debug-optimized-builds
Debugging optimized builds skill for diagnosing issues in release code. Use when debugging RelWithDebInfo builds, using -Og for debuggable optimization, working with split-DWARF, applying GDB scheduler-locking, reading inlined frames, or understanding "value optimized out" messages. Activates on queries about debugging optimized code, RelWithDebInfo, -Og, inlined functions in GDB, value optimized out, GDB with -O2, or debugging release builds.
Works with
---
name: debug-optimized-builds
description: Debugging optimized builds skill for diagnosing issues in release code. Use when debugging RelWithDebInfo builds, using -Og for debuggable optimization, working with split-DWARF, applying GDB scheduler-locking, reading inlined frames, or understanding "value optimized out" messages. Activates on queries about debugging optimized code, RelWithDebInfo, -Og, inlined functions in GDB, value optimized out, GDB with -O2, or debugging release builds.
license: MIT
---
# Debugging Optimized Builds
## Purpose
Guide agents through debugging code compiled with optimization: choosing the right debug-friendly optimization level, reading inlined frames, diagnosing "value optimized out", using split-DWARF for faster debug builds, and applying GDB techniques specific to optimized code.
## Triggers
- "GDB says 'value optimized out' — what does that mean?"
- "How do I debug a release build?"
- "How do I see inlined function frames in GDB?"
- "What's the difference between -O0 and -Og for debugging?"
- "How do I use RelWithDebInfo with CMake?"
- "Breakpoints in optimized code land on wrong lines"
## Workflow
### 1. Choose the right build configuration
```
Goal?
├── Full debuggability, no optimization
│ → -O0 -g (slowest, all vars visible)
├── Debuggable, some optimization (recommended for most dev work)
│ → -Og -g (-Og keeps debug experience good)
├── Release build with debug info (shipped, debuggable crashes)
│ → -O2 -g -gsplit-dwarf (or -O2 -g1 for lighter info)
└── Full release (no debug symbols)
→ -O2 -DNDEBUG
```
**`-Og`**: GCC's "debug-friendly optimization" — enables optimizations that don't interfere with debugging. Variables stay in registers where GDB can see them. Line numbers stay accurate. **Best balance for development.**
```bash
# GCC / Clang
gcc -Og -g -Wall main.c -o prog
# CMake build types
cmake -S . -B build -DCMAKE_BUILD_TYPE=Debug # -O0 -g
cmake -S . -B build -DCMAKE_BUILD_TYPE=RelWithDebInfo # -O2 -g -DNDEBUG
cmake -S . -B build -DCMAKE_BUILD_TYPE=Release # -O2 -DNDEBUG
```
### 2. "Value optimized out" — causes and workarounds
```text
(gdb) print my_variable
$1 = <optimized out>
```
This means the compiler decided the variable's value doesn't need to be stored at this point — it might be:
- Kept only in a register (not the one GDB is looking at)
- Folded into a constant by constant propagation
- Eliminated because it's not used after this point
- Replaced by a later optimized value
**Workarounds:**
```c
// 1. Mark variable volatile (prevents optimization away)
volatile int counter = 0;
// Use sparingly — changes semantics
// 2. Use GCC attribute
int counter __attribute__((used)) = 0;
// 3. Compile problematic TU at lower optimization
// In CMake:
set_source_files_properties(tricky.c PROPERTIES COMPILE_FLAGS "-O0")
// 4. Use -Og instead of -O2 for the whole build
// 5. Look at register values directly
// (gdb) info registers
// (gdb) p/x $rax # value may be in a register
```
### 3. Reading inlined frames in GDB
With optimization, frequently-called small functions get inlined. GDB shows these as extra frames:
```text
(gdb) bt
#0 process_packet (data=0x7ff..., len=<optimized out>)
at network.c:45
#1 0x0000... in dispatch_handler (pkt=0x7ff...)
