Windows Installer

A single, installed, working program (or set of programs) is a product. A product is identified by a unique GUID (the ProductCode property) providing an authoritative identity throughout the world. The GUID, in combination with the version number (ProductVersion property), allows for release management of the product's files and registry keys.

A package includes the package logic and other metadata that relates to how the package executes when running. For example, changing an EXE file in the product may require the ProductCode or ProductVersion to be changed for the release management. However, merely changing or adding a launch condition (with the product remaining exactly the same as the previous version) would still require the PackageCode to change for release management of the MSI file itself.

A feature is a hierarchical group of components. A feature may contain any number of components and other sub-features. Smaller packages can consist of a single feature. More complex installers may display a "custom setup" dialog box, from which the user can select which features to install or remove.

The package author defines the product features. A word processor, for example, might place the program's core file into one feature, and the program's help files, optional spelling checker and stationery modules into additional features.

A component is the basic unit of a product. Each component is treated by Windows Installer as a unit. The installer cannot install just part of a component.[7] Components can contain program files, folders, COM components, registry keys, and shortcuts. The user does not directly interact with components.

Components are identified globally by GUIDs; thus the same component can be shared among several features of the same package or multiple packages, ideally through the use of Merge Modules.

A key path is a specific file, registry key, or ODBC data source that the package author specifies as critical for a given component. Because a file is the most common type of key path, the term key file is commonly used. A component can contain at most one key path; if a component has no explicit key path, the component's destination folder is taken to be the key path. When an MSI-based program is launched, Windows Installer checks the existence of key paths. If there is a mismatch between the current system state and the value specified in the MSI package (e.g., a key file is missing), the related feature is re-installed. This process is known as self-healing or self-repair. No two components should use the same key path.

The user interface phase typically queries the target system, displays an installation wizard and enables the user to change various options that will affect the installation.

However, the user interface sequence should not make any changes to the system, for the following reasons:

1. A user may install or uninstall an MSI package in passive mode or silent mode, bypassing this phase entirely. (Passive mode bypasses the user interface phase but shows a graphical progress bar. Silent mode shows nothing.) During installation, information gathered in this phase may be supplied beforehand through the command-line interface.
2. The user interface sequence runs with user privileges, and not with the elevated privileges required during installation.

When the user clicks the "Install" button in a typical MSI installation wizard, installation proceeds to the Execute phase, in which software components are actually installed. The Execute phase makes system changes, but it does not display any user interface elements.

The Execute phase happens in two steps:

• Immediate mode. In this phase, Windows Installer receives instructions, either from a user or an application, to install or uninstall features of a product. The requests cause the execution of sequences of actions, which query the installation database to build an internal script describing the execution phase in detail.
• Deferred mode. In this phase, the script built in immediate mode is executed in the context of the privileged Windows Installer service. The script must be executed by a privileged account because of the heterogeneity of the scenarios in which a setup operation is initiated. For example, elevated privileges are necessary to serve on-demand installation requests from non-privileged users. (To run with elevated privileges, however, the package must be deployed by a local administrator or advertised by a system administrator using Group Policy.)

All installation operations are transactional.[8] In other words, for each operation that Windows Installer performs, it generates an equivalent undo operation that would revert the change made to the system. In case any script action fails during deferred execution, or the operation is cancelled by the user, all the actions performed until that point are rolled back, restoring the system to its original state. Standard Windows Installer actions automatically write information into a rollback script; package authors who create custom actions that change the target system should also create corresponding rollback actions (as well as uninstall actions and uninstallation-rollback actions). As a design feature, if applied correctly this mechanism will also roll back a failed uninstall of an application to a good working state.

Windows Installer can advertise a product rather than actually install it.[9] The product will appear installed to the user, but it will not actually be installed until it is run for the first time by triggering an entry point (by means of a Start menu shortcut, by opening a document that the product is configured to handle, or by invoking an advertised COM class). A package can be advertised by an administrator using Group Policy or other deployment mechanism, or by running the msiexec executable with the /jm (for per-machine advertisement) or /ju (for per-user advertisement) switch. Some MSI packages authored in InstallShield may prevent the use of these and other native MSI features.

The user must have administrator privileges to complete the advertised installation.

Similar to advertising, it installs a feature as soon as the user tries to use it.[10]

An administrative installation creates an uncompressed source image for a product, typically to be used for installing or running an application from a network location.[11] An administrative installation is not a typical installation, in that it does not create any shortcuts, register COM servers, create an Add or Remove Programs entry, and so on. Often an administrative installation enables a user to install the product in such a way that its features run from the uncompressed installation source.

Administrative installations are also useful when creating a Windows Installer patch, as this requires uncompressed images of the earlier and current versions of a product in order to compute binary file differences. An administrative installation is performed by running the msiexec executable with the /a switch.

The developer of an installer package may write code to serve their own purpose, delivered as a DLL, EXE, VBScript or JavaScript. This can be executed during the installation sequences, including when the user clicks a button in the user interface, or during the InstallExecuteSequence. Custom Actions typically validate product license keys, or initialize more complex services. Developers should normally provide inverse custom actions for use during uninstall.

Msiexec provides a way to break after loading a specified custom action DLL but before invoking the action.[12]

A Windows Installer package may contain another package to be installed at the same time. These are ideally provided as a .msm file component, but may also be a separate executable program which will be unpacked from the installer package during the InstallExecuteSequence and can be run immediately. The file can then optionally be deleted before the end of the InstallExecuteSequence, and so is ideal for using with older installers. .msm file component is called Microsoft Merge Module and is utilized to single out and integrate common components. It contains images of all tables, which have relevant information regarding the common components. Ready-made merge modules are supplied with Visual Studio or can be downloaded from the alternative sources.[13]

Windows Installer 4.0, which was shipped with Windows Vista, incorporates capabilities to take advantage of the User Account Control feature. If an application can install without elevated privileges, its MSI package can be marked as such, thus allowing install without prompting the user for Administrator credentials. Windows Installer also works in conjunction with the Restart Manager; when installing or updating an application or system component with "full" user interface mode, the user will be displayed a list of affected applications that can be shut down, and then restarted after files have been updated. Installer actions running in silent mode perform these application restarts automatically. System services and tray applications can also be restarted in this manner.

Microsoft provides a set of Internal Consistency Evaluators (ICE) that can be used to detect potential problems with an MSI database.[16] The ICE rules are combined into CUB files, which are stripped-down MSI files containing custom actions that test the target MSI database's contents for validation warnings and errors. ICE validation can be performed with the Platform SDK tools Orca and msival2, or with validation tools that ship with the various authoring environments.

For example, some of the ICE rules are:

• ICE09: Validates that any component destined for the System folder is marked as being permanent.
• ICE24: Validates that the product code, product version, and product language have appropriate formats.
• ICE33: Validates that the Registry table is not used for data better suited for another table (Class, Extension, Verb, and so on).

Addressing ICE validation warnings and errors is an important step in the release process.