Version 2.5.0

* Fix missing dependency feature for objdiff-gui

* Update .gitignore

* PPC: Display data values on hover for pools as well

* Tooltip data display: Format floats and doubles better

Floats and doubles will now always be displayed with a decimal point and one digit after it, even if they are whole numbers. Floats will also have the f suffix. This is so you can tell the data type just by glancing at the value.

* Move big functions to bottom ppc.rs

* Clear pool relocs in volatile registers on function call

This fixes some false positives.

* Revert ObjArch API changes, add fake target symbol hack

Because we no longer have access to the actual symbol name via sections, guess_data_type can no longer detect the String data type for pooled references.

* Add hack to detect strings via the addi opcode

* Move hack to resolve placeholder symbol into process_code_symbol

* Merge reloc and fake_pool_reloc fields of ObjIns

.gitignore

@@ -18,4 +18,4 @@ android.keystore
 *.frag
 *.vert
 *.metal
-.vscode/launch.json
+.vscode/

Cargo.toml

@@ -13,7 +13,7 @@ strip = "debuginfo"
 codegen-units = 1
 
 [workspace.package]
-version = "2.3.4"
+version = "2.5.0"
 authors = ["Luke Street <luke@street.dev>"]
 edition = "2021"
 license = "MIT OR Apache-2.0"

README.md

@@ -20,6 +20,7 @@ Supports:
 - MIPS (N64, PS1, PS2, PSP)
 - x86 (COFF only at the moment)
 - ARM (GBA, DS, 3DS)
+- ARM64 (Switch, experimental)
 
 See [Usage](#usage) for more information.
 

deny.toml

@@ -73,6 +73,7 @@ ignore = [
     #{ id = "RUSTSEC-0000-0000", reason = "you can specify a reason the advisory is ignored" },
     #"a-crate-that-is-yanked@0.1.1", # you can also ignore yanked crate versions if you wish
     #{ crate = "a-crate-that-is-yanked@0.1.1", reason = "you can specify why you are ignoring the yanked crate" },
+    { id = "RUSTSEC-2024-0384", reason = "Unmaintained indirect dependency" },
 ]
 # If this is true, then cargo deny will use the git executable to fetch advisory database.
 # If this is false, then it uses a built-in git library.
@@ -97,7 +98,7 @@ allow = [
     "BSL-1.0",
     "CC0-1.0",
     "MPL-2.0",
-    "Unicode-DFS-2016",
+    "Unicode-3.0",
     "Zlib",
     "0BSD",
     "OFL-1.1",

objdiff-cli/Cargo.toml

@@ -14,13 +14,13 @@ publish = false
 
 [dependencies]
 anyhow = "1.0"
-argp = "0.3"
+argp = "0.4"
 crossterm = "0.28"
 enable-ansi-support = "0.2"
 memmap2 = "0.9"
 objdiff-core = { path = "../objdiff-core", features = ["all"] }
 prost = "0.13"
-ratatui = "0.28"
+ratatui = "0.29"
 rayon = "1.10"
 serde = { version = "1.0", features = ["derive"] }
 serde_json = "1.0"

objdiff-cli/src/main.rs

@@ -1,6 +1,7 @@
 mod argp_version;
 mod cmd;
 mod util;
+mod views;
 
 // musl's allocator is very slow, so use mimalloc when targeting musl.
 // Otherwise, use the system allocator to avoid extra code size.

objdiff-cli/src/views/mod.rs

@@ -0,0 +1,25 @@
+use anyhow::Result;
+use crossterm::event::Event;
+use ratatui::Frame;
+
+use crate::cmd::diff::AppState;
+
+pub mod function_diff;
+
+#[derive(Default)]
+pub struct EventResult {
+    pub redraw: bool,
+    pub click_xy: Option<(u16, u16)>,
+}
+
+pub enum EventControlFlow {
+    Break,
+    Continue(EventResult),
+    Reload,
+}
+
+pub trait UiView {
+    fn draw(&mut self, state: &AppState, f: &mut Frame, result: &mut EventResult);
+    fn handle_event(&mut self, state: &mut AppState, event: Event) -> EventControlFlow;
+    fn reload(&mut self, state: &AppState) -> Result<()>;
+}

objdiff-core/Cargo.toml

@@ -16,15 +16,17 @@ documentation = "https://docs.rs/objdiff-core"
 crate-type = ["cdylib", "rlib"]
 
