math/rand (or crypto/rsa): non-deterministic output after go1.11.2

[lukateras]

What version of Go are you using (go version)?

# Both versions are affected:
$ go version
go version go1.11.2 linux/amd64
$ go version
go version go1.11.4 linux/amd64

Does this issue reproduce with the latest release?

Yes.

What operating system and processor architecture are you using (go env)?

go env Output

$ go env
GOARCH="amd64"
GOBIN=""
GOCACHE="/home/yegortimoshenko/.cache/go-build"
GOEXE=""
GOFLAGS=""
GOHOSTARCH="amd64"
GOHOSTOS="linux"
GOOS="linux"
GOPATH="/home/yegortimoshenko/.local/share/go:/nix/store/yh6p6ky7w2nfa64r3r7vhr5d5s1p87bk-go-1.11.2/share/go"
GOPROXY=""
GORACE=""
GOROOT="/nix/store/yh6p6ky7w2nfa64r3r7vhr5d5s1p87bk-go-1.11.2/share/go"
GOTMPDIR=""
GOTOOLDIR="/nix/store/yh6p6ky7w2nfa64r3r7vhr5d5s1p87bk-go-1.11.2/share/go/pkg/tool/linux_amd64"
GCCGO="gccgo"
CC="gcc"
CXX="g++"
CGO_ENABLED="1"
GOMOD=""
CGO_CFLAGS="-g -O2"
CGO_CPPFLAGS=""
CGO_CXXFLAGS="-g -O2"
CGO_FFLAGS="-g -O2"
CGO_LDFLAGS="-g -O2"
PKG_CONFIG="pkg-config"
GOGCCFLAGS="-fPIC -m64 -pthread -fmessage-length=0 -fdebug-prefix-map=/run/user/1000/go-build002404125=/tmp/go-build -gno-record-gcc-switches"

What did you do?

https://play.golang.org/p/v9ajjMHKx5M

What did you expect to see?

Consistent output between invocations, across all platforms.

What did you see instead?

When I run this program built locally with go1.11.2 or go1.11.4, I'm sometimes getting 1810a846745d4bc329c923c76786416d75dbd711eb369a824ebcff3078d22519, sometimes b2d4c5a6c6683cab81962ba61e71a2dec2869ce947078958413b686a0ca48c7f.

When I run this program built locally with go1.10.3, I'm always getting 1810a846745d4bc329c923c76786416d75dbd711eb369a824ebcff3078d22519.

When I run this program on play.golang.org, I'm always getting b2d4c5a6c6683cab81962ba61e71a2dec2869ce947078958413b686a0ca48c7f.

Additional notes

Creating a wrapped Source that locks and unlocks mutex on each Source.Int63() call doesn't resolve the issue: https://play.golang.org/p/JgJ4kM3UjY8

This is not specific to PRNG implementation, either. When I replace math/rand Source PRNG with PCG RNG, the issue still persists.

Some invocations seem to consume more RNG values, as seen in this example: https://play.golang.org/p/W1bTSIb5FYD (always 426 on play.golang.org, always 206 on go1.10.3, 206 or 426 on go1.11.2 and go1.11.4).

[magical]

I believe this is intentional. From the Go 1.11 release notes:

Certain crypto operations, including ecdsa.Sign, rsa.EncryptPKCS1v15 and rsa.GenerateKey, now randomly read an extra byte of randomness to ensure tests don't rely on internal behavior.

See also #21915

[lukateras]

You are right, thanks.