Port f3probe, f3brew, and f3fix to macOS

[sbrunner-atx]

What

Ports the three extra tools - f3probe, f3brew, and f3fix - to macOS, as invited by #8 and requested in #135.

How

The probing algorithm (libprobe.c) is already platform-independent; only the device-access layer needed work. All changes are behind APPLE guards:

• open() with F_NOCACHE replaces O_DIRECT (plus _DARWIN_C_SOURCE so Darwin extensions remain visible under _POSIX_C_SOURCE)
• DKIOCGETBLOCKCOUNT / DKIOCGETBLOCKSIZE replace BLKGETSIZE64 / BLKSSZGET
• whole-disk detection by device-name convention (/dev/rdiskN vs /dev/rdiskNsM) replaces the udev devtype check
• manual USB reset (used only by f3brew; f3probe uses RT_NONE) polls the device node instead of using a udev monitor
• software USB reset reports EOPNOTSUPP on macOS

f3fix on macOS is a new libparted-free backend (f3fix_darwin.c) that writes the MBR directly - fixing a fake drive only ever needs a single MBR partition ending at the last good sector from f3probe. MBR only for now; GPT support could follow if there is demand.

Validation

On macOS 26.5 (Intel, MacBookPro16,1), clang + brew argp-standalone:

• make extra builds all three tools with no warnings
• f3probe --debug-unit-test: SUMMARY: Perfect!
• Real hardware (SanDisk Ultra 128 GB USB 3.0, genuine):

$ sudo ./build/f3probe --destructive --time-ops /dev/rdisk7
Good news: The device /dev/rdisk7 is the real thing
Device geometry:
         *Usable* size: 114.60 GB (240328704 blocks)
        Announced size: 114.60 GB (240328704 blocks)
Probe time: 1 min 36 s

The Linux build is unchanged: gcc on Ubuntu compiles all targets with no new warnings, and the Linux code path is untouched.

Notes

• Happy to add a macos-latest job to the GitHub Actions workflow in a follow-up commit so the port cannot regress.
• Disclosure: developed with AI assistance (Claude); reviewed, built, and hardware-validated by the author.

[sbrunner-atx]

For anyone who wants to try this on macOS without building manually, I have published a Homebrew tap pinned to this PR branch:

brew tap sbrunner-atx/f3
brew install sbrunner-atx/f3/f3-macos

Tap: https://github.com/sbrunner-atx/homebrew-f3 (README includes macOS usage notes: raw whole-disk device, unmount first, sudo). I will retire the tap once this is merged and the official formula picks up make extra.

Full f3probe hardware-validation transcript (genuine SanDisk Ultra 128 GB, macOS 26.5 Intel) available on request.

[sbrunner-atx]

Rebased onto current master; builds clean on macOS with -Wall -Wextra -pedantic (f3write, f3read, f3probe, f3brew, f3fix).

I see #285 covers much of the same ground for f3probe and f3brew. Rather than have you compare two black boxes, here is my honest read:

• Build f3probe and f3brew on macOS #285 has a macOS CI job and this PR does not. That is the better approach, and I would rather it win than duplicate it.
• Build f3probe and f3brew on macOS #285 fixes the keg-only argp-standalone lookup (brew --prefix argp-standalone). This PR does not. Mine builds on my Intel Mac only because /usr/local/include/argp.h happens to exist there; on Apple Silicon, where keg-only formulae are not linked into the prefix, it would fail. That fix is correct and needed either way.
• What this PR has that Build f3probe and f3brew on macOS #285 does not is f3fix. Build f3probe and f3brew on macOS #285 leaves it out because of libparted. I replaced libparted with a small MBR writer (src/f3fix_darwin.c), so f3fix builds on macOS with no new dependency.

Testing was on Intel macOS against five physical SanDisk Ultra 128 GB drives, three good and two defective. The bad ones accept a full-capacity write and then drop off the USB bus 37 to 100 seconds into read-back. f3probe reported the good ones correctly: usable size = announced size, 114.60 GB.

One gap I should name: the two defective drives went back to the seller before I thought to run f3probe against them, so I have no transcript of the port's error paths under a mid-probe disconnect. If that matters to you, say so and I will reproduce it the next time a drive fails on me.

Happy to rebase the f3fix part on top of #285 so it lands separately, if that is easier for you to review.

[AltraMayor]

Hi @sbrunner-atx,

Given your effort to port all three extra tools to macOS, I'm assuming that you're willing to join forces to get your code merged.

My main concern with the proposed code at this point is the widespread of #if[n]def __APPLE__. This makes maintenance harder and will be a nightmare when someone decides to port the extra tools to another platform. I propose isolating the platform-dependent code in separate files. For example, src/libdevs.c could include the platform-dependent code using the following preprocessor directives:

#if defined(__linux__)
#include "linux_libdevs.c"
#elif defined(__APPLE__)
#include "darwin_libdevs.c"
#else
#error "Platform not supported"
#endif

I'm open to evaluating other ways to organize the code if you have alternative approaches to propose.

If you choose to move forward with this work, keep this pull request open and create another pull request following the new approach to organize the code and only port f3probe. Once we work together to review and eventually merge this new pull request, it should be easier later to have a pull request to port f3brew and another to port f3fix.

Finally, we should invite the developers of pull requests #279 and #285 to collaborate with us. They can help by reviewing code, testing, and, perhaps, even writing some of the code. Their help would be important because I'm not a Mac developer and don't have a Mac.

If we all pull this off, we may motivate other developers to port the extra tools to their platforms. So many more users will benefit from the extra tools on their platform.