 [features]
-all = ["config", "dwarf", "mips", "ppc", "x86", "arm", "bindings"]
+all = ["config", "dwarf", "mips", "ppc", "x86", "arm", "arm64", "bindings", "build"]
 any-arch = ["config", "dep:bimap", "dep:strum", "dep:similar", "dep:flagset", "dep:log", "dep:memmap2", "dep:byteorder", "dep:num-traits"] # Implicit, used to check if any arch is enabled
+bindings = ["dep:serde_json", "dep:prost", "dep:pbjson", "dep:serde", "dep:prost-build", "dep:pbjson-build"]
+build = ["dep:shell-escape", "dep:path-slash", "dep:winapi", "dep:notify", "dep:notify-debouncer-full", "dep:reqwest", "dep:self_update", "dep:tempfile", "dep:time"]
 config = ["dep:bimap", "dep:globset", "dep:semver", "dep:serde_json", "dep:serde_yaml", "dep:serde", "dep:filetime"]
 dwarf = ["dep:gimli"]
 mips = ["any-arch", "dep:rabbitizer"]
 ppc = ["any-arch", "dep:cwdemangle", "dep:cwextab", "dep:ppc750cl"]
 x86 = ["any-arch", "dep:cpp_demangle", "dep:iced-x86", "dep:msvc-demangler"]
 arm = ["any-arch", "dep:cpp_demangle", "dep:unarm", "dep:arm-attr"]
-bindings = ["dep:serde_json", "dep:prost", "dep:pbjson", "dep:serde", "dep:prost-build", "dep:pbjson-build"]
+arm64 = ["any-arch", "dep:cpp_demangle", "dep:yaxpeax-arch", "dep:yaxpeax-arm"]
 wasm = ["bindings", "any-arch", "dep:console_error_panic_hook", "dep:console_log", "dep:wasm-bindgen", "dep:tsify-next", "dep:log"]
 
 [package.metadata.docs.rs]
@@ -76,6 +78,31 @@ msvc-demangler = { version = "0.10", optional = true }
 unarm = { version = "1.6", optional = true }
 arm-attr = { version = "0.1", optional = true }
 
+# arm64
+yaxpeax-arch = { version = "0.3", default-features = false, features = ["std"], optional = true }
+yaxpeax-arm = { version = "0.3", default-features = false, features = ["std"], optional = true }
+
+# build
+notify = { git = "https://github.com/notify-rs/notify", rev = "128bf6230c03d39dbb7f301ff7b20e594e34c3a2", optional = true }
+notify-debouncer-full = { git = "https://github.com/notify-rs/notify", rev = "128bf6230c03d39dbb7f301ff7b20e594e34c3a2", optional = true }
+shell-escape = { version = "0.1", optional = true }
+tempfile = { version = "3.14", optional = true }
+time = { version = "0.3", optional = true }
+
+[target.'cfg(windows)'.dependencies]
+path-slash = { version = "0.2", optional = true }
+winapi = { version = "0.3", optional = true }
+
+# For Linux static binaries, use rustls
+[target.'cfg(target_os = "linux")'.dependencies]
+reqwest = { version = "0.12", default-features = false, features = ["blocking", "json", "multipart", "rustls-tls"], optional = true }
+self_update = { version = "0.41", default-features = false, features = ["rustls"], optional = true }
+
+# For all other platforms, use native TLS
+[target.'cfg(not(target_os = "linux"))'.dependencies]
+reqwest = { version = "0.12", default-features = false, features = ["blocking", "json", "multipart", "default-tls"], optional = true }
+self_update = { version = "0.41", optional = true }
+
 [build-dependencies]
 prost-build = { version = "0.13", optional = true }
 pbjson-build = { version = "0.7", optional = true }

