build, ipc, doc: fix the OpenBSD and FreeBSD builds, including multiprocess

[jamescowens]

Neither BSD could be built as documented. FreeBSD failed to configure at all without a workaround flag, and OpenBSD could not compile the multiprocess build. Both turned out to be portability bugs in this tree rather than packaging problems on those systems.

Verified on OpenBSD 7.8 (amd64) and FreeBSD 15.0-RELEASE (amd64), 4 vCPU / 4 GB each.

FreeBSD — Boost was not found

CMake Error at BoostConfig.cmake:141 (find_package):
  No suitable build variant has been found.
  * libboost_filesystem.so.1.88.0 (release runtime, Boost_USE_DEBUG_RUNTIME=TRUE)
  * libboost_filesystem.a (static, default is shared, ...)

CMake's FindBoost defaults Boost_USE_DEBUG_RUNTIME to TRUE when the caller has not set it (FindBoost.cmake:1699). That selects the MSVC /MDd runtime and means nothing elsewhere — but Boost's own CMake config files honour it on every platform; each variant file opens with:

if(Boost_USE_DEBUG_RUNTIME)
  _BOOST_SKIPPED("libboost_foo.so" "release runtime, ...")

FreeBSD builds only release-runtime variants, so every variant was rejected. We reach that default because CMP0167 OLD routes the version probe through the FindBoost module.

Set OFF before the probe rather than cleared afterwards: FindBoost only assigns its default under if(NOT DEFINED ...), so a value set here survives, it also covers the pre-1.70 module-mode branch, and it doesn't discard a value the user passed with -D.

This only bites where FindBoost runs its own module search — it returns early when a BoostConfig.cmake is found, which is why the failure doesn't reproduce everywhere.

OpenBSD — multiprocess did not compile

src/ipc/peercred.cpp:58: error: variable has incomplete type 'struct ucred'

The peer-uid check selected the Linux path on #if defined(SO_PEERCRED), assuming that option implies Linux's struct ucred. OpenBSD also defines SO_PEERCRED (sys/socket.h:118) but its payload is struct sockpeercred (line 306). FreeBSD and macOS don't define it at all, which is why only OpenBSD broke.

The condition now keys on __linux__ and is defined once as GRIDCOIN_PEERCRED_SO_PEERCRED, because CheckPeerCredentials() and PeerCredentialEnforcement() must agree — the latter is what the startup log advertises as the enforcement mechanism in force, and a silent disagreement there would misstate a security property. Every non-Linux target we support provides getpeereid(3).

Documentation

Both guides claimed verification against 5.5.0.0, neither mentioned multiprocess or Cap'n Proto, and both pointed at the wrong output directory (build/src/... rather than build/bin/...). Beyond fixing those:

• Cap'n Proto recorded as the multiprocess-only dependency on both.
• OpenBSD needs ulimit -d raised before a multiprocess build. login.conf's default class caps datasize at 1536M and clang needs more for the generated Cap'n Proto proxies (one preprocessed TU is ~9 MB), dying with LLVM ERROR: out of memory. That's the per-process limit, not a RAM shortage — worth stating because the message reads like one. Fewer jobs than cores is also recommended: -j2 completed reliably on 4 vCPU / 4 GB, -j4 did not.
• OpenBSD's clang accepts but ignores -fstack-clash-protection (many argument unused warnings), so that mitigation is not applied there despite being requested. Recorded so nobody assumes it is.
• The FreeBSD guide's existing Boost workaround was wrong. It required -DBoost_USE_DEBUG_RUNTIME=OFF, described it as accepting release Boost "even when building with debug symbols", and offered Release as an alternative that avoids it. I tested all three build types — Release, RelWithDebInfo and Debug all failed identically; the build type is irrelevant. The flag is no longer needed and the note now says what actually happened.
• FreeBSD 15.0's capnproto package warns of a kernel bug fixed in 15.1. Reproduced with the finding that Gridcoin's IPC worked anyway, so readers know why pkg shows it rather than assuming it blocks multiprocess.