at handler.c:102
#2 (inlined by) event_loop () at main.c:78
#3 0x0000... in main () at main.c:200
# (inlined by) frames are virtual — they show the call chain
# that was inlined into the actual frame above
```
```bash
# Navigate inlined frames
(gdb) frame 2 # jump to the inlined frame
(gdb) up # move up through frames (including inlined)
(gdb) down # move down
# Show all frames including inlined
(gdb) backtrace full
# Set breakpoint inside inlined function
(gdb) break network.c:45 # may hit multiple inlined call sites
(gdb) break process_packet # hits all inline expansions
```
### 4. Line number discrepancies
Optimizers reorder instructions, so the "current line" in GDB may jump around:
```bash
# See which instructions map to which source lines
(gdb) disassemble /s function_name # interleaved source and asm
# Step by machine instruction (more accurate in optimized code)
(gdb) si # stepi — one machine instruction
(gdb) ni # nexti — one machine instruction (no step into)
# Show mixed source/asm at current point
(gdb) layout split # TUI mode: source + asm side by side
(gdb) set disassemble-next-line on
# Jump to specific address (when line stepping is unreliable)
(gdb) jump *0x400a2c
```
### 5. GDB scheduler-locking for optimized multithreaded code
With optimization, threads may race in unexpected ways when stepping:
```bash
# Lock the scheduler — only the current thread runs while stepping
(gdb) set scheduler-locking on
# Modes:
# off — all threads run freely (default)
# on — only current thread runs while stepping
# step — only current thread runs while single-stepping
# (all run on continue)
# replay — for reverse debugging
# Common debugging session
(gdb) set scheduler-locking step # prevent other threads interfering with step
(gdb) break my_function
(gdb) continue
(gdb) set scheduler-locking on # lock while examining
(gdb) next
(gdb) set scheduler-locking off # unlock to continue normally
```
### 6. split-DWARF — faster debug builds
Split DWARF offloads debug info to `.dwo` files, reducing linker input:
```bash
# Compile with split DWARF
gcc -g -gsplit-dwarf -O2 -c file.c -o file.o
# Creates: file.o (object) + file.dwo (DWARF sidecar)
# Link — no debug info in final binary, just references
gcc -g -gsplit-dwarf file.o -o prog
# GDB finds .dwo files via the path embedded in the binary
gdb prog # works automatically if .dwo files are next to the binary
# Package all .dwo into a single .dwp for distribution
dwp -o prog.dwp prog # GNU dwp tool
gdb prog # with .dwp in same directory
# CMake
add_compile_options(-gsplit-dwarf)
```
### 7. Useful GDB commands for optimized builds
```bash
# Show where variables actually live (register vs stack)
(gdb) info locals # all locals (may show <optimized out>)
(gdb) info args # function arguments
# Force evaluation of an expression
(gdb) call (int)my_func(42) # call actual function to get value
# Watch a memory address directly (not a variable name)
(gdb) watch *0x7fffffffe430
# Print memory contents
(gdb) x/10xw $rsp # 10 words at stack pointer (hex)
(gdb) x/s 0x4008a0 # string at address
# Catch crashes without debug symbols
(gdb) bt # backtrace — shows addresses even without symbols
(gdb) info sharedlibrary # shows loaded libs for symbol resolution
# .gdbinit helpers for optimized debugging
# set print pretty on
# set print array on
# set disassembly-flavor intel
```
## Related skills
- Use `skills/debuggers/gdb` for full GDB session management
- Use `skills/debuggers/dwarf-debug-format` for DWARF debug info details
- Use `skills/debuggers/core-dumps` for post-mortem debugging of optimized crashes
- Use `skills/compilers/gcc` for `-Og`, `-g`, and debug flag selectionMore Debugging skills
diagnosing-bugs
mattpocock/skills
Diagnosis loop for hard bugs and performance regressions. Use when the user says "diagnose"/"debug this", or reports something broken/throwing/failing/slow.
explore-code
lllllllama/rigorpilot-skills
Rigor Improve implementation leaf skill for auditable candidate implementation in deep learning research repositories. Use when the researcher explicitly authorizes exploratory work on an isolated branch or worktree to transplant modules, adapt a backbone, add LoRA or adapter layers, replace a head, or stitch together meaningful low-risk migration ideas with rollback-aware records in `explore_outputs/`. Do not use for end-to-end exploration orchestration on top of `current_research`, trusted baseline reproduction, conservative debugging, environment setup, verified contribution claims, or default repository analysis.
safe-debug
lllllllama/rigorpilot-skills
Rigor Debug / Rigor Audit skill for deep learning research work. Use when the user pastes a traceback, terminal error, CUDA OOM, checkpoint load failure, shape mismatch, NaN loss symptom, or training failure and wants conservative diagnosis before any patching, with debug fixes clearly separated from research contributions. Do not use for broad refactoring, speculative adaptation, automatic exploratory patching, or general repository familiarization.