objdiff-core/README.md

@@ -11,4 +11,5 @@ objdiff-core contains the core functionality of [objdiff](https://github.com/enc
 - **`ppc`**: Enables the PowerPC backend powered by [ppc750cl](https://github.com/encounter/ppc750cl).
 - **`x86`**: Enables the x86 backend powered by [iced-x86](https://crates.io/crates/iced-x86).
 - **`arm`**: Enables the ARM backend powered by [unarm](https://github.com/AetiasHax/unarm).
+- **`arm64`**: Enables the ARM64 backend powered by [yaxpeax-arm](https://github.com/iximeow/yaxpeax-arm).
 - **`bindings`**: Enables serialization and deserialization of objdiff data structures.

objdiff-core/src/arch/arm.rs

@@ -124,11 +124,9 @@ impl ObjArch for ObjArchArm {
             .get(&SectionIndex(section_index))
             .map(|x| x.as_slice())
             .unwrap_or(&fallback_mappings);
-        let first_mapping_idx =
-            match mapping_symbols.binary_search_by_key(&start_addr, |x| x.address) {
-                Ok(idx) => idx,
-                Err(idx) => idx - 1,
-            };
+        let first_mapping_idx = mapping_symbols
+            .binary_search_by_key(&start_addr, |x| x.address)
+            .unwrap_or_else(|idx| idx - 1);
         let first_mapping = mapping_symbols[first_mapping_idx].mapping;
 
         let mut mappings_iter =
@@ -215,7 +213,7 @@ impl ObjArch for ObjArchArm {
                 address: address as u64,
                 size: (parser.address - address) as u8,
                 op: ins.opcode_id(),
-                mnemonic: parsed_ins.mnemonic.to_string(),
+                mnemonic: Cow::Borrowed(parsed_ins.mnemonic),
                 args,
                 reloc,
                 branch_dest,
@@ -234,7 +232,7 @@ impl ObjArch for ObjArchArm {
         section: &ObjSection,
         address: u64,
         reloc: &Relocation,
-    ) -> anyhow::Result<i64> {
+    ) -> Result<i64> {
         let address = address as usize;
         Ok(match reloc.flags() {
             // ARM calls

objdiff-core/src/arch/mips.rs

@@ -119,7 +119,7 @@ impl ObjArch for ObjArchMips {
             let op = instruction.unique_id as u16;
             ops.push(op);
 
-            let mnemonic = instruction.opcode_name().to_string();
+            let mnemonic = instruction.opcode_name();
             let is_branch = instruction.is_branch();
             let branch_offset = instruction.branch_offset();
             let mut branch_dest = if is_branch {
@@ -202,7 +202,7 @@ impl ObjArch for ObjArchMips {
                 address: cur_addr as u64,
                 size: 4,
                 op,
-                mnemonic,
+                mnemonic: Cow::Borrowed(mnemonic),
                 args,
                 reloc: reloc.cloned(),
                 branch_dest,

objdiff-core/src/arch/mod.rs

@@ -12,6 +12,8 @@ use crate::{
 
 #[cfg(feature = "arm")]
 mod arm;
+#[cfg(feature = "arm64")]
+mod arm64;
 #[cfg(feature = "mips")]
 pub mod mips;
 #[cfg(feature = "ppc")]
@@ -34,13 +36,22 @@ pub enum DataType {
 
