helper.exe contains the uninstaller, plus a routine that’s run by the application updater after it applies an update, as well as a few utilities used for default browser handling and shortcut maintenance. It mainly consists of two files, uninstaller.nsi, which is the main script and contains the entry point and the uninstall logic, and shared.nsh, which contains most of the logic for the other functions.
The uninstaller may be the most straightforward of the installer components. The only complexity comes from a need to avoid accidentally removing any user files that may have found their way into the installation directory; the general philosophy of the uninstaller is to remove everything that the installer creates, and nothing that it doesn’t.
Any registry entries the installer would have created are removed, even ones only used by very old installer versions.
All the files that the uninstaller knows were created by the installer or are owned by the application are deleted, or flagged for deletion on reboot if any are in use. There are a few hard-coded directories that are known to be safe to delete (for example, the distribution directory, and any temporary directories created by the updater). For a list of application files that are safe to uninstall, we read a file from the application directory called
precomplete. This file is mainly used to tell the updater what it should do to clean out the directory when applying a complete update (one that replaces all application files), but that means it contains a handy auto-generated list of all application files, so it can be reused for uninstallation.
If the application directory is empty after that, then it is removed, but it’s left alone if any files are still present.
If the copy of Firefox that was just uninstalled is the only one that was using the maintenance service, the maintenance service uninstaller is also run.
Any BITS jobs from this installation of Firefox are cancelled.
The first “uninstall” ping found is uploaded, and all uninstall pings are deleted (for this installation).
Note that profiles and any other user-generated files (e.g., crash reports) are specifically not uninstalled.
At the end of an application update cycle, after the new files are in place, the updater invokes the helper with the
/PostUpdate command-line switch. The PostUpdate function fills a grab bag of responsibilities which are all focused around maintaining system integration objects created by the installer. For example, a number of registry entries contain the version number, so that has to be changed on updates. If the branding name of the application changes and shortcuts have to be renamed, that’s done here as well. For one counterexample, changing icons does not require any code in PostUpdate, or anywhere else; new icons are automatically picked up by the shell. The PostUpdate function also keeps the maintenance service up to date.
It’s important to remember that PostUpdate is, indeed, post-update. It doesn’t run until after its own code has already been updated. This makes it really the only phase of the update process where changes can go into affect immediately in the first build that contains a patch, instead of having to wait for the next update after that. This makes it a good place to put anything that needs to be done before the new version of the application can run; this includes things like registering DLL’s, which the installer also handles, but that PostUpdate has to take care of for existing installations.
PostUpdate actually runs two times after each update: once in an elevated context, and once as a regular user who performed the update. Both runs do similar work - the main difference being that we update HKLM registry entries are updated by the elevated instance, while HKCU ones are updated by the non-elevated instance.
It’s also important to remember that PostUpdate, being part of the installer code, only exists on Windows, so it can’t be used to fix things up on other platforms the same way.
Default Browser and Shortcut Handling¶
Windows versions older than 10 contain a control panel called Set Program Access and Defaults, or SPAD. As the name suggests, this was the UI for setting default programs for classes of activities (“web browser” or “e-mail client” for example), as the Windows 10 default program settings page is, but it also controls program “access,” which is typically defined as whether or not shortcuts for the program exist. To support this interface, an application has to register a set of commands that the interface can invoke to hide or show the shortcuts, and to have the application make itself the default. We implement these actions in helper.exe; they’re triggered by invoking it with the command-line switches
The helper also implements the
/SetAsDefaultAppUser switch, which is invoked by the “Make Default” button in the Firefox preferences UI.
On Windows 10 neither SetAsDefaultAppUser nor SetAsDefaultAppGlobal is effective because the default programs settings can only be modified by the Windows settings app. However they do still write the registry entries that are needed to get us an entry in the system default browser menu, should those entries not already exist (the installer always creates them, but running Firefox without having run the installer is supported). ShowShortcuts and HideShortcuts are never called on Windows 10 because the SPAD control panel no longer exists.