Joy of Setup

Google today released (after a bit of a comic-book pre-release, presumably due to the Labor Day holiday in the US) Google Chrome, its long-rumored open-source browser. Plenty of people will talk (endlessly) about the implications of another browser and how well Google Chrome and Chromium (the open source project) do the job. Blah, blah. Whatever. What’s really interesting is a couple of choices Google made about deployment:

1. The Google Chrome download is a svelte 474K bootstrapper that downloads the setup bits. No offline installer is available (unless it’s well-hidden).
2. Google Chrome is a "composite" setup: The guts of the application are installed by a non-MSI self-extractor. However, Google Chrome includes Google Gears, the browser add-in/library that adds a bunch of functionality for making apps-in-the-browser more powerful. The Gears in Google Chrome is installed by an MSI package. And yes, it’s built with WiX.
3. The Google Updater is no longer a LocalSystem service; instead, it starts at logon from the HKEY_CURRENT_USER\Software\Microsoft\Windows\CurrentVersion\Run registry key.
4. Last and absolutely not least: Google Chrome is a per-user application. It even installs in the per-user LocalAppDataFolder. (The included Google Gears is marked as "UAC compliant.")

That Google Chrome is a per-user app is amazing. Even with UAC on Windows Vista and Windows Server 2008, it’s so easy to say that "everyone’s used to needing admin privileges to install." That Google took the extra effort to limit themselves to the capabilities of a per-user app says a lot about their desire to have:

• a low-impact setup
• and absolutely no barriers to entry.

I wonder if it’s the start of a trend…

MSI has some funky syntax to support checking the action state (what MSI is planning to do during the current install transaction) and install state (what MSI already did, in a previous install transaction) of both features and components. See Conditional Statement Syntax for the details and Examples of Conditional Statement Syntax for some examples. (As an aside, would it have been too much to ask to avoid the Perl-like prefix characters and have something function-like such as FeatureActionState(featureId)?)

A couple of people on the wix-users mailing list have asked about using the feature action state in component conditions. There’s good news and bad news.

Good news for simple cases

The good news is that in simple cases with simple conditions, there’s probably another way to get the behavior you’re looking for. For example, one component can be assigned to multiple features, so if you’re trying to install a component when any of several features are being installed, just reference the component in all the features. MSI will install the component once, regardless of how many features reference it.

The bad news

If you want to conditionalize a component based on whether a particular feature is being installed (i.e., its action state), you can’t rely solely on the component’s condition. The Component Table doc has this to say:

The Condition field enables or disables a component only during the CostFinalize action. To enable or disable a component after CostFinalize, you must use a custom action or the DoAction ControlEvent to call MsiSetComponentState.

The CostFinalize action doc says:

The CostFinalize action queries the Condition table to determine which features are scheduled to be installed. Costing is done for each component in the Component table.

Unfortunately, it doesn’t state explicitly that the order of the two sentences is the order of execution. But, even if it did, there’s this table in Conditional Statement Syntax:

Feature and Component State Values

The following table shows where it is valid to use the feature and component operator symbols.

Operator <state>	Where this syntax is valid
$component-action	In the Condition table, and in the sequence tables, after the CostFinalize action.
&feature-action	In the Condition table, and in the sequence tables, after the CostFinalize action.
!feature-state	In the Condition table, and in the sequence tables, after the CostFinalize action.
?component-state	In the Condition table, and in the sequence tables, after the CostFinalize action.

So there are two fairly explicit strikes against being able to use a feature action state in a component condition:

• Conditional Statement Syntax doesn’t list the Component table as a valid place to use the &feature-action syntax.
• It says the syntax is valid only after CostFinalize but according to Component Table, that’s too late to use the component condition to enable/disable the component.

The Component Table says to use a “custom action or the DoAction ControlEvent to call MsiSetComponentState.” I haven’t quite deciphered whether the doc is saying to use DoAction to call a custom action from the UI or whether there’s a special, secret way of calling MsiSetComponentState via DoAction. If it’s the latter, I’m not aware of it (hence “special, secret”).

