Go modules whose own _test.go suites go2cs converted with go2cs -tests -recurse,
built against the converted standard library, ran under the Go-semantics test host, and compared
verdict for verdict against a clean go test -json baseline of the same sources — the same bar the
standard-library roster holds every package to. A module appears here
only when, in every package the run reached, each test either matched go test or is disclosed, by its exact
failure, as something the converted code does not do.
Each module's pages are written by the converter beside its conversion and copied here unedited:
the module summary sums the per-package proof pages, and each proof page lists every verdict. Every
module's summary is the VALIDATION.md its NuGet package carries, below the package's notice.
| Module | Version | Packages | Matched | Disclosed | Platform | Proof |
|---|---|---|---|---|---|---|
golang.org/x/sync |
v0.19.0 | 4 | 27 | 1 | linux/amd64 |
summary |
github.com/google/uuid |
v1.6.0 | 1 | 54 | 0 | linux/amd64 |
summary |
github.com/golang-jwt/jwt/v5 |
v5.3.1 | 2 | 189 | 0 | linux/amd64 |
summary |
github.com/ritchiecarroll/hashset |
v1.0.0 | 1 | 37 | 0 | linux/amd64 |
summary |
github.com/spf13/pflag |
v1.0.9 | 1 | 177 | 0 | linux/amd64 |
summary |
Measured with Go 1.24.13 at Release with the tiered JIT off, the test host's default.
golang.org/x/sync: semaphore's TestWeightedAcquire gives each acquire attempt a 10 ms deadline. With
the tiered JIT off, the first attempt compiles the acquire path at full optimization before it can wait, which
on the linux/amd64 machine measured takes longer than 10 ms, so the test fails there and is disclosed as
runtime-capability; the semaphore proof page pins the exact failure. With the tiered JIT on (-test-tiered),
the same machine matches all 8 of semaphore's tests and all 28 of the module's. On windows/amd64, two runs on
their own each matched all 28 verdicts, and a third, made while the machine was running other heavy test jobs,
failed TestWeightedAcquire once; those proof pages are not reproduced here. That test depends on timing, so the
result above is claimed only for a machine that is not busy with other work.
github.com/google/uuid and github.com/golang-jwt/jwt/v5: the proof pages linked above are the linux/amd64
readings their packages carry. Each module was also read on windows/amd64 from the same module sources (their
module sums match), with the same counts and no test differing; those proof pages are not reproduced here.
uuid's TestClockSeqRace skips on both sides. uuid's TestVersion6 has an upstream race that can fail it on
either runtime; its proof page cites the race without disclosing it, and the proof is a run in which both sides
passed. jwt/v5 also holds cmd/jwt and test, which have no tests of their own.