/ crates / tor-basic-utils / src / test_rng.rs
test_rng.rs
  1  //! Code for deterministic and/or reproducible use of PRNGs in tests.
  2  //!
  3  //! Often in testing we want to test a random scenario, but we want to be sure
  4  //! of our ability to reproduce the scenario if the test fails.
  5  //!
  6  //! To achieve this,  just have your test use [`testing_rng()`] in place of
  7  //! [`rand::thread_rng()`].  Then the test will (by default) choose a new random
  8  //! seed for every run, and print that seed to standard output.  If the test
  9  //! fails, the seed will be displayed as part of the failure message, and you
 10  //! will be able to use it to recreate the same PRNG seed as the one that caused
 11  //! the failure.
 12  //!
 13  //! If you're running your tests in a situation where deterministic behavior is
 14  //! key, you can also enable this via the environment.
 15  //!
 16  //! The run-time behavior is controlled using the `ARTI_TEST_PRNG` variable; you
 17  //! can set it to any of the following:
 18  //!   * `random` for a randomly seeded PRNG. (This is the default).
 19  //!   * `deterministic` for an arbitrary seed that is the same on every run of
 20  //!     the program. (You can use this in cases where even a tiny chance of
 21  //!     stochastic behavior in your tests is unacceptable.)
 22  //!   * A hexadecimal string, to specify a given seed to reuse from a previous
 23  //!     test run.
 24  //!
 25  //! # WARNING
 26  //!
 27  //! This is for testing only!  Never ever use it in non-testing code.  Doing so
 28  //! may compromise your security.
 29  //!
 30  //! You may wish to use clippy's `disallowed-methods` lint to ensure you aren't
 31  //! using it outside of your tests.
 32  //!
 33  //! # Examples
 34  //!
 35  //! Here's a simple example of a test that verifies that integer sorting works
 36  //! correctly by shuffling a short sequence and then re-sorting it.
 37  //!
 38  //! ```
 39  //! use tor_basic_utils::test_rng::testing_rng;
 40  //! use rand::{seq::SliceRandom};
 41  //! let mut rng = testing_rng();
 42  //!
 43  //! let mut v = vec![-10, -3, 0, 1, 2, 3];
 44  //! v.shuffle(&mut rng);
 45  //! v.sort();
 46  //! assert_eq!(&v, &[-10, -3, 0, 1, 2, 3])
 47  //! ```
 48  //!
 49  //! Here's a trickier example of how you might write a test to override the
 50  //! default behavior.  (For example, you might want to do this if the test is
 51  //! unreliable and you don't have time to hunt down the issues.)
 52  //!
 53  //! ```
 54  //! use tor_basic_utils::test_rng::Config;
 55  //! let mut rng = Config::from_env()
 56  //!     .unwrap_or(Config::Deterministic)
 57  //!     .into_rng();
 58  //! ```
 59  
 60  // We allow printing to stdout and stderr in this module, since it's intended to
 61  // be used by tests, where this is the preferred means of communication with the user.
 62  #![allow(clippy::print_stdout, clippy::print_stderr)]
 63  
 64  use rand::{RngCore, SeedableRng};
 65  // We'll use the same PRNG as the (current) standard.  We specify it here rather
 66  // than using StdRng, since we want determinism in the future.
 67  pub use rand_chacha::ChaCha12Rng as TestingRng;
 68  
 69  /// The seed type for the RNG we're returning.
 70  type Seed = <TestingRng as SeedableRng>::Seed;
 71  
 72  /// Default seed for deterministic RNG usage.
 73  ///
 74  /// This is the seed we use when we're told to use a deterministic RNG with no
 75  /// specific seed.
 76  const DEFAULT_SEED: Seed = *b"4   // chosen by fair dice roll.";
 77  
 78  /// The environment variable that we inspect.
 79  const PRNG_VAR: &str = "ARTI_TEST_PRNG";
 80  
 81  /// Return a new, possibly deterministic, RNG for use in tests.
 82  ///
 83  /// This function is **only** for testing: using it elsewhere may make your code
 84  /// insecure!
 85  ///
 86  /// The type of this RNG will depend on the value of `ARTI_TEST_PRNG`:
 87  ///   * If ARTI_TEST_PRNG is `random` or unset, we'll use a real seeded PRNG.
 88  ///   * If ARTI_TEST_PRNG is `deterministic`, we'll use a standard canned PRNG
 89  ///     seed.
 90  ///   * If ARTI_TEST_PRNG is a hexadecimal string, we'll use that as the PRNG
 91  ///     seed.
 92  ///
 93  /// We'll print the value of this RNG seed to stdout, so that if the test fails,
 94  /// you'll know what seed to use in reproducing it.
 95  ///
 96  /// # Panics
 97  ///
 98  /// Panics if the environment variable is set to an invalid value.
