This article possibly contains original research. (August 2009) (Learn how and when to remove this template message)
This article may be confusing or unclear to readers. (April 2018) (Learn how and when to remove this template message)
A portable application (portable app), sometimes also called standalone, is a program designed to read and write its configuration settings into an accessible folder in the computer, usually in the folder where the portable application can be found. This makes it easier to transfer the program with the user's preferences and data between different computers. A program that doesn't have any configuration options can also be a portable application.
Portable applications can be stored on any data storage device, including internal mass storage, a file share, cloud storage or external storage such as USB drives and floppy disks--storing its program files and any configuration information and data on the storage medium alone. If no configuration information is required a portable program can be run from read-only storage such as CD-ROMs and DVD-ROMs. Some applications are available in both installable and portable versions.
Some applications which are not portable by default do support optional portability through other mechanisms, the most common being command-line arguments. Examples might include
/portable to simply instruct the program to behave as a portable program, or
--cfg=/path/inifile to specify the configuration file location.
Like any application, portable applications must be compatible with the computer system hardware and operating system.
Depending on the operating system, portability is more or less complex to implement; to operating systems such as AmigaOS, all applications are by definition portable.
A portable application does not leave its files or settings on the host computer or modify the existing system and its configuration. The application does not write to the Windows registry or store its configuration files (such as an INI file) in the user's profile; instead, it stores its configuration files in the portable directory. Another requirement, since file paths will often differ on changing computers due to variation in Windows drive letter assignments, is the need for applications to store them in a relative format. While some applications have options to support this behavior, many programs are not designed to do this. A common technique for such programs is the use of a launcher program to copy necessary settings and files to the host computer when the application starts and move them back to the application's directory when it closes.
An alternative strategy for achieving application portability within Windows, without requiring application source code changes, is application virtualization: An application is "sequenced" or "packaged" against a runtime layer that transparently intercepts its file system and registry calls, then redirects these to other persistent storage without the application's knowledge. This approach leaves the application itself unchanged, yet portable.
The same approach is used for individual application components: run-time libraries, COM components or ActiveX, not only for the entire application. As a result, when individual components are ported in such manner they are able to be: integrated into original portable applications, repeatedly instantiated (virtually installed) with different configurations/settings on the same operating system (OS) without mutual conflicts. As the ported components do not affect the OS-protected related entities (registry and files), the components will not require administrative privileges for installation and management.
Microsoft saw the need for an application-specific registry for its Windows operating system as far back as 2005. It eventually incorporated some of this technology, using the techniques mentioned above, via its Application Compatibility Database  using its Detours  code library, into Windows XP. It did not make any of this technology available via its system APIs.
Programs written with a Unix-like base in mind often do not make any assumptions. Whereas many Windows programs assume the user is an administrator--something very prevalent in the days of Windows 95/98/ME (and to some degree in Windows XP/2000, though not in Windows Vista or Windows 7)--such would quickly result in "Permission denied" errors in Unix-like environments since users will be in an unprivileged state much more often. Programs are therefore generally designed to use the
HOME environment variable to store settings (e.g.
$HOME/.w3m for the w3m browser). The dynamic linker provides an environment variable
LD_LIBRARY_PATH that programs can use to load libraries from non-standard directories. Assuming
/mnt contains the portable programs and configuration, a command line may look like:
HOME=/mnt/home/user LD_LIBRARY_PATH=/mnt/usr/lib /mnt/usr/bin/w3m www.example.com
A Linux application without need for a user-interaction (e.g. adapting a script or environment variable) on varying directory paths can be achieved with the GCC Linker option $ORIGIN which allows a relative library search path.
Not all programs honor this--some completely ignore $HOME and instead do a user look-up in
/etc/passwd to find the home directory, therefore thwarting portability.
There are also cross-distro package formats that do not require admin rights to run, like Autopackage, klik (now called AppImage), or CDE, but which gained only limited acceptance and support in the Linux community in the 2000s. Around 2015 the idea of portable and distro independent packing for the Linux ecosystem got more traction when Linus Torvalds discussed this topic on the DebConf 2014 and endorsed later AppImage for his dive log application Subsurface. For instance, MuseScore and Krita followed in 2016 and started to use AppImage builds for software deployment. RedHat released in 2016 the Flatpak system, which is an successor of Alexander Larsson's glick project which was inspired by klik (now AppImage). Similarly, Canonical released in 2016 Snap packages for Ubuntu and many other Linux distros.
Many Mac applications that can be installed by drag-and-drop are inherently portable as Mac application bundles. Examples include Mozilla Firefox, Skype and Google Chrome which do not require admin access and do not need to be placed into a central, restricted area. Applications placed into
~/Applications) are registered with macOS LaunchServices in the same way as applications placed into the main
/Applications folder. For example, right-clicking a file in Finder and then selecting "Open With..." will show applications available from both /Applications and ~/Applications. Developers can create Mac product installers which allow the user to perform a home directory install, labelled "Install for me only" in the Installer user interface. Such an installation is performed as the user.
You can use a special keyword $ORIGIN to say 'relative to the actual location of the executable'. Suddenly we found we could use -rpath $ORIGIN/lib and it worked. The game was loading the correct libraries, and so was stable and portable, but was also now completely in the spirit of the LGPL as well as the letter!
The Linux community, in their infinite wisdom, proceeds to flame the hell out of CDE. [...] "We should all just be using package management." Here is what I want to say, and let my words be carried down from the mountaintops, written on tiny stone tablets: Package management is not a universal panacea.
If Hearn is correct, the real lesson of Autopackage is not how to improve software installation, but the difficulty -- perhaps the impossibility -- of large-scale changes in Linux architecture this late in its history. It's a sobering, disappointing conclusion to a project that once seemed so promising.
I have seen this first hand with the other project I'm involved with, which is my dive log app. We make binaries for Windows and OSX, we basically don't make binaries for Linux. Why? Because making binaries for Linux desktop applications is a major fucking pain in the ass.CS1 maint: uses authors parameter (link)
I finally got around to play with the "AppImage" version of +Subsurface, and it really does seem to "just work".
I, as the app maintainer, don't want my app bundled in a distribution anymore. Way to much pain for absolutely zero gain. Whenever I get a bug report my first question is "oh, which version of which distribution? which version of which library? What set of insane patches were applied to those libraries?". No, Windows and Mac get this right. I control the libraries my app runs against. [...] With an AppImage I can give them just that. Something that runs on their computer.