cacadump

A terminal hex editor with binary templates.

• gtcaca — a libcaca-based TUI toolkit — draws the interface.
• libpcapng / pycapng and its POSA decoder language provide the parsing engine.

Binary structure is described declaratively in .posa files (not code); the POSA engine parses a buffer into a field tree, and that same tree drives both the coloured hex grid and the template pane.

 cacadump   photo.png   69 bytes   fmt:PNG                              F1 Help
 00000000  89 50 4E 47 0D 0A 1A 0A  00 00 00 0D 49 48 44 52 |.PNG........IHDR| │ Inspector @ 0x10
 00000010  00 00 00 04 00 00 00 04  08 06 00 00 00 A9 F1 9E |................| │ uint32   4
 00000020  7E 00 00 00 0C 49 44 41  54 78 9C 63 60 20 0D 00 |~....IDATx.c` ..| │ …
                                                                            │ Template: PNG
                                                                            │ PNG
                                                                            │   Signature: 8 bytes
                                                                            │   Chunks
                                                                            │     IHDR (13 bytes)
                                                                            │       Width: 4
                                                                            │       Color type: Truecolor+Alpha (6)
 0x10 (16)  HEX OVR                                                     PNG

Features

• Hex + ASCII panes with a movable caret, insert/overwrite editing, undo/redo.
• Binary templates via POSA: autodetects the format from the buffer's magic, dissects it, colours each byte by the field it belongs to, and shows the parsed structure in a live tree that follows the cursor.
• Data inspector: the bytes at the cursor decoded as int8…int64 / uint / float32 / float64 (both byte orders), char and binary — all at once.
• Goto offset, find (text or 0x.. hex) with find-next, selection anchor, bookmarks.
• Ships decoders for PNG, GIF, BMP, GZIP, WAV and — because POSA is libpcapng's own decoder language — every network protocol libpcapng/carcal ships (DNS, HTTP, TLS, DHCP, SMB, …), usable on any selected buffer.

Install

pip install cacadump      # pulls in gtcaca and pycapng wheels
cacadump path/to/file

The .posa decoders live in the fileformat.posa submodule, so clone with --recurse-submodules (or run git submodule update --init in an existing checkout).

For local development against checkouts of gtcaca and libpcapng:

git clone --recurse-submodules https://github.com/stricaud/cacadump
python -m venv .venv
PKG_CONFIG_PATH=$(brew --prefix libcaca)/lib/pkgconfig \
  .venv/bin/pip install scikit-build-core pybind11 setuptools-scm
PKG_CONFIG_PATH=$(brew --prefix libcaca)/lib/pkgconfig \
  .venv/bin/pip install --no-build-isolation -e ../gtcaca -e ../libpcapng
.venv/bin/pip install -e .

Keys

Everything on the menu bar is also a key, and vice versa — File, Edit, View, Template and Help route to the same commands. F10 is gtcaca's own menu binding; F9 is wired to the same thing because F10 is intercepted by some terminals.

Wrapped files

Some formats are a header around a compressed stream. A NoteWorthy score is six bytes then zlib, and NWC1_SCORE only matches what comes out of it — so cacadump opens such a file at its payload and the template just works. The title bar shows the layer stack ([1/1 file > NWC (zlib)]); Ctrl-U drops back to the bytes on disk, Ctrl-D goes in again, and --raw skips decoding entirely.

Decoding is automatic only when the outer bytes have no decoder of their own. A .gz keeps its gzip.posa view — inflating it silently would hide the very structure that decoder exists to show — so you descend into it with Ctrl-D.

Each layer is a separate buffer with its own offsets, cursor and undo history, and offsets are never mapped across layers: DEFLATE has no byte-to-byte correspondence between input and output, so "the compressed bytes behind this field" would be a fiction. Saving re-encodes edited payloads back through their containers; a layer you did not edit is written back exactly as it was read, so opening and saving a file you didn't change never rewrites it.

Writing a template

Templates are .posa files (see the src/cacadump/protos/ submodule and libpcapng's doc/posa.md). A decoder is declarative — types, enums, repeat, when/else, scope, bits, and a label/info line:

Object<main> PNG
    abbrev "png"
    col "PNG"
    required bytes<8> signature "Signature"
    repeat until end as chunk "Chunks"
        label "%s (%u bytes)" ctype, length
        required uint32 length "Length"
        required str<4>  ctype  "Type"
        when ctype == 0x49484452:            # "IHDR"
            required uint32 width "Width"
            required uint32 height "Height"
            required uint8  bit_depth "Bit depth"
            required uint8  color_type "Color type"
                Truecolor+Alpha = 6
        else:
            required bytes[length] data "Data"
        required uint32 crc hex "CRC"

Point CACADUMP_PROTOS_DIR at a directory of your own (or several, separated by :) and the format is understood on the next open — no code change. Decoders meant for everyone belong upstream in fileformat.posa rather than dropped into the submodule checkout, where they would be lost on the next git submodule update.

Notes / limitations

• POSA is byte-order-per-type, so formats that switch endianness or word size on a runtime field (ELF/PE/Mach-O 32-vs-64) are only partially expressible today.
• Editing operates on an in-memory buffer; very large files load fully into memory.

License

MIT.