 impl DataType {
     pub fn display_bytes<Endian: ByteOrder>(&self, bytes: &[u8]) -> Option<String> {
-        // TODO: Attempt to interpret large symbols as arrays of a smaller type,
-        // fallback to intrepreting it as bytes.
-        // https://github.com/encounter/objdiff/issues/124
-        if self.required_len().is_some_and(|l| bytes.len() != l) {
-            log::warn!("Failed to display a symbol value for a symbol whose size doesn't match the instruction referencing it.");
+        if self.required_len().is_some_and(|l| bytes.len() < l) {
+            log::warn!("Failed to display a symbol value for a symbol whose size is too small for instruction referencing it.");
             return None;
         }
+        let mut bytes = bytes;
+        if self.required_len().is_some_and(|l| bytes.len() > l) {
+            // If the symbol's size is larger a single instance of this data type, we take just the
+            // bytes necessary for one of them in order to display the first element of the array.
+            bytes = &bytes[0..self.required_len().unwrap()];
+            // TODO: Attempt to interpret large symbols as arrays of a smaller type and show all
+            // elements of the array instead. https://github.com/encounter/objdiff/issues/124
+            // However, note that the stride of an array can not always be determined just by the
+            // data type guessed by the single instruction accessing it. There can also be arrays of
+            // structs that contain multiple elements of different types, so if other elements after
+            // the first one were to be displayed in this manner, they may be inaccurate.
+        }
 
         match self {
             DataType::Int8 => {
@@ -84,10 +95,10 @@ impl DataType {
                 }
             }
             DataType::Float => {
-                format!("Float: {}", Endian::read_f32(bytes))
+                format!("Float: {:?}f", Endian::read_f32(bytes))
             }
             DataType::Double => {
-                format!("Double: {}", Endian::read_f64(bytes))
+                format!("Double: {:?}", Endian::read_f64(bytes))
             }
             DataType::Bytes => {
                 format!("Bytes: {:#?}", bytes)
@@ -165,6 +176,8 @@ pub fn new_arch(object: &File) -> Result<Box<dyn ObjArch>> {
         Architecture::I386 | Architecture::X86_64 => Box::new(x86::ObjArchX86::new(object)?),
         #[cfg(feature = "arm")]
         Architecture::Arm => Box::new(arm::ObjArchArm::new(object)?),
+        #[cfg(feature = "arm64")]
+        Architecture::Aarch64 => Box::new(arm64::ObjArchArm64::new(object)?),
         arch => bail!("Unsupported architecture: {arch:?}"),
     })
 }

objdiff-core/src/arch/ppc.rs

@@ -1,4 +1,7 @@
-use std::{borrow::Cow, collections::BTreeMap};
+use std::{
+    borrow::Cow,
+    collections::{BTreeMap, HashMap},
+};
 
 use anyhow::{bail, ensure, Result};
 use byteorder::BigEndian;
@@ -7,7 +10,7 @@ use object::{
     elf, File, Object, ObjectSection, ObjectSymbol, Relocation, RelocationFlags, RelocationTarget,
     Symbol, SymbolKind,
 };
-use ppc750cl::{Argument, InsIter, Opcode, GPR};
+use ppc750cl::{Argument, InsIter, Opcode, ParsedIns, GPR};
 
 use crate::{
     arch::{DataType, ObjArch, ProcessCodeResult},
@@ -49,6 +52,8 @@ impl ObjArch for ObjArchPpc {
         let ins_count = code.len() / 4;
         let mut ops = Vec::<u16>::with_capacity(ins_count);
         let mut insts = Vec::<ObjIns>::with_capacity(ins_count);
+        let fake_pool_reloc_for_addr =
+            generate_fake_pool_reloc_for_addr_mapping(address, code, relocations);
         for (cur_addr, mut ins) in InsIter::new(code, address as u32) {
             let reloc = relocations.iter().find(|r| (r.address as u32 & !3) == cur_addr);
             if let Some(reloc) = reloc {
@@ -143,9 +148,9 @@ impl ObjArch for ObjArchPpc {
             insts.push(ObjIns {
                 address: cur_addr as u64,
                 size: 4,
-                mnemonic: simplified.mnemonic.to_string(),
+                mnemonic: Cow::Borrowed(simplified.mnemonic),
                 args,
-                reloc: reloc.cloned(),
+                reloc: reloc.or(fake_pool_reloc_for_addr.get(&cur_addr)).cloned(),
                 op: ins.op as u16,
                 branch_dest,
                 line,
@@ -173,6 +178,7 @@ impl ObjArch for ObjArchPpc {
     fn display_reloc(&self, flags: RelocationFlags) -> Cow<'static, str> {
         match flags {
             RelocationFlags::Elf { r_type } => match r_type {
+                elf::R_PPC_NONE => Cow::Borrowed("R_PPC_NONE"), // We use this for fake pool relocs
                 elf::R_PPC_ADDR16_LO => Cow::Borrowed("R_PPC_ADDR16_LO"),
                 elf::R_PPC_ADDR16_HI => Cow::Borrowed("R_PPC_ADDR16_HI"),
                 elf::R_PPC_ADDR16_HA => Cow::Borrowed("R_PPC_ADDR16_HA"),
@@ -188,26 +194,22 @@ impl ObjArch for ObjArchPpc {
     }
 
