Known quirks¶

The port deliberately reproduces the original's behavior — including its quirks. These are features for parity, verified byte-for-byte by the harness. Each entry says what the original does, what the port does, and why it stays.

Faithfully reproduced¶

1. NULL-group console keys resolve via `atoi`¶

find_console_id only matches keys whose console has a non-empty RA group. Everything else falls back to atoi — so RetroAchievementsSharp 3DS file silently hashes as console 3 (SNES), and Oric/TI83/TIC-80/ESCV/DOS keys print usage + exit 1. Use numeric ids for NULL-group consoles.

2. 64 MiB whole-file cap¶

MAX_BUFFER_SIZE truncates whole-file/buffered/.neo/disc-file hashes. Verbose output reports the truncated size — byte-identical to the C.

3. Wildcard path construction¶

On Windows the original scans the full pattern (FindFirstFile accepts forward slashes) but builds each open path from the backslash-split directory. Therefore dir/*.bin finds dir\a.bin yet tries to open .\a.bin. The port reproduces this exactly (the parity case for it expects both binaries to fail identically).

4. `.sms` had no extension mapping (until 12.4.0)¶

Pre-12.4.0, .sms files fell through to the generic fallback; 12.4.0 maps them to Master System. The port follows 12.4.0 (same hash either way — the mapping makes console resolution explicit).

5. CHD `LARGEST`-track stale offsets¶

The 1.8.3 HashCHD.cpp re-fetches metadata after computing offsets, so the LARGEST track selection can use stale values — both binaries report Not a PC-FX CD on the 3-track test CHD. Reproduced 1:1 (parity case expects the identical failure).

6. CD images need real MSF¶

The cdreader derives track_first_sector from the sector-16 header MSF (msf_to_lba − 150). Images whose sector headers say 00:00:00 produce −150 and break absolute-LBA reads. Real dumps encode 00:02:00 for LBA 0; the corpus fixtures follow the real convention.

7. AUDIO tracks are headerless¶

AUDIO cue tracks read as a raw 2352-byte stream (no header skip) — the corpus builds Jaguar CD bins accordingly.

8. `merge_callbacks` (12.2.1 bug)¶

The 12.2.1 C assigned error_message into verbose_message; the port inherited the bug and fixed it in Part II with 12.4.0 (the C# bug was found by re-auditing, and the fix is covered by the sync).

9. Zip entries >= 2 GiB (fixed in the port)¶

The C extracts a zip entry into a malloc'd buffer and can hold multi-GiB dumps; a C# byte[] is capped below 2 GiB, so the original port failed on 2 GiB+ 3DS dumps (new byte[entry.Length] throws). Fixed with a disk-backed fallback (LoadZippedFileToTemp + GenerateHashes) that hashes the extracted temp file — same bytes, same hash. Verified against real 2 GiB 3DS dumps in the real-ROM suite.

10. RVZ/WIA (supported) vs. WUX/WUD (unsupported)¶

rcheevos defines no reader for Dolphin's RVZ, but RetroAchievementsSharp adds one via RVZSharp (RvzFilereader, GPL-2.0-or-later): GameCube/Wii .rvz/.wia files are decoded on read and hashed exactly like the plain ISO — validated against DolphinTool conversions (6/6). The CLI selects the RVZ reader per file by extension (? auto-detect maps .rvz to GameCube + Wii). What the binaries actually do:

RVZ/WIA: fully supported — RetroAchievementsSharp 16 game.rvz, RetroAchievementsSharp 19 game.wia, or RetroAchievementsSharp ? game.rvz hash the decoded disc image (same hash as the converted ISO).
WUX/WUD (Wii U): no reader exists in either engine. RetroAchievementsSharp 20 game.wux hits the console with no hasher at all — id 20 (Wii U) exists in the console table but has no FromFile case (Unsupported console for file hash: 20).
? auto-detect for other unknown containers: an unknown extension falls through to the generic whole-file MD5 fallback with a warning (No console mapping specified for wux file extension - trying full file hash). The result is the hash of the compressed container bytes, not a game hash — do not use it for identification.

The WUX/WUD behavior is byte-identical to the 1.8.3 oracle (verified). The practical route for such files is conversion (WUX → raw image), then the existing iso handler applies.

Intentional differences (documented, not bugs)¶

Difference	Why
`-s` with no value prints usage + exit 1	the original segfaults (reads `argv[argc]`); crashing is not behavior worth replicating
directories matching a wildcard are skipped	the original emits a `????` line for them; noted edge case, kept for simplicity
Wii/WiiWare malloc-failure paths	managed code has no allocation failures; noted in comments
`RC_CCONV` calling convention	ABI detail, no native interop in the port
error/verbose callbacks route to stderr/stdout via `Console`	matches the C's `fprintf(stderr, ...)` / `printf(...)` streams