skills/debuggers/debug-optimized-builds/SKILL.md
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.
npx skillsauth add mohitmishra786/low-level-dev-skills debug-optimized-buildsInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
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.
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.
# 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
(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:
Workarounds:
// 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
With optimization, frequently-called small functions get inlined. GDB shows these as extra frames:
(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
# 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
Optimizers reorder instructions, so the "current line" in GDB may jump around:
# 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
With optimization, threads may race in unexpected ways when stepping:
# 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
Split DWARF offloads debug info to .dwo files, reducing linker input:
# 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)
# 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
skills/debuggers/gdb for full GDB session managementskills/debuggers/dwarf-debug-format for DWARF debug info detailsskills/debuggers/core-dumps for post-mortem debugging of optimized crashesskills/compilers/gcc for -Og, -g, and debug flag selectiondevelopment
QEMU/KVM skill for virtualization and kernel development. Use when running qemu-system-x86_64 with KVM, configuring virtio devices, VFIO passthrough, QMP monitor, libvirt, or booting custom kernels. Activates on queries about QEMU, KVM, virtio, VFIO, virsh, virt-install, or -kernel -append.
development
Hardware virtualization internals skill for Intel VT-x and AMD-V. Use when studying VMCS/VMCB, EPT/NPT page tables, VMEXIT handling, APIC virtualization, or building minimal hypervisors. Activates on queries about VMX, SVM, VMCS, EPT, NPT, VMEXIT, or type-1 hypervisor.
testing
Linux containers internals skill for namespaces, cgroups, and OCI. Use when understanding clone/unshare namespaces, cgroups v2 limits, overlayfs, runc, seccomp profiles, capabilities, or escape mitigations. Activates on queries about namespaces, cgroups, overlayfs, runc, seccomp-bpf, OCI spec, or container escape.
tools
Reverse engineering skill for binary analysis. Use when decompiling with Ghidra, analyzing with radare2, scripting RE tools, triaging with strings/file/xxd, or diffing binaries. Activates on queries about Ghidra, radare2, r2, decompiler, Binary Ninja, Diaphora, or stripped binary analysis.