     fn guess_data_type(&self, instruction: &ObjIns) -> Option<super::DataType> {
-        // Always shows the first string of the table. Not ideal, but it's really hard to find
-        // the actual string being referenced.
         if instruction.reloc.as_ref().is_some_and(|r| r.target.name.starts_with("@stringBase")) {
             return Some(DataType::String);
         }
 
-        match Opcode::from(instruction.op as u8) {
-            Opcode::Lbz | Opcode::Lbzu | Opcode::Lbzux | Opcode::Lbzx => Some(DataType::Int8),
-            Opcode::Lhz | Opcode::Lhzu | Opcode::Lhzux | Opcode::Lhzx => Some(DataType::Int16),
-            Opcode::Lha | Opcode::Lhau | Opcode::Lhaux | Opcode::Lhax => Some(DataType::Int16),
-            Opcode::Lwz | Opcode::Lwzu | Opcode::Lwzux | Opcode::Lwzx => Some(DataType::Int32),
-            Opcode::Lfs | Opcode::Lfsu | Opcode::Lfsux | Opcode::Lfsx => Some(DataType::Float),
-            Opcode::Lfd | Opcode::Lfdu | Opcode::Lfdux | Opcode::Lfdx => Some(DataType::Double),
-
-            Opcode::Stb | Opcode::Stbu | Opcode::Stbux | Opcode::Stbx => Some(DataType::Int8),
-            Opcode::Sth | Opcode::Sthu | Opcode::Sthux | Opcode::Sthx => Some(DataType::Int16),
-            Opcode::Stw | Opcode::Stwu | Opcode::Stwux | Opcode::Stwx => Some(DataType::Int32),
-            Opcode::Stfs | Opcode::Stfsu | Opcode::Stfsux | Opcode::Stfsx => Some(DataType::Float),
-            Opcode::Stfd | Opcode::Stfdu | Opcode::Stfdux | Opcode::Stfdx => Some(DataType::Double),
-            _ => None,
+        let op = Opcode::from(instruction.op as u8);
+        if let Some(ty) = guess_data_type_from_load_store_inst_op(op) {
+            Some(ty)
+        } else if op == Opcode::Addi {
+            // Assume that any addi instruction that references a local symbol is loading a string.
+            // This hack is not ideal and results in tons of false positives where it will show
+            // garbage strings (e.g. misinterpreting arrays, float literals, etc).
+            // But not all strings are in the @stringBase pool, so the condition above that checks
+            // the target symbol name would miss some.
+            Some(DataType::String)
+        } else {
+            None
         }
     }
 
