[scottanderson]

Problem

A primitive declared in a match arm reads its value from the main section. This occurs when its struct is placed in a custom section with @ ... in section. Members in an if body, in a try/catch body, and structs in a match arm read from the correct section.

struct S {
    u32 plain;
    match (plain) { (_): u32 inMatch; }
    if (true) { u32 inIf; }
};
S s @ 0 in section;

Here s.plain and s.inIf read from section, but s.inMatch reads from the main section at the same offset. The value is wrong during evaluation (for example std::print) and afterwards (for example the JSON formatter).

This breaks palworld.hexpat in ImHex-Patterns. It inflates a save into a section and parses it with gvas.hexpat. Each IntProperty, FloatProperty and similar property there is a primitive in a match arm.

Summary

ASTNodeConditionalStatement::createPatterns and ASTNodeTryCatchStatement::createPatterns set the current section on each new pattern before they push it into the scope. ASTNodeMatchStatement::createPatterns did not. This PR adds the same line:

 for (auto &pattern : newPatterns) {
+    pattern->setSection(evaluator->getSectionId());
     scope.push_back(std::move(pattern));
 }

No other statement node has this gap. Compound statements and multi-variable declarations do not push into the scope. They return their patterns to the enclosing struct or union, which sets the section.

The new SectionStatements test places a struct in a section whose bytes differ from the main data. It asserts that each of these reads from the section: a plain member, a primitive in a match arm, an if body, a try body, a catch body, a nested struct, and a struct in a match arm.

Evidence

• Without the fix, SectionStatements fails with assertion failed "Match arm member did not read from section". All other tests pass.
• With the fix, all 80 tests pass.

Merge Danger

Door: two-way

A primitive in a match arm in a custom section now reads from that section. Reverting gives back the old behavior.

Blast Radius: small

Only patterns that place a struct in a custom section and declare a primitive directly in a match arm see a change. Those patterns read wrong values today.