Run and filter tests

Options go after --, exactly as with silly:

dub test :base                       # run everything, in parallel
dub test :base -- -i "Buffer"        # only tests matching a regex
dub test :base -- -e "slow"          # skip tests matching a regex
dub test :base -- -v                 # durations, locations, full traces
dub test :base -- -t 1               # single-threaded (main thread, 8 MiB stack)
dub test :base -- -l                 # list tests without running them

Filtering

-i/-e are regular expressions matched against <fully.qualified.symbol> <test name> — so both -i "buffer" (module) and -i "Buffer.basic" (name UDA) work. When both are given they combine: a test must match -i and not match -e.

Listing

--list prints every discovered test with its special-handling markers:

$ dub test :test-runner -- -l
 sparkles.test_runner.discovery discovery.selfTest @betterC
 sparkles.test_runner.discovery discovery.moduleTests
 sparkles.test_runner.discovery.NestedCtfeHost discovery.nestedCtfe @ctfe

Verbose output and source links

-v appends per-test durations and [file:line] locations. On a terminal, locations are OSC 8 hyperlinks (file:// URIs) when the tested package has sparkles:core-cli in its dependency closure.

Worker stacks

-t 1 runs on the process main thread (~8 MiB on every OS we ship). Any -t N with N > 1 pins each worker to 512 KiB, matching macOS's default pthread stack — Linux's 8 MiB default would otherwise hide a megabyte-sized stack local that SIGSEGVs on a developer Mac. Crossing a 384 KiB watermark fails that test with stack budget exceeded and the suite continues. Put large scratch on the heap (Unique, HeapBuffer); do not T x; a workspace in a unittest.

ci --test runs each package twice after one compile: -t 1, then -t N with N = max(2, hwParallelism()).

The runner's own tests

The runner is compiled into every in-tree test build, so its self-tests are discoverable everywhere; they are hidden unless you ask:

dub test :test-runner        # the runner's own suite (runs by default here)
dub test :base -- --self-test # base's tests + the runner's, in one process

Exit status

The test binary exits non-zero when any test fails, in every mode — safe for CI and git bisect run. A failing run reprints every failed test (name, location, throwable, stack trace) after the summary, so a CI log's tail names the failures even when hundreds of passing lines came first.