probe-ida-function

---

name: probe-ida-function description: Headlessly decompile a function in an IDA database and dump its pseudocode, lvars, ctree AST, calls, and microcode — so you can iterate on the ioshelper hex-rays plugin from a shell without staring at IDA's GUI. Use when the user asks to "probe a function", "check what hex-rays produces", "see the pseudo for X", or wants to verify that a plugin change took effect on a specific binary/EA. Two paths: a fast IPC server for iteration (idat_ipc_launch.sh + idat_ipc_client.py, ~50ms/request) and a cold standalone probe for verification (probe_func.sh, ~30-90s per invocation). Both load the ioshelper hooks (SwiftClassCallHook, SwiftPrologRewriteHook, SwiftLogRewriteHook) and run fix_swift_types().

Probe an IDA function headlessly

Two-tier setup:

1. IPC server (idat_ipc_launch.sh + idat_ipc_client.py) — long-running idat with a Unix-socket listener. Each reload / decompile / eval is 30-100ms instead of paying 30-90s cold-start cost. Use this for iteration.
2. Standalone probe (probe_func.sh) — spawns a fresh idat per query. Use this only for verification once iteration is done — it proves the result reproduces on a clean session.

IPC server (iteration path)

Launch once per IDB (keeps running until you quit it):

bash src/scripts/idat_ipc_launch.sh <binary_or_idb>

The launcher blocks. Either run it in another terminal or background it (&).

Wait for /tmp/ioshelper-idat.sock to appear (initial auto-analysis on a fresh binary still takes 60-120s on the cold IDB build; the cost is paid ONCE).

Drive from a separate shell:

python3 src/scripts/idat_ipc_client.py ping                       # liveness check
python3 src/scripts/idat_ipc_client.py reload                     # re-import plugin sources
python3 src/scripts/idat_ipc_client.py decompile 0x1000173C0      # pseudo dump
python3 src/scripts/idat_ipc_client.py decompile 0x1000173C0 --sections pseudo lvars
python3 src/scripts/idat_ipc_client.py eval "idc.get_type(0x1000318A0)"
python3 src/scripts/idat_ipc_client.py quit                       # graceful shutdown

reload re-imports swift_types.py, prolog_rewrite.py, log_hook.py and re-installs hooks — so edits to those files take effect without restarting idat. It also calls clear_cached_cfuncs() so subsequent decompiles rebuild from scratch.

IOSHELPER_IDAT_SOCK=/tmp/other.sock lets you run multiple servers in parallel (different IDBs).

Known limitation

The decompile's stored type (line 2 of every decompile response) reflects the current IDB state correctly. The rendered pseudo header sometimes lags one decompile pass behind when the plugin changes a function's prototype mid-flight — only a fresh IDA process (i.e. the cold path) re-renders the header from the up-to-date stored type. Use idc.get_type(ea) via eval or the stored-type line to confirm what's actually persisted; use the cold path for final visual verification.

Standalone probe (verification path)

bash src/scripts/probe_func.sh <binary_or_idb> <ea> [section ...]

• <binary_or_idb> — path to the Mach-O binary or an existing .i64. If you pass a binary and no .i64 exists yet, idat creates one (initial auto-analysis can take a minute or two).
• <ea> — function entry point, hex (0x10001A41C) or decimal.
• [section ...] — any subset of pseudo lvars ast calls mc (default: all). Add --all-mc after mc to dump every microcode maturity instead of just MMAT_CALLS + MMAT_GLBOPT3.

Sections are delimited with === NAME … / --- end NAME … banners so a shell consumer can awk or sed -n the chunk it cares about.

Output goes to stdout; idat's own progress noise is captured and dropped.

When you need a clean slate

If a prior run left the IDB in a confused state — .i64 missing but .id0/.id1/.nam/.til still present, or a stale type the plugin set is interfering — delete the aux files and let idat re-create the database from the binary:

cd <dir>; rm -f <name>.id0 <name>.id1 <name>.nam <name>.til <name>.i64

Then re-run the probe. Re-analysis can take 60–120 s on the first call.

Typical patterns

Just see the pseudo:

bash src/scripts/probe_func.sh /path/to/Binary 0x10001A41C pseudo > /tmp/p.txt

Pseudo + AST around one call site:

bash src/scripts/probe_func.sh /path/to/Binary.i64 0x10001A41C pseudo ast > /tmp/p.txt
grep -n -A 30 "lookForPattern" /tmp/p.txt

Check what hex-rays saw at MMAT_CALLS (before optimizations) vs MMAT_GLBOPT3:

bash src/scripts/probe_func.sh /path/to/Binary.i64 0x10001A41C mc > /tmp/mc.txt

Environment

• $IDAT — override the idat binary path. Otherwise the script auto-discovers macOS install locations from IDA Professional 9.0–9.3 and IDA Pro 8.4.
• Requires that the project's plugin code is importable from the IDA Python environment (it is — src/scripts/probe_func.py adds src/ to sys.path).

What it actually does

1. ida_auto.auto_wait() to let initial analysis finish.
2. Instantiates and .hook()s SwiftClassCallHook, SwiftPrologRewriteHook, SwiftLogRewriteHook (headless idat -A doesn't auto-install Hexrays_Hooks subclasses).
3. Calls fix_swift_types() once (the StartupScript path doesn't run in headless mode either).
4. ida_hexrays.mark_cfunc_dirty(ea, False) to invalidate any cached cfunc, then decompile(ea).
5. Dumps the requested sections.

The hook installation and fix_swift_types() invocation are what make the output match what a user with a real IDA GUI would see after the plugin loaded. Without them, you're testing pure stock hex-rays — not the plugin.