@@ -381,3 +383,196 @@ fn make_symbol_ref(symbol: &Symbol) -> Result<ExtabSymbolRef> {
     let demangled_name = cwdemangle::demangle(&name, &cwdemangle::DemangleOptions::default());
     Ok(ExtabSymbolRef { original_index: symbol.index().0, name, demangled_name })
 }
+
+fn guess_data_type_from_load_store_inst_op(inst_op: Opcode) -> Option<DataType> {
+    match inst_op {
+        Opcode::Lbz | Opcode::Lbzu | Opcode::Lbzux | Opcode::Lbzx => Some(DataType::Int8),
+        Opcode::Lhz | Opcode::Lhzu | Opcode::Lhzux | Opcode::Lhzx => Some(DataType::Int16),
+        Opcode::Lha | Opcode::Lhau | Opcode::Lhaux | Opcode::Lhax => Some(DataType::Int16),
+        Opcode::Lwz | Opcode::Lwzu | Opcode::Lwzux | Opcode::Lwzx => Some(DataType::Int32),
+        Opcode::Lfs | Opcode::Lfsu | Opcode::Lfsux | Opcode::Lfsx => Some(DataType::Float),
+        Opcode::Lfd | Opcode::Lfdu | Opcode::Lfdux | Opcode::Lfdx => Some(DataType::Double),
+
+        Opcode::Stb | Opcode::Stbu | Opcode::Stbux | Opcode::Stbx => Some(DataType::Int8),
+        Opcode::Sth | Opcode::Sthu | Opcode::Sthux | Opcode::Sthx => Some(DataType::Int16),
+        Opcode::Stw | Opcode::Stwu | Opcode::Stwux | Opcode::Stwx => Some(DataType::Int32),
+        Opcode::Stfs | Opcode::Stfsu | Opcode::Stfsux | Opcode::Stfsx => Some(DataType::Float),
+        Opcode::Stfd | Opcode::Stfdu | Opcode::Stfdux | Opcode::Stfdx => Some(DataType::Double),
+        _ => None,
+    }
+}
+
+// Given an instruction, determine if it could accessing data at the address in a register.
+// If so, return the offset added to the register's address, the register containing that address,
+// and (optionally) which destination register the address is being copied into.
+fn get_offset_and_addr_gpr_for_possible_pool_reference(
+    opcode: Opcode,
+    simplified: &ParsedIns,
+) -> Option<(i16, GPR, Option<GPR>)> {
+    let args = &simplified.args;
+    if guess_data_type_from_load_store_inst_op(opcode).is_some() {
+        match (args[1], args[2]) {
+            (Argument::Offset(offset), Argument::GPR(addr_src_gpr)) => {
+                // e.g. lwz. Immediate offset.
+                Some((offset.0, addr_src_gpr, None))
+            }
+            (Argument::GPR(addr_src_gpr), Argument::GPR(_offset_gpr)) => {
+                // e.g. lwzx. The offset is in a register and was likely calculated from an index.
+                // Treat the offset as being 0 in this case to show the first element of the array.
+                // It may be possible to show all elements by figuring out the stride of the array
+                // from the calculations performed on the index before it's put into offset_gpr, but
+                // this would be much more complicated, so it's not currently done.
+                Some((0, addr_src_gpr, None))
+            }
+            _ => None,
+        }
+    } else {
+        // If it's not a load/store instruction, there's two more possibilities we need to handle.
+        // 1. It could be loading a pointer to a string.
+        // 2. It could be moving the relocation address plus an offset into a different register to
+        //    load from later.
+        // If either of these match, we also want to return the destination register that the
+        // address is being copied into so that we can detect any future references to that new
+        // register as well.
+        match (opcode, args[0], args[1], args[2]) {
+            (
+                Opcode::Addi,
+                Argument::GPR(addr_dst_gpr),
+                Argument::GPR(addr_src_gpr),
+                Argument::Simm(simm),
+            ) => Some((simm.0, addr_src_gpr, Some(addr_dst_gpr))),
+            (
+                Opcode::Or,
+                Argument::GPR(addr_dst_gpr),
+                Argument::GPR(addr_src_gpr),
+                Argument::None,
+            ) => Some((0, addr_src_gpr, Some(addr_dst_gpr))), // `mr` or `mr.`
+            _ => None,
+        }
+    }
+}
+
+// We create a fake relocation for an instruction, vaguely simulating what the actual relocation
+// might have looked like if it wasn't pooled. This is so minimal changes are needed to display
+// pooled accesses vs non-pooled accesses. We set the relocation type to R_PPC_NONE to indicate that
+// there isn't really a relocation here, as copying the pool relocation's type wouldn't make sense.
+// Also, if this instruction is accessing the middle of a symbol instead of the start, we add an
+// addend to indicate that.
+fn make_fake_pool_reloc(offset: i16, cur_addr: u32, pool_reloc: &ObjReloc) -> Option<ObjReloc> {
+    let offset_from_pool = pool_reloc.addend + offset as i64;
+    let target_address = pool_reloc.target.address.checked_add_signed(offset_from_pool)?;
+    let orig_section_index = pool_reloc.target.orig_section_index?;
+    // We also need to create a fake target symbol to go inside our fake relocation.
+    // This is because we don't have access to list of all symbols in this section, so we can't find
+    // the real symbol yet. Instead we make a placeholder that has the correct `orig_section_index`
+    // and `address` fields, and then later on when this information is displayed to the user, we
+    // can find the real symbol by searching through the object's section's symbols for one that
+    // contains this address.
+    let fake_target_symbol = ObjSymbol {
+        name: "".to_string(),