 99  ///
100  /// (If your code must not panic, then it is not test code, and you should not
101  /// be using this function.)
102  pub fn testing_rng() -> TestingRng {
103      // Somewhat controversially, we prefer a Random prng by default.  Our
104      // rationale is that, if this weren't the default, nobody would ever set it,
105      // and we'd never find out about busted tests or code.
106      Config::from_env().unwrap_or(Config::Random).into_rng()
107  }
108  
109  /// Type describing a testing_rng configuration.
110  ///
111  /// This is a separate type so that you can pick different defaults, or inspect
112  /// the configuration before using it.
113  #[derive(Debug, Clone, Copy, Eq, PartialEq)]
114  #[non_exhaustive]
115  pub enum Config {
116      /// Use a PRNG with a randomly chosen seed.
117      Random,
118      /// Use a PRNG with a (default) pre-selected seed.
119      Deterministic,
120      /// Use a specific seed value for the PRNG.
121      Seeded(Seed),
122  }
123  
124  impl Config {
125      /// Return the testing PRNG from the environment, if one is configured.
126      ///
127      /// # Panics
128      ///
129      /// Panics if the environment variable is set to an invalid value.
130      ///
131      /// (If your code must not panic, then it is not test code, and you should not
132      /// be using this function.)
133      pub fn from_env() -> Option<Self> {
134          match Self::from_env_result(std::env::var(PRNG_VAR)) {
135              Ok(c) => c,
136              Err(e) => {
137                  panic!(
138                      "Bad value for {}: {}\n\
139                      We recognize `random`, `deterministic`, or a hexadecimal seed.",
140                      PRNG_VAR, e
141                  );
142              }
143          }
144      }
145  
146      /// Read the configuration from the result of `std::env::var()`.
147      ///
148      /// Return None if there was no option.
149      fn from_env_result(var: Result<String, std::env::VarError>) -> Result<Option<Self>, Error> {
150          match var {
151              Ok(s) if s.is_empty() => Ok(None),
152              Ok(s) => Ok(Some(Config::from_str(&s)?)),
153              Err(std::env::VarError::NotPresent) => Ok(None),
154              Err(std::env::VarError::NotUnicode(_)) => Err(Error::InvalidUnicode),
155          }
156      }
157  
158      /// Read the configuration from a provided string.
159      ///
160      /// The string format is as described in [`testing_rng`].
161      ///
162      /// Return None if this string can't be interpreted as a [`Config`]
163      fn from_str(s: &str) -> Result<Self, Error> {
164          Ok(if s == "random" {
165              Self::Random
166          } else if s == "deterministic" {
167              Self::Deterministic
168          } else if let Some(seed) = decode_seed_bytes(s) {
169              Self::Seeded(seed)
170          } else {
171              return Err(Error::UnrecognizedValue(s.to_string()));
172          })
173      }
174  
175      /// Consume this `Config` and return a `Seed`.
176      fn into_seed(self) -> Seed {
177          match self {
178              Config::Deterministic => DEFAULT_SEED,
179              Config::Seeded(seed) => seed,
180              Config::Random => {
181                  let mut seed = Seed::default();
182                  rand::thread_rng().fill_bytes(&mut seed[..]);
183                  seed
184              }
185          }
186      }
187  
188      /// Consume this `Config` and return a `TestingRng`.
189      pub fn into_rng(self) -> TestingRng {
190          let seed = self.into_seed();
191          println!("  Using RNG seed {}={}", PRNG_VAR, format_seed_bytes(&seed));
192          TestingRng::from_seed(seed)
193      }
194  }
195  
196  /// Format `seed` in the format expected by [`decode_seed_bytes`].
197  ///
198  /// This is a separate function to make it clearer what the tests are testing.
199  fn format_seed_bytes(seed: &Seed) -> String {
200      hex::encode(seed)
201  }
202  
203  /// Try to see whether a literal seed can be decoded from a given string.  If
204  /// so, return it.
205  ///
206  /// We currently use a hex encoding, truncating or zero-extending the provided
207  /// seed as needed.
208  fn decode_seed_bytes(s: &str) -> Option<Seed> {
209      if s.is_empty() {
210          // Do not accept the empty string.
211          return None;
212      }
213      let bytes = hex::decode(s).ok()?;
214      let mut seed = Seed::default();
215      let n = std::cmp::min(seed.len(), bytes.len());
216      seed[..n].copy_from_slice(&bytes[..n]);
217      Some(seed)
218  }
219  
220  /// An error from trying to decode a [`Config`] from a string.
221  #[derive(Clone, Debug, thiserror::Error, Eq, PartialEq)]
222  enum Error {
223      /// We got a value that wasn't unicode.
224      #[error("Value was not UTF-8")]
225      InvalidUnicode,
226      /// We got a value that we otherwise couldn't decode.
