Read-only mirror of file CVS repository, updated every half hour. NOTE: do not make pull requests here, nor comment any commits, submit them usual way to bug tr
Read-only mirror of file CVS repository, updated every half hour. NOTE: do not make pull requests here, nor comment any commits, submit them usual way to bug tr
@(#) $File: README.md,v 1.8 2026/08/28 15:26:23 christos Exp $Phone: Do not even think of telephoning me about this program. Send cash first!
This is Release 5.x of Ian Darwin's (copyright but distributable) file(1) command, an implementation of the Unix File(1) command. It knows the 'magic number' of several thousands of file types. This version is the standard "file" command for Linux, *BSD, and other systems. (See "patchlevel.h" for the exact release number).
The major changes for 5.x are CDF file parsing, indirect magic, name/use (recursion) and overhaul in mime and ascii encoding handling.
The major feature of 4.x is the refactoring of the code into a library, and the re-write of the file command in terms of that library. The library itself, libmagic can be used by 3rd party programs that wish to identify file types without having to fork() and exec() file. The prime contributor for 4.0 was Mans Rullgard.
UNIX is a trademark of UNIX System Laboratories.
The prime contributor to Release 3.8 was Guy Harris, who put in megachanges including byte-order independence.
The prime contributor to Release 3.0 was Christos Zoulas, who put in hundreds of lines of source code changes, including his own ANSIfication of the code (I liked my own ANSIfication better, but his (__P()) is the "Berkeley standard" way of doing it, and I wanted UCB to include the code...), his HP-like "indirection" (a feature of the HP file command, I think), and his mods that finally got the uncompress (-z) mode finished and working.
This release has compiled in numerous environments; see PORTING for a list and problems.
This fine freeware file(1) follows the USG (System V) model of the file command, rather than the Research (V7) version or the V7-derived 4.[23] Berkeley one. That is, the file /etc/magic contains much of the ritual information that is the source of this program's power. My version knows a little more magic (including tar archives) than System V; the /etc/magic parsing seems to be compatible with the (poorly documented) System V /etc/magic format (with one exception; see the man page).
In addition, the /etc/magic file is built from a subdirectory for easier(?) maintenance. I will act as a clearinghouse for magic numbers assigned to all sorts of data files that are in reasonable circulation. Send your magic numbers, in magic(5) format please, to the maintainer, Christos Zoulas.
COPYING - read this first.
ChangeLog - log of important changesREADME.md - read this second (you are currently reading this file).INSTALL - read on how to installsrc/apprentice.c - parses /etc/magic to learn magicsrc/apptype.c - used for OS/2 specific application type magicsrc/ascmagic.c - third & last set of tests, based on hardwired assumptions.src/asctime_r.c - replacement for OS's that don't have it.src/asprintf.c - replacement for OS's that don't have it.src/buffer.c - buffer handling functions.src/cdf.[ch] - parser for Microsoft Compound Document Filessrc/cdf_time.c - time converter for CDF.src/compress.c - handles decompressing files to look inside.src/ctime_r.c - replacement for OS's that don't have it.src/der.[ch] - parser for Distinguished Encoding Rulessrc/dprintf.c - replacement for OS's that don't have it.src/elfclass.h - common code for elf 32/64.src/encoding.c - handles unicode encodingssrc/file.c - the main programsrc/file.h - header filesrc/file_opts.h - list of optionssrc/fmtcheck.c - replacement for OS's that don't have it.src/fsmagic.c - first set of tests the program runs, based on filesystem infosrc/funcs.c - utilility functionssrc/getline.c - replacement for OS's that don't have it.src/getopt_long.c - replacement for OS's that don't have it.src/gmtime_r.c - replacement for OS's that don't have it.src/is_csv.c - knows about Comma Separated Value file format (RFC 4180).src/is_json.c - knows about JavaScript Object Notation format (RFC 8259).src/is_simh.c - knows about SIMH tape file format.src/is_tar.c, tar.h - knows about Tape ARchive format (courtesy John Gilmore).src/landlock.c - linux landlock protectionsrc/localtime_r.c - replacement for OS's that don't have it.src/magic.h.in - source file for magic.hsrc/mygetopt.h - replacement for OS's that don't have it.src/magic.c - the libmagic apisrc/names.h - header file for ascmagic.csrc/pread.c - replacement for OS's that don't have it.src/print.c - print results, errors, warnings.src/readcdf.c - CDF wrapper.src/readelf.[ch] - Stand-alone elf parsing code.src/softmagic.c - 2nd set of tests, based on /etc/magicsrc/swap.h - byte swappingsrc/swap.c - byte swappingsrc/mygetopt.h - replacement for OS's that don't have it.src/seccomp.c - linux seccomp protectionsrc/strcasestr.c - replacement for OS's that don't have it.src/strlcat.c - replacement for OS's that don't have it.src/strlcpy.c - replacement for OS's that don't have it.src/strndup.c - replacement for OS's that don't have it.src/tar.h - tar file definitionssrc/vasprintf.c - for systems that don't have it.doc/file.man - man page for the commanddoc/magic.man - man page for the magic file, courtesy Guy Harris.