Here be dragons

Having been a writer for many years, I try to be nice about doc that is correct, even if in retrospect it’s not as helpful as it could have been. So here’s me being nice. The following sentence from Component Table easily leads a trusting reader down the wrong path:

The Condition column in the Component table accepts conditional expressions containing references to the installed states of features and components.

So here we have a sentence that contradicts the table in Conditional Statement Syntax and it’s really easy to misread the “installed state[s]” as “action states.” At the very least, it deserves clarification.

A question that’s come up on the wix-users mailing list several times lately has been about how to use feature conditions with properties set in the UI. It doesn’t work as most people expect so I thought I’d dig a little deeper than I would in a bunch of replies on the mailing list.

The basic idea is that you have some kind of optional functionality — like a Web site or Visual Studio integration — that you want the user to be able to enable or disable from the installer UI. (I’m assuming here that the functionality is in reality just a set of discrete components organized into a feature.)

The easiest way to get this working is to use MSI’s built-in SelectionTree control — the normal, boring, not-quite-easy-to-use feature tree that you see in most installers. That’s the approach we used with the WiX installer; in addition to features for plain sets of files, there are features for Visual Studio, Votive, and MSBuild integration.

The thrust of this question, though, is generally about using some other UI, like checkboxes or groups of radio buttons, to offer a better user experience during setup. But if you replace the SelectionTree control, you need to also replace how it manages feature states for you. Feature conditions seem like a fairly cheap way of accomplishing that, assuming you use public (i.e., all UPPERCASE) and secure (Property/@Secure=”yes”) properties to ensure they’re always passed to the MSI server for the installation execution sequence.

But it doesn’t work and as is usual with MSI, reading the SDK doc and examining a verbose log tell us why — almost.

Let’s start with the doc on the Condition table, which is the table where feature conditions are stored. A promising blurb:

The Level may be set based on any conditional statement, such as a test for platform, operating system, or a particular property setting.

A nice feature of the MSI SDK doc is that each table lists the actions that refer to the table. For the Condition table, it says:

This table is referred to when the CostFinalize action is executed.

Follow that link and you get another couple of interesting blurbs:

The CostFinalize action must be executed before starting any user interface sequence which allows the user to view or modify Feature table selections or directories.

and

The CostFinalize action queries the Condition table to determine which features are scheduled to be installed.

So far, everything sounds like it will work. It doesn’t, though, so clearly the docs aren’t telling the whole story. Verbose logs are our next step. Starting with MSI 3.1, they include entries when properties are added, changed, or deleted, so you can verify that the properties are being set correctly during the UI sequence. You can also verify that MSI is passing those properties to the execution sequence. Search the log for “Switching to server” to see the list of properties being passed. For example, from a complete install of the Windows Vista SDK version of Orca:

MSI (c) (6C:24) [14:54:54:015]: Switching to server: CUBDIR=”C:\Program Files (x86)\Orca\” ORCADIRECTORY=”C:\Program Files (x86)\Orca\” TARGETDIR=”V:\” MS.51D569E0_8A28_11D2_B962_006097C4DE24=”C:\WINNT\SysWOW64\” MS.51D569E2_8A28_11D2_B962_006097C4DE24=”C:\WINNT\SysWOW64\” MS.7EBEDD6A_AA66_11D2_B980_006097C4DE24=”C:\WINNT\SysWOW64\” MS.7EBEDD3E_AA66_11D2_B980_006097C4DE24=”C:\WINNT\SysWOW64\” INSTALLLEVEL=”1000″ COMPANYNAME=”MS” USERNAME=”Bob Arnson” CURRENTDIRECTORY=”X:\” CLIENTUILEVEL=”0″ CLIENTPROCESSID=”3180″ SOURCEDIR=”X:\” ACTION=”INSTALL” EXECUTEACTION=”INSTALL” ROOTDRIVE=”V:\” SECONDSEQUENCE=”1″ ADDLOCAL=OrcaHelp,Orca,EvalComServer,MergeModServer,CUBFiles,FullCUBFile,LogoCUBFile,XPLogoCUBFile,MMCUBFile

Notice the ADDLOCAL property setting. It’s a comma-delimited list of feature names that are to be installed. I didn’t pick them individually; instead, I used the “complete” button to say I wanted everything. The orca.msi package publishes a SetInstallLevel control event with a high install level to let MSI decide to install all features rather than using an AddLocal control event to manually list them.

