rand v0.9.5: backport #1790 - #1803
Merged
Merged
Conversation
Prevent memory safety violation in `UniformChar` via deserialization.
vks
approved these changes
Jul 10, 2026
This was referenced Jul 27, 2026
elvis-sik
added a commit
to baixada-cards/truco-solver
that referenced
this pull request
Jul 31, 2026
…e format (#14) Replaces #9 (Dependabot could only bump the versions; both bumps are breaking and needed an API migration). ## rand 0.8.7 → 0.9.5 - `rng.gen()` → `rng.random()`, `rng.gen::<bool>()` → `rng.random::<bool>()`, `rand::thread_rng()` → `rand::rng()`. `StdRng::seed_from_u64` is unchanged. - Security value: 0.9.5 includes the `UniformChar` deserialization memory-safety fix (rust-random/rand#1803). 0.8.7 is the last 0.8.x release, so there is no patched 0.8 — 0.9 is the only way to pick the fix up. - **Seeded reproducibility is preserved.** rand 0.8.7 and 0.9.5 `StdRng::seed_from_u64` produce bit-identical streams for the three draw shapes this crate uses (`next_u64`, `f64`, `bool`), verified across seeds 0/1/42/12345/u64::MAX. Previously recorded seeded `mccfr-bench` results stay reproducible. ### Note: rand 0.8.7 is still linked, via truco-engine `cargo tree -i rand@0.8.7 -e normal` shows it arriving as a **normal** dependency of the pinned `truco-engine` rev, not as a dev-only dependency — so solver binaries link both 0.8.7 and 0.9.5. That is sound (the two RNGs are