The IANA type registration says that the file must begin with the string
"(This file must be converted with BinHex 4.0)", but the binhex utility
generates files that begin with "(This file must be converted; you knew
that already.)".

https://www.iana.org/assignments/media-types/application/mac-binhex40
https://datatracker.ietf.org/doc/html/rfc1741
http://fileformats.archiveteam.org/wiki/Binhex
https://github.com/VorpalBlade/macutils/blob/640782c6698a4d360711bbfb71a0d27dbd182d8f/binhex/dofile.c#L30

application/x-msdownload is what Microsoft Windows uses for EXE and DLL
files, so that type is commonly found on the WWW and in other contexts.
It is an umbrella type for anything with a .exe or .dll file extension,
among others.

application/x-dosexec is what the `file` utility uses for DOS
executables specifically. According to the File Format Problem wiki:
"It's not clear if there is any completely reliable way to identify a
file as strictly DOS EXE, except in the negative (i.e., it looks like
EXE, and is not a valid NE, PE, etc., file)...If the relocation table
offset is from 28 to 63...it is pretty certainly DOS EXE."

application/x-ms-ne-executable is what the `file` utility uses for
16-bit Windows libraries and executables.

application/vnd.microsoft.portable-executable is the IANA standard type
for 32-bit and 64-bit Windows libraries and executables, as well as EFI
executables.

With these changes, application/x-ms-dos-executable is considered
deprecated. For now it is an alias of application/x-msdownload, but in
the future it may change to be an alias of application/x-dosexec.

helloworld-*.exe was compiled with Open Watcom from the following C
source code:

    #include <stdio.h>

    int main(void)
    {
        puts("Hello World!");
        return 0;
    }

https://github.com/file/file/blob/171c73b67ca692550882e0b8ab3c2a1f16ab1e00/magic/Magdir/msdos
https://www.iana.org/assignments/media-types/application/vnd.microsoft.portable-executable
http://fileformats.archiveteam.org/wiki/MS-DOS_EXE
http://fileformats.archiveteam.org/wiki/New_Executable
http://fileformats.archiveteam.org/wiki/Portable_Executable

MSIX and APPX have the same descriptions because MSIX is essentially the
new version of APPX. For now, app store apps have to be unzipped before
running them in Wine, but I expect that a future version of Wine will
support double-click installation.

https://learn.microsoft.com/en-us/windows/msix/app-installer/web-install-iis#step-7---configure-the-web-app-for-app-package-mime-types

FORMAT= in the first 128 bytes is very unspecific magic, it happens
to match pcb-drillFile.drl too (which is a application/x-excellon file,
and can be detected as such because it starts with M48)

Found when porting QtCore to shared-mime-info 2.3

PowerShell is a cross-platform scripting language with implementations
for Windows, Mac, and Linux.

application/x-powershell is not an official IANA media type, but is used
by GitHub and others.

https://github.com/github-linguist/linguist/blob/7ca3799b8b5f1acde1dd7a8dfb7ae849d3dfb4cd/lib/linguist/languages.yml#L5320
https://learn.microsoft.com/en-us/powershell/module/microsoft.powershell.core/about/about_requires?view=powershell-7.3#-psedition-psedition-name

WASM (WebAssembly) is a portable object format.

This is an official IANA media type.

Use bzip1 for (these days rarely used) *.bz files

This preserves source compatiblity for code that was
using application/x-bzip-compressed-tar to detect tar.bz2 files.

This is MPlayer's own subtitle format.

These subtitles do not belong to either MPlayer or its fork MPlayer2.

"HTML Components" is a file format and API promoted by Microsoft that
never became popular. Nonetheless, Wine supports it and would benefit
from having the MIME type in the shared database.

https://www.w3.org/TR/NOTE-HTMLComponents
https://learn.microsoft.com/en-us/previous-versions/windows/it-pro/windows-2000-server/bb742440(v=technet.10)#mime-formats-in-iis-in-extension-order

Otherwise it's unclear which of image/png and image/apng should win.
The spec doesn't say that one should check if one match is a subclass
of another, for magic matching, it says priority is used for this:

"Low numbers should be used for more generic types (such as 'gzip
compressed data') and higher values for specific subtypes (such as a
word processor format that happens to use gzip to compress the file)."

Found when porting QtCore to shared-mime-info-2.3.