Verification

	OpenBSD 7.8	FreeBSD 15.0
Monolithic build	✅	✅
Multiprocess build	✅	✅
Binaries run	✅	✅
MP GUI attaches over IPC	✅ GUI loaded.	✅ GUI loaded.
Peer-cred mechanism	getpeereid	getpeereid

Linux reconfigured and rebuilt with these changes: 0 errors, Boost found identically, core suite No errors detected, Qt suite 8/8. Lint clean with a populated COMMIT_RANGE.

The hardening probe is fixed, not just documented

An earlier revision of this PR documented that OpenBSD's clang silently ignores
-fstack-clash-protection and left it at that. That was the weaker answer: the build was
still requesting a mitigation it did not get, and emitting 316 warnings saying so.

check_cxx_compiler_flag only proves the driver did not reject a flag's spelling. The
probes now run with -Werror=unused-command-line-argument, so a flag the compiler accepts
and then ignores fails the check and is not applied — reported once at configure time
instead of once per translation unit:

-- Hardening: -fstack-clash-protection not applied (compiler accepts no working
   implementation for this target)

GCC has no such warning and errors on the option, so it is probed for first. Setting it
unconditionally would make every hardening probe fail on GCC and silently disable
hardening on Linux — the exact opposite of the intent, and the thing most worth getting
wrong here.

This does not replace the Mach-O platform gate. -mbranch-protection=standard on
macOS arm64 is accepted and honoured and wrong — it emits real pointer-authentication
code that defeats libunwind. No probe can see that, so it stays gated by platform.

Verified on three toolchains

Toolchain	probe	stack-protector	cf-protection	stack-clash
Linux GCC 15.3	Failed (required)	✅	✅	✅
Linux clang 19.1.7	Success	✅	✅	✅
OpenBSD 7.8 clang 19.1.7	Success	✅	✅	correctly withheld

On OpenBSD a 208-object build produced 0 "argument unused" warnings (down from 316
across a full build) with 0 errors, and the compile options carry only the two flags that
actually work there. Linux is unchanged on both compilers.

Not addressed

Neither BSD is in CI, so nothing here is regression-protected — the guides still say so. The -fstack-clash-protection gap is now fixed rather than documented (see above), but the mitigation itself remains unavailable on OpenBSD — that is a toolchain property the build cannot work around.

[Copilot]

Pull request overview

This pull request improves BSD portability for Gridcoin’s CMake-based build and multiprocess IPC credential handling, and updates the OpenBSD/FreeBSD build guides to reflect the corrected build steps and multiprocess requirements.

Changes:

• Fix Boost discovery on platforms where CMake falls back to FindBoost by setting Boost_USE_DEBUG_RUNTIME=OFF only when the user hasn’t specified it.
• Fix OpenBSD multiprocess compilation by gating the Linux SO_PEERCRED/struct ucred path on __linux__ instead of SO_PEERCRED alone.
• Update OpenBSD/FreeBSD build documentation for multiprocess (Cap’n Proto), correct output paths, and add platform-specific notes.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
CMakeLists.txt	Sets Boost_USE_DEBUG_RUNTIME up-front (when unspecified) to avoid Boost variant rejection during module-mode FindBoost resolution.
src/ipc/peercred.cpp	Introduces a single Linux-only SO_PEERCRED gate macro to avoid OpenBSD’s incompatible SO_PEERCRED payload type.
src/ipc/peercred.h	Updates platform-coverage documentation to match the Linux-only SO_PEERCRED/ucred usage.
doc/build-openbsd.md	Documents multiprocess prerequisites and OpenBSD-specific build constraints; corrects run paths.
doc/build-freebsd.md	Updates guidance for Boost and adds multiprocess/Cap’n Proto notes; corrects run paths.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.