Backport #1032 to support/3.2: respect auto_select in unattended installations #1089

Summary

The conditional module selection bug fixed in #1032 (N°9984, commit 9c6baae) still affects support/3.2.

The fix is already present in develop, but the current support/3.2 branch still selects conditional modules as part of unpackaged extensions before evaluating their auto_select expressions.

Checked against Combodo/iTop branch support/3.2 at da4a81053411314812f5db055391a8e8edc99ed3 on 2026-10-22. The checkout declares iTop core version 3.2.3. Local unit tests use PHP 8.3.6 and PHPUnit 9.6.34; these are test-environment versions, not a claim about an affected deployed installation.

Reproduction at the service level

1. Provide an eligible, visible unpackaged extension containing a regular module and conditional modules with auto_select expressions false and true.
2. Leave the extension unspecified in the installation choices.
3. Run ProcessDefaultModules(), followed by ProcessExtensionModulesNotSpecifiedInChoices() and ProcessAutoSelectModules(), as in the unattended-install flow.
4. The extension-selection step selects both conditional modules prematurely. The later auto-selection step does not remove the module whose expression evaluates to false.

Expected: the regular module is selected by extension processing; conditional modules are selected only when their auto_select expressions evaluate to true.

Actual on the unpatched 3.2 code: the module with the false expression is also selected.

This reproduction is covered by a PHPUnit regression test. An end-to-end reproduction on an installed iTop instance has not yet been performed.

Proposed backport

Skip modules listed in aAutoSelectModules during ProcessExtensionModulesNotSpecifiedInChoices(), leaving selection to ProcessAutoSelectModules(). This is functionally equivalent to #1032, with an explicit guard and explanatory comment.

The prepared change also adds a regression test covering regular modules and both boolean outcomes. The affected test class passes locally: 31 tests, 106 assertions. Without the guard, the new test fails because both conditional modules are selected prematurely.

A companion PR targeting support/3.2 provides the backport and regression test.