Skip to content

CLAUDE.md

Latest commit

 

History

History
50 lines (30 loc) · 4.32 KB

CLAUDE.md

File metadata and controls

50 lines (30 loc) · 4.32 KB
Raw

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Project

Windows C port of the Linux hexdump CLI. Intentionally a stripped port: only file input is supported (no stdin), and custom-format strings (-e, -f) from the GNU original are deliberately omitted. Adds a -h switch in place of the man hexdump page that doesn't exist on Windows. Built from scratch without referencing the original sources, so don't assume behavioral parity beyond what's documented in README.md / info.c's help text.

Build

MSBuild project (hexdump.sln / hexdump.vcxproj), Platform Toolset v142 (Visual Studio 2019). Four configurations: {Debug,Release} x {Win32,x64}.

# from a Developer PowerShell / VS dev cmd
msbuild hexdump.sln /p:Configuration=Release /p:Platform=x64
# or open hexdump.sln in Visual Studio and build

Output binary: <Platform>\<Configuration>\hexdump.exe (e.g. x64\Release\hexdump.exe).

There are no tests, no linter config, and no CI. Validate changes by running the built binary against sample files and comparing output to the expected format documented in info.c::print_help.

Charset quirk: Debug|Win32 passes /execution-charset:cp852 to cl. The other three configurations don't. Source files contain Czech characters in author-attribution comments (Marek Polášek) — if you save them in a different encoding the Win32 Debug build can break in surprising ways. Prefer not to re-save source files unless you know what encoding they're in.

Architecture

Three translation units, dispatched from main.c:

  • main.c — argv parsing as a character-by-character state machine. Sets bitflags from switches.h (SW_DUMP_BYTE_OCTAL, SW_VERBOSE, SW_INVALID, SW_FILEUNKNOWN, SW_OFFSETFORMAT, ...) into a single unsigned int switches register. After parsing, error flags are checked in priority order (help → invalid → length format → offset format → missing file → misplaced file), each producing a specific exit code (1–5). Then the appropriate dump_* function is dispatched based on the dump flag bits.
  • dumps.c / dumps.h — one dump_* function per output mode (dump_plain_hex, dump_byte_octal, dump_byte_char, dump_canonical, dump_word_decimal, dump_word_octal, dump_word_hex). All read 16-byte rows from FILE*, format them, and update *position. The non-verbose collapse-to-asterisk behavior (suppress runs of identical lines) is implemented inside each dump function via buffer_current / buffer_previous comparison. Format constants (column widths, escape sequences for -c, ASCII printable boundaries) live in dumps.h.
  • info.c / info.h — print_help (the man replacement), print_error, print_file_error (formats errno via strerror_s), print_error_memory. All prepend a header banner. VERSION macro lives here.

Argument-parsing rules baked into main.c

  • The file path must be the last argument; appearing earlier triggers SW_FILEELSEWHERE.
  • -h, -n, -s, -v must each be a standalone switch (not combined like -hv); detected by checking j == 1 in the per-character loop. Dump-mode flags (-bcCdox) can be combined.
  • -s <offset> parses C-style numeric prefixes itself: leading 0x/0X → hex, leading 0 → octal, otherwise decimal; trailing b/k/m multiplies by 512 / 1024 / 1048576.

Version bumps

VERSION string lives in two places that must stay in sync: info.h (#define VERSION) and resources.rc (FILEVERSION, PRODUCTVERSION, and the StringFileInfo block).

Output encoding

Console output is not UTF-8 — it follows the active OEM/DOS codepage. Bytes ≥ 0x80 in -c mode are emitted raw, so they render per the user's national codepage. This is intentional; don't "fix" it to UTF-8 without considering the design note in README.md.

Known rough edges (to be aware of when editing)

  • dumps.h declares dump_byte_octal and dump_word_decimal without an offset parameter while the other dump prototypes take one — easy to mis-call. Check signatures before adding new dump variants.
  • print_file_error reads errno after a malloc call; if the malloc itself fails, errno may have been overwritten before strerror_s runs. Be careful when reordering allocations near error paths.