Install as magic.4 on USG and magic.5 on V7 or Berkeley; cf Makefile.If you submit a new magic entry please make sure you read the following guidelines:
Initial match is preferably at least 32 bits long, and is a unique match
If this is not feasible, use additional check
Match of <= 16 bits are not accepted
Delay printing string as much as possible, don't print output too early
Avoid printf arbitrary byte as string, which can be a source of crash and buffer overflow
Provide complete information with entry:
$ gpg --verify file-X.YY.tar.gz.asc file-X.YY.tar.gz
gpg: assuming signed data in `file-X.YY.tar.gz'
gpg: Signature made WWW MMM DD HH:MM:SS YYYY ZZZ using DSA key ID KKKKKKKKTo download the key:
$ gpg --keyserver hkp://keys.gnupg.net --recv-keys KKKKKKKKParts of this software were developed at SoftQuad Inc., developers of SGML/HTML/XML publishing software, in Toronto, Canada. SoftQuad was swallowed up by Corel in 2002 and does not exist any longer.
libmagic and file cross-compile for AmigaOS 3 (m68k), AmigaOS 4 (PPC) and
MorphOS (PPC) using the sacredbanana/amiga-compiler toolchain images:
| system | image tag | host triplet |
|---|---|---|
| AmigaOS 3 | sacredbanana/amiga-compiler:m68k-amigaos |
m68k-unknown-amigaos |
| AmigaOS 4 | sacredbanana/amiga-compiler:ppc-amigaos |
ppc-amigaos |
| MorphOS | sacredbanana/amiga-compiler:ppc-morphos |
ppc-morphos |
Three scripts at the repository root drive this; run autoreconf -i once first.
amiga-variants.sh -- the shared table of every variant that is built, one
entry per tree|slot|configure args, plus the configure flags common to
all of them. Both scripts below source it, so they can never disagree.build-amiga-sdk.sh [OS ...] -- builds every variant and installs
libmagic.a plus magic.h into an SDK tree. The location comes from
$AMIGA_SDK (default /opt/amiga); either a <target>/ or an
AmigaSDK-gcc-style amigaos3/sdk/<target>/ layout is accepted, and an OS
whose subdirectory is absent is skipped rather than failing.package-amiga.sh [OS ...] -- builds every variant plus one file binary
per OS and produces the two distribution archives in dist/.Both default to all three systems. The archives are laid out as:
magic.readme
magic/C/magic.mgc shared -- it is not machine code
magic/AmigaOS3/{libnix,clib2}/{include,lib/...}
magic/AmigaOS4/{include,lib/clib4}
magic/MorphOS/{include,lib,lib/libb32,lib/libnix,lib/libb32/libnix}
file.readme
file/{AmigaOS3,AmigaOS4,MorphOS}/C/{file,magic.mgc}package-amiga.sh needs a native magic.mgc to ship, so build the host copy
into build-native/ first; it is refused otherwise.
Note that m68k and PPC lay their C runtimes out differently, and the variant
table follows each. On m68k a runtime is a separate tree (libnix/, clib2/),
each with its own include/ and lib/. On PPC a runtime is a multilib slot
beneath a single shared lib/ -- -mcrt=clib4 searches <target>/lib/clib4/
and takes headers from <target>/include/. This was read off
ppc-amigaos-gcc -mcrt=clib4 -print-search-dirs rather than assumed, because
installing to the wrong one produces a library the compiler silently never
finds.
.github/workflows/amiga-packages.yml runs the whole packaging path on every
push to master (and on demand via Run workflow), so the commit gets a tick
when both archives build and verify and a cross when anything fails. It
uploads magic.lha and file.lha as run artifacts, named magic-lha and
file-lha; GitHub always wraps an artifact in a zip, so the download is a zip
containing the untouched .lha.
The runner builds LHa for UNIX (jca02266/lha, pinned to a release tag) from
source, because Debian and Ubuntu only ship lhasa, which extracts .lha but
cannot create one. After packaging, the workflow lists both archives and checks
the counts inside them -- one libmagic.a per variant, one magic.h per
per-OS tree, one file and one magic.mgc per OS, one copy of each readme --
with the expected numbers derived from amiga-variants.sh so the check cannot
go stale when a slot is added. That matters because a build failure now makes
package-amiga.sh exit non-zero, but a missing copy would otherwise leave a
perfectly valid, quietly incomplete archive.
docker run --rm -v ${PWD}:/work -e USER=$( id -u ) -e GROUP=$( id -g ) \
-it sacredbanana/amiga-compiler:m68k-amigaos bash
mkdir build-amiga && cd build-amiga
../configure --host=m68k-unknown-amigaos --disable-shared --enable-static \
--disable-zlib --disable-bzlib --disable-xzlib --disable-zstdlib \
--disable-lzlib --disable-lrziplib --disable-lz4lib \
--disable-libseccomp --disable-landlock CFLAGS="-O2"
make -C src magic.h && make -C src libmagic.lalibmagic.a is left in build-amiga/src/.libs/, and the public header
magic.h is generated into build-amiga/src/.
These change the ABI, so they are applied to the compiler driver before any
of configure's feature tests run -- which matters because the C runtime
choice selects a different set of target headers, and a probe run with the
wrong ones produces a wrong config.h. configure classifies the target from
--host and only honours the options for that system.
AmigaOS 3 (m68k)
No open issues yet, or sync has not completed.