+        demangled_name: None,
+        address: target_address,
+        section_address: 0,
+        size: 0,
+        size_known: false,
+        kind: Default::default(),
+        flags: Default::default(),
+        orig_section_index: Some(orig_section_index),
+        virtual_address: None,
+        original_index: None,
+        bytes: vec![],
+    };
+    // The addend is also fake because we don't know yet if the `target_address` here is the exact
+    // start of the symbol or if it's in the middle of it.
+    let fake_addend = 0;
+    Some(ObjReloc {
+        flags: RelocationFlags::Elf { r_type: elf::R_PPC_NONE },
+        address: cur_addr as u64,
+        target: fake_target_symbol,
+        addend: fake_addend,
+    })
+}
+
+// Searches through all instructions in a function, determining which registers have the addresses
+// of pooled data relocations in them, finding which instructions load data from those addresses,
+// and constructing a mapping of the address of that instruction to a "fake pool relocation" that
+// simulates what that instruction's relocation would look like if data hadn't been pooled.
+// Limitations: This method currently only goes through the instructions in a function in linear
+// order, from start to finish. It does *not* follow any branches. This means that it could have
+// false positives or false negatives in determining which relocation is currently loaded in which
+// register at any given point in the function, as control flow is not respected.
+// There are currently no known examples of this method producing inaccurate results in reality, but
+// if examples are found, it may be possible to update this method to also follow all branches so
+// that it produces more accurate results.
+fn generate_fake_pool_reloc_for_addr_mapping(
+    address: u64,
+    code: &[u8],
+    relocations: &[ObjReloc],
+) -> HashMap<u32, ObjReloc> {
+    let mut active_pool_relocs = HashMap::new();
+    let mut pool_reloc_for_addr = HashMap::new();
+    for (cur_addr, ins) in InsIter::new(code, address as u32) {
+        let simplified = ins.simplified();
+        let reloc = relocations.iter().find(|r| (r.address as u32 & !3) == cur_addr);
+
+        if let Some(reloc) = reloc {
+            // This instruction has a real relocation, so it may be a pool load we want to keep
+            // track of.
+            let args = &simplified.args;
+            match (ins.op, args[0], args[1], args[2]) {
+                (
+                    Opcode::Addi,
+                    Argument::GPR(addr_dst_gpr),
+                    Argument::GPR(_addr_src_gpr),
+                    Argument::Simm(_simm),
+                ) => {
+                    active_pool_relocs.insert(addr_dst_gpr.0, reloc.clone()); // `lis` + `addi`
+                }
+                (
+                    Opcode::Ori,
+                    Argument::GPR(addr_dst_gpr),
+                    Argument::GPR(_addr_src_gpr),
+                    Argument::Uimm(_uimm),
+                ) => {
+                    active_pool_relocs.insert(addr_dst_gpr.0, reloc.clone()); // `lis` + `ori`
+                }
+                (Opcode::B, _, _, _) => {
+                    if simplified.mnemonic == "bl" {
+                        // When encountering a function call, clear any active pool relocations from
+                        // the volatile registers (r0, r3-r12), but not the nonvolatile registers.
+                        active_pool_relocs.remove(&0);
+                        for gpr in 3..12 {
+                            active_pool_relocs.remove(&gpr);
+                        }
+                    }
+                }
+                _ => {}
+            }
+        } else if let Some((offset, addr_src_gpr, addr_dst_gpr)) =
+            get_offset_and_addr_gpr_for_possible_pool_reference(ins.op, &simplified)
+        {
+            // This instruction doesn't have a real relocation, so it may be a reference to one of
+            // the already-loaded pools.
+            if let Some(pool_reloc) = active_pool_relocs.get(&addr_src_gpr.0) {
+                if let Some(fake_pool_reloc) = make_fake_pool_reloc(offset, cur_addr, pool_reloc) {
+                    pool_reloc_for_addr.insert(cur_addr, fake_pool_reloc);
+                }
+                if let Some(addr_dst_gpr) = addr_dst_gpr {
+                    // If the address of the pool relocation got copied into another register, we
+                    // need to keep track of it in that register too as future instructions may
+                    // reference the symbol indirectly via this new register, instead of the
+                    // register the symbol's address was originally loaded into.
+                    // For example, the start of the function might `lis` + `addi` the start of the
+                    // ...data pool into r25, and then later the start of a loop will `addi` r25
+                    // with the offset within the .data section of an array variable into r21.
+                    // Then the body of the loop will `lwzx` one of the array elements from r21.
+                    let mut new_reloc = pool_reloc.clone();
+                    new_reloc.addend += offset as i64;
+                    active_pool_relocs.insert(addr_dst_gpr.0, new_reloc);
+                }
+            }
+        }
+    }
+
+    pool_reloc_for_addr
+}