227      #[error("Could not interpret {0:?} as a PRNG seed.")]
228      UnrecognizedValue(String),
229  }
230  
231  #[cfg(test)]
232  mod test {
233      // @@ begin test lint list maintained by maint/add_warning @@
234      #![allow(clippy::bool_assert_comparison)]
235      #![allow(clippy::clone_on_copy)]
236      #![allow(clippy::dbg_macro)]
237      #![allow(clippy::mixed_attributes_style)]
238      #![allow(clippy::print_stderr)]
239      #![allow(clippy::print_stdout)]
240      #![allow(clippy::single_char_pattern)]
241      #![allow(clippy::unwrap_used)]
242      #![allow(clippy::unchecked_duration_subtraction)]
243      #![allow(clippy::useless_vec)]
244      #![allow(clippy::needless_pass_by_value)]
245      //! <!-- @@ end test lint list maintained by maint/add_warning @@ -->
246      use std::env::VarError;
247  
248      use super::*;
249  
250      #[test]
251      fn from_str() {
252          assert_eq!(Ok(Config::Deterministic), Config::from_str("deterministic"));
253          assert_eq!(Ok(Config::Random), Config::from_str("random"));
254          assert_eq!(Ok(Config::Seeded([0x00; 32])), Config::from_str("00"));
255          {
256              let s = "aaaaaaaa";
257              let seed = [
258                  0xaa, 0xaa, 0xaa, 0xaa, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
259                  0, 0, 0, 0, 0, 0, 0, 0,
260              ];
261              assert_eq!(Ok(Config::Seeded(seed)), Config::from_str(s));
262          }
263          {
264              let seed = *b"hello world. this is a longer st";
265              let mut s = hex::encode(seed);
266              assert_eq!(Ok(Config::Seeded(seed)), Config::from_str(&s));
267              // we can make it longer, and it just gets truncated.
268              s.push_str("aabbccddeeff");
269              assert_eq!(Ok(Config::Seeded(seed)), Config::from_str(&s));
270          }
271  
272          assert_eq!(
273              Err(Error::UnrecognizedValue("".to_string())),
274              Config::from_str("")
275          );
276  
277          assert_eq!(
278              Err(Error::UnrecognizedValue("return 4".to_string())),
279              Config::from_str("return 4")
280          );
281      }
282  
283      #[test]
284      fn from_env() {
285          assert_eq!(
286              Ok(Some(Config::Deterministic)),
287              Config::from_env_result(Ok("deterministic".to_string()))
288          );
289          assert_eq!(
290              Ok(Some(Config::Random)),
291              Config::from_env_result(Ok("random".to_string()))
292          );
293          assert_eq!(
294              Ok(Some(Config::Seeded([0xcd; 32]))),
295              Config::from_env_result(Ok("cd".repeat(32)))
296          );
297          assert_eq!(Ok(None), Config::from_env_result(Ok("".to_string())));
298          assert_eq!(Ok(None), Config::from_env_result(Err(VarError::NotPresent)));
299          assert_eq!(
300              Err(Error::InvalidUnicode),
301              Config::from_env_result(Err(VarError::NotUnicode("3".into())))
302          );
303          assert_eq!(
304              Err(Error::UnrecognizedValue("123".to_string())),
305              Config::from_env_result(Ok("123".to_string()))
306          );
307      }
308  
309      #[test]
310      fn make_seed() {
311          assert_eq!(Config::Deterministic.into_seed(), DEFAULT_SEED);
312          assert_eq!(Config::Seeded([0x24; 32]).into_seed(), [0x24; 32]);
313  
314          let s1 = Config::Random.into_seed();
315          let s2 = Config::Random.into_seed();
316          assert_ne!(s1, s2);
317      }
318  
319      #[test]
320      fn code_decode() {
321          assert_eq!(
322              decode_seed_bytes(&format_seed_bytes(&DEFAULT_SEED)).unwrap(),
323              DEFAULT_SEED
324          );
325      }
326  
327      #[test]
328      fn determinism() {
329          let mut d_rng = Config::Deterministic.into_rng();
330          let values: Vec<_> = std::iter::repeat_with(|| d_rng.next_u32())
331              .take(8)
332              .collect();
333  
334          // This should be the same every time.
335          let deterministic_values = vec![
336              4222362647, 2976626662, 1407369338, 1087750672, 196711223, 996083910, 836259566,
337              2589890951,
338          ];
339          assert_eq!(values, deterministic_values);
340  
341          // But if we use a random RNG, we'll get different values
342          // (with P=1-2^-256)
343          let mut r_rng = Config::Random.into_rng();
344          let values: Vec<_> = std::iter::repeat_with(|| r_rng.next_u32())
345              .take(8)
346              .collect();
347          assert_ne!(values, deterministic_values);
348      }
349  }