Let’s assume that MSI uses the ADDLOCAL property and its friends the REMOVE and ADDSOURCE properties (among other friends) to communicate to the MSI server the feature choices the user made during the UI sequence. If you think about the many different ways the UI can implicitly and explicitly set which features get installed — from “typical” and “complete” buttons publishing SetInstallLevel control events and SelectionTree controls letting users pick and choose individual features — it makes sense that MSI would need a rich way of controlling feature installation and it makes sense for MSI to use its own properties mechanism rather than invent a new one. (And note that the same thing applies to uninstallation and reinstallation — those are all handled on a per-feature basis via properties.)

OK, so what does that have to do with feature conditions? (I’m getting there, really.) All the feature-selection property topics have this blurb:

The installer sets the Preselected Property to a value of “1″ during the resumption of a suspended installation, or when any of the above properties are specified on the command line.

Follow the Preselected link and you get this blurb:

The Preselected property indicates that features have already been selected and that the selection dialog need not be shown.

Now it’s time to enter the murky world of speculation. There’s no doc to indicate that the feature-selection properties or the Preselected property have any affect on feature-condition evaluation in CostFinalize. But I think it’s reasonable to infer that an explicit feature selection (via ADDLOCAL, for example) would override the more implicit feature-selection options available, including the Condition table and INSTALLLEVEL property.

And, to make a long story short — too late! — Carolyn, MSI Team Dev Lead, confirmed that’s the case. Feature conditions are evaluated only if no feature-selection properties are set. And, as the UI sequence converts SetInstallLevel control events, feature-selection control events, and SelectionTree control settings into the corresponding feature-selection properties, feature conditions that include properties set during the UI sequence won’t work as expected.

Note that an installation that doesn’t run the UI sequence (e.g., using the /qb command-line switch to run a basic-UI installation) can use properties set during the execute sequence in feature conditions. But the same caveat applies: If any feature-selection properties are set, feature conditions aren’t evaluated.

OK, fine, what now?

So how do you get this working? As I said near the beginning of this much-longer-than-expected tome, “if you replace the SelectionTree control, you need to also replace how it manages feature states for you.” As SelectionTree controls end up setting the feature-selection properties, you need to do the same thing, directly or indirectly. The easiest way is to publish feature-selection control events from your feature-selection dialog.

For example, the WixUI dialog set WixUI_Mondo uses SetInstallLevel published from SetupTypeDlg:

<Control Id=”CompleteButton” Type=”PushButton” X=”40″ Y=”171″ Width=”80″ Height=”17″ ToolTip=”!(loc.SetupTypeDlgCompleteButtonTooltip)” Text=”!(loc.SetupTypeDlgCompleteButton)”>
<Publish Property=”WixUI_InstallMode” Value=”InstallComplete”>1</Publish>
<Publish Event=”SetInstallLevel” Value=”1000″>1</Publish>
</Control>

If you wanted to explicitly list features, the easiest way to do so is to use an AddLocal control event with an argument of ALL to install all features locally, then use individual Remove control events to remove the features that don’t apply. Doing so covers lets the user both install and uninstall features. For example:

<Control Id=”Next” Type=”PushButton” X=”235″ Y=”243″ Width=”57″ Height=”18″ Default=”yes” Text=”&Next >”>
<Publish Event=”AddLocal” Value=”ALL”>1</Publish>
<Publish Event=”Remove” Value=”FeatureX”>NOT FEATUREX_CHECKBOX</Publish>
<Publish Event=”Remove” Value=”FeatureY”>NOT FEATUREY_CHECKBOX</Publish>
<Publish Event=”Remove” Value=”FeatureZ”>NOT FEATUREZ_CHECKBOX</Publish>
</Control>

Hope this helps understand why things aren’t always as simple as a first glance might indicate.