struct indicates a C program. These tests are less reliable than the previous two groups, so they are performed last. The language test routines also test for some miscellany (such as tar (1) archives) and determine whether an unknown file should be labelled as `ascii text' or `data'.
8 -v Print the version of the program and exit.
8 -m list Specify an alternate list of files containing magic numbers. This can be a single file, or a colon-separated list of files.
8 -z Try to look inside compressed files.
8 -c Cause a checking printout of the parsed form of the magic file. This is usually used in conjunction with -m to debug a new magic file before installing it.
8 -f namefile Read the names of the files to be examined from namefile (one per line) before the argument list. Either namefile or at least one filename argument must be present; to test the standard input, use ``-'' as a filename argument.
8 -L option causes symlinks to be followed, as the like-named option in ls (1). (on systems that support symbolic links).
strings (1), " od" (1) - tools for examining non-textfiles.
The one significant difference between this version and System V is that this version treats any white space as a delimiter, so that spaces in pattern strings must be escaped. For example,
>10 string language impress (imPRESS data)
in an existing magic file would have to be changed to
>10 string language\e impress (imPRESS data)
In addition, in this version, if a pattern string contains a backslash, it must be escaped. For example
0 string \ebegindata Andrew Toolkit document
in an existing magic file would have to be changed to
0 string \e\ebegindata Andrew Toolkit document
SunOS releases 3.2 and later from Sun Microsystems include a file (1) command derived from the System V one, but with some extensions. My version differs from Sun's only in minor ways. It includes the extension of the `&' operator, used as, for example,
>16 long&0x7fffffff >0 not stripped
The order of entries in the magic file is significant. Depending on what system you are using, the order that they are put together may be incorrect. If your old file command uses a magic file, keep the old magic file around for comparison purposes (rename it to /etc/magic.orig ).
This program, based on the System V version, was written by Ian Darwin without looking at anybody else's source code.
John Gilmore revised the code extensively, making it better than the first version. Geoff Collyer found several inadequacies and provided some magic file entries. The program has undergone continued evolution since.
Altered by Rob McMahon, cudcv@warwick.ac.uk, 1989, to extend the `&' operator from simple `x&y != 0' to `x&y op z'.
Altered by Guy Harris, guy@auspex.com, 1993, to:
put the ``old-style'' `&' operator back the way it was, because 1) Rob McMahon's change broke the previous style of usage, 2) the SunOS ``new-style'' `&' operator, which this version of file supports, also handles `x&y op z', and 3) Rob's change wasn't documented in any case;
put in multiple levels of `>';
put in ``beshort'', ``leshort'', etc. keywords to look at numbers in the file in a specific byte order, rather than in the native byte order of the process running file .
Changes by Ian Darwin and various authors including Christos Zoulas (christos@deshaw.com), 1990-1997.
A few support files (getopt, strtok) distributed with this package are by Henry Spencer and are subject to the same terms as above.
A few simple support files (strtol, strchr) distributed with this package are in the public domain; they are so marked.
The files tar.h and is_tar.c were written by John Gilmore from his public-domain tar program, and are not covered by the above restrictions.
File uses several algorithms that favor speed over accuracy, thus it can be misled about the contents of ASCII files.
The support for ASCII files (primarily for programming languages) is simplistic, inefficient and requires recompilation to update.
There should be an ``else'' clause to follow a series of continuation lines.
The magic file and keywords should have regular expression support. Their use of "ASCII TAB" as a field delimiter is ugly and makes it hard to edit the files, but is entrenched.
It might be advisable to allow upper-case letters in keywords for e.g., troff (1) commands vs man page macros. Regular expression support would make this easy.
The program doesn't grok \s-2FORTRAN\s0. It should be able to figure \s-2FORTRAN\s0 by seeing some keywords which appear indented at the start of line. Regular expression support would make this easy.
The list of keywords in ascmagic probably belongs in the Magic file. This could be done by using some keyword like `*' for the offset value.
Another optimisation would be to sort the magic file so that we can just run down all the tests for the first byte, first word, first long, etc, once we have fetched it. Complain about conflicts in the magic file entries. Make a rule that the magic entries sort based on file offset rather than position within the magic file?
The program should provide a way to give an estimate of ``how good'' a guess is. We end up removing guesses (e.g. ``From '' as first 5 chars of file) because they are not as good as other guesses (e.g. ``Newsgroups:'' versus "Return-Path:"). Still, if the others don't pan out, it should be possible to use the first guess.
This program is slower than some vendors' file commands.
This manual page, and particularly this section, is too long.