JFIF is the technical name for the on-disk JPEG file format, and
consequently JPEG files are occasionally found with a .jfif extension.

https://www.iana.org/assignments/media-types/application/its+xml

To avoid misdetection as application/x-virtual-boy-rom, look for the
three things that are most commonly at the beginning of a VB.NET source
file: An "Imports" statement, a "Module" statement, or a "REM" comment.

https://github.com/github-linguist/linguist/blob/7ca3799b8b5f1acde1dd7a8dfb7ae849d3dfb4cd/lib/linguist/languages.yml#L7419

Files with a .bas extension contain code in the BASIC programming
language or one of its derivatives, such as QBasic, Visual Basic, or
QB64.

https://ciam-tika.straumann.com/mime-types/text/x-basic

Fixes: #209

This is the Quake PAK format, and it does start with PACK,
but the real signature requires parsing the file deeper,
as explained on https://web.archive.org/web/20070701221700/http://mediasrv.ns.ac.yu:80/extra/fileformat/games/pak/pak.txt

References:
* https://bugs.freedesktop.org/show_bug.cgi?id=10030
* https://invent.kde.org/unmaintained/kdelibs/-/blob/KDE/3.5/kio/magic?ref_type=heads

Fixes #210

JSON5 is a superset of standard JSON that permits comments and trailing
commas. Although application/json5 is not an official IANA media type,
it is used by Nim, Fastify, Altova, and others.

https://nim-lang.org/1.6.6/mimetypes.html
https://www.jsdelivr.com/package/npm/fastify-json5
https://www.altova.com/manual/Mapforce/mapforceenterprise/mff_json5.html

This is the MIME type that Internet Explorer accepts for
pseudo-base64-encoded VBScript files, which typically have a .vbe
extension. Use of this file format is discouraged and files of this type
should be treated with suspicion.

https://book.hacktricks.xyz/windows-hardening/ntlm/places-to-steal-ntlm-creds#vbscript

This is the MIME type that Internet Explorer accepts for
pseudo-base64-encoded JScript files, which typically have a .jse
extension. Use of this file format is discouraged and files of this type
should be treated with suspicion.

https://www.virusbulletin.com/virusbulletin/2006/01/inside-microsoft-script-encoder/
https://book.hacktricks.xyz/windows-hardening/ntlm/places-to-steal-ntlm-creds#jscript

text/jscript is used for Microsoft's implementation of ECMAScript, which
has a few quirks and extra features compared to JavaScript, but no
obvious distinguishing marks.

All ECMAScript programs are JavaScript, but not all JavaScript programs
are ECMAScript.

https://www.iana.org/assignments/media-types/application/vnd.cups-ppd

This is more or less the Windows equivalent of a .desktop file. Opening
a .lnk file in Wine opens the target program, file, or directory.

C.lnk is a shortcut to C:\ made on Windows 10.

https://gitlab.winehq.org/wine/wine/-/blob/209b2c23c5e302ed9211d67ab044386d3fe0bef4/programs/winemenubuilder/winemenubuilder.c#L2017
https://invent.kde.org/frameworks/kcoreaddons/-/blob/66983a7c188d1b19c5dabbb056f6fac90803db5f/src/mimetypes/kde6.xml#L2591
https://winprotocoldoc.blob.core.windows.net/productionwindowsarchives/MS-SHLLINK/[MS-SHLLINK].pdf#page=11

Now that there is a MIME type for bzip, the deprecated predecessor
of bzip2, it is necessary to keep bzip and bzip2 compression apart.

This fixes the compiler warnings similar as following.

../src/update-mime-database.cpp:393:50: warning: missing sentinel in function call [-Wsentinel]
  393 |                         g_strconcat(namespaceURI, " ", localName, NULL),
      |                                                                       ^
      |                                                                       , nullptr

The `has_function` feature in meson uses different detection methods
depending on the contents of the `prefix` kwarg [1]:

 * if it contains `#include` directives it will copy the prefix into
   the test code and check if it compiles
 * if it doesn't contain an include or isn't specified, `has_function`
   will forward declare the function and test for it's existence by
   trying to link it to the default libraries

The latter approach wrongly succeeds for `fdatasync` on darwin because
the linker binds the function to a system call of the same name. Note
that this result really is wrong because that system call has not
the expected semantics of `fdatasync`.

By adding an include for `unistd.h` we can get meson to use the
first approach and the detection fails.

Note that this has gone unnoticed so far because only recent versions
of clang (the default compiler on darwin) started to treat implicit
function declarations as an error.

[1] https://github.com/mesonbuild/meson/blob/583d2815d1a130227f0f4db47e4ab2e80ebb6a61/mesonbuild/compilers/mixins/clike.py#L839-L846

Fixes #211

Clang is not able to disambiguate between multiple string literatals
and C++11 user defined literals. Spaces help.