Fix scan_instructions_internal when reading 2-bytes instruction

Author: macabeus

objdiff-core/src/arch/arm.rs

@@ -224,7 +224,9 @@ impl Arch for ArchArm {
             }
 
             // Check how many bytes we can/should read
-            let num_code_bytes = if data.len() >= 4 {
+            let bytes_until_next_mapping =
+                next_mapping.map(|m| (m.address - address) as usize).unwrap_or(usize::MAX);
+            let num_code_bytes = if data.len() >= 4 && bytes_until_next_mapping >= 4 {
                 if mode == unarm::ParseMode::Data && address & 3 != 0 {
                     // 32-bit .word value should be aligned on a 4-byte boundary, otherwise use .hword
                     2

objdiff-core/tests/arch_arm.rs

@@ -57,6 +57,31 @@ fn combine_text_sections() {
     insta::assert_snapshot!(output);
 }
 
+#[test]
+#[cfg(feature = "arm")]
+fn thumb_short_data_mapping() {
+    // When a .2byte directive is used in Thumb code, the assembler emits
+    // $d/$t mapping symbols for a 2-byte data region. The disassembler must
+    // not read 4 bytes as a .word when the next mapping symbol limits the
+    // data region to 2 bytes.
+    let diff_config = diff::DiffObjConfig::default();
+    let obj = obj::read::parse(
+        include_object!("data/arm/code_1_vblank.o"),
+        &diff_config,
+        diff::DiffSide::Base,
+    )
+    .unwrap();
+    let symbol_idx = obj.symbols.iter().position(|s| s.name == "VBlankDMA_Level1").unwrap();
+    let diff = diff::code::no_diff_code(&obj, symbol_idx, &diff_config).unwrap();
+    let output = common::display_diff(&obj, &diff, symbol_idx, &diff_config);
+    // .2byte data followed by Thumb code must not be merged into a 4-byte .word
+    assert!(
+        !output.contains(".word"),
+        "2-byte data regions should not be decoded as 4-byte .word values.\n\
+         The disassembler must respect mapping symbol boundaries."
+    );
+}
+
 #[test]
 #[cfg(feature = "arm")]
 fn trim_trailing_hword() {