objdiff-core/src/arch/x86.rs

@@ -51,7 +51,7 @@ impl ObjArch for ObjArchX86 {
                 address: 0,
                 size: 0,
                 op: 0,
-                mnemonic: String::new(),
+                mnemonic: Cow::Borrowed("<invalid>"),
                 args: vec![],
                 reloc: None,
                 branch_dest: None,
@@ -76,7 +76,7 @@ impl ObjArch for ObjArchX86 {
                 address,
                 size: instruction.len() as u8,
                 op,
-                mnemonic: String::new(),
+                mnemonic: Cow::Borrowed("<invalid>"),
                 args: vec![],
                 reloc: reloc.cloned(),
                 branch_dest: None,
@@ -242,7 +242,8 @@ impl FormatterOutput for InstructionFormatterOutput {
 
     fn write_mnemonic(&mut self, _instruction: &Instruction, text: &str) {
         self.formatted.push_str(text);
-        self.ins.mnemonic = text.to_string();
+        // TODO: can iced-x86 guarantee 'static here?
+        self.ins.mnemonic = Cow::Owned(text.to_string());
     }
 
     fn write_number(

objdiff-core/src/bindings/diff.rs

@@ -1,3 +1,4 @@
+#![allow(clippy::needless_lifetimes)] // Generated serde code
 use crate::{
     diff::{
         ObjDataDiff, ObjDataDiffKind, ObjDiff, ObjInsArgDiff, ObjInsBranchFrom, ObjInsBranchTo,
@@ -132,7 +133,7 @@ impl Instruction {
             address: instruction.address,
             size: instruction.size as u32,
             opcode: instruction.op as u32,
-            mnemonic: instruction.mnemonic.clone(),
+            mnemonic: instruction.mnemonic.to_string(),
             formatted: instruction.formatted.clone(),
             arguments: instruction.args.iter().map(Argument::new).collect(),
             relocation: instruction.reloc.as_ref().map(|reloc| Relocation::new(object, reloc)),

objdiff-core/src/bindings/report.rs

@@ -1,3 +1,4 @@
+#![allow(clippy::needless_lifetimes)] // Generated serde code
 use std::ops::AddAssign;
 
 use anyhow::{bail, Result};
@@ -173,8 +174,7 @@ impl Report {
                 continue;
             }
             fn is_sub_category(id: &str, parent: &str, sep: char) -> bool {
-                id.starts_with(parent)
-                    && id.get(parent.len()..).map_or(false, |s| s.starts_with(sep))
+                id.starts_with(parent) && id.get(parent.len()..).is_some_and(|s| s.starts_with(sep))
             }
             let mut sub_categories = self
                 .categories