test: eliminate intermittent macOS gtest-discovery build failure

[gmarzot]

Problem

The macos job intermittently fails at build time with:

CMake Error at /opt/homebrew/share/cmake/Modules/GoogleTest/ParseTestList.cmake:96 (string):
  string sub-command JSON failed parsing json string:
  * Line 1, Column 1
    Syntax error: value, object or array expected.

(e.g. this run on #533 — passed on rerun, unrelated to the PR's changes.)

Root cause

CMake 4.4.0 (current Homebrew) has a regression in the GoogleTest module refactor: the generated *_discovery.cmake scripts omit TEST_TARGET, so the per-target hash that keys the discovery JSON file is computed from the empty string. Every test target in build/test therefore shares one file, cmake_test_discovery_e3b0c44298.json (sha256("")), defeating the race protection that hash exists for.

In default POST_BUILD mode, discovery runs right after each test binary links — and Ninja links many test binaries near-simultaneously (the failing run linked three within 0.2 s). Concurrent discoveries then race on the shared file: one truncates/removes it while another parses → empty JSON → the error above. Intermittent by nature; the losing target varies. Fixed upstream in cmake 4.4.1, which Homebrew doesn't ship yet.

Fix

Switch discovery to PRE_TEST mode (one set() before the gtest_discover_tests() calls). Discovery then runs serially at ctest startup — immune to the race on any cmake version.

Cost

• Build step: slightly faster — the ~28 post-link discovery executions disappear from the build.
• Test step: +0.8 s measured to enumerate all 772 tests on first ctest invocation; cached via an IS_NEWER_THAN guard thereafter (measured 0.12 s), re-discovering only binaries that actually relinked.

Validation (local, linux, cmake 4.2.3)

• Reconfigured + full build: discovery steps gone from the build graph; generated *_include.cmake files now carry the guarded gtest_discover_tests_impl call.
• ctest -N: all 772 tests enumerated.
• Targeted run of 306 tests: all pass. (UpstreamProviderTest network tests fail/hang on this dev box with or without the change — pre-existing local-env issue; CI is the gate for those.)

Note for #519: moqx_add_gtest() there still uses default POST_BUILD discovery, so this line should survive the rework (or move into the helper as DISCOVERY_MODE PRE_TEST).

---

This change is 

[michalhosna]

@michalhosna made 1 comment.
Reviewable status: 0 of 1 files reviewed, 1 unresolved discussion (waiting on gmarzot).

---

-- commits line 14 at r1:
Homebrew now ships 4.4.2 (See https://formulae.brew.sh/formula/cmake).

So, is this still a problem?

[afrind]

I'm ok with this, but maybe it's not actually going to fix it?

@afrind reviewed 1 file and all commit messages, and made 2 comments.
Reviewable status: all files reviewed, 1 unresolved discussion (waiting on michalhosna).

---

-- commits line 14 at r1:

Previously, michalhosna (Michal Hošna) wrote…

Homebrew now ships 4.4.2 (See https://formulae.brew.sh/formula/cmake).

So, is this still a problem?

What's on our CI runner?

[gmarzot]

I'm ok with this, but maybe it's not actually going to fix it?

@afrind reviewed 1 file and all commit messages, and made 2 comments.
Reviewable status: all files reviewed, 1 unresolved discussion (waiting on michalhosna).

-- commits line 14 at r1:

Previously, michalhosna (Michal Hošna) wrote…

Homebrew now ships 4.4.2 (See https://formulae.brew.sh/formula/cmake).
So, is this still a problem?

What's on our CI runner?

I am reading now that this problem will fade out in a few weeks. the runners are mixed and the bad version can still crop up.

the restructure is slightly better even after 4.4.2 roles out... conserves runner cycles.. but doesnt really translate into any observable build perf benefit ~1sec

we can close or merge.. you guys decide? eh i see a check mark.. will merge

[gmarzot]

we should not see it again w/ this change and some minor optimization bonus

[gmarzot]

@gmarzot made 1 comment.
Reviewable status: all files reviewed, 1 unresolved discussion (waiting on afrind and michalhosna).

[gmarzot]

@gmarzot resolved 1 discussion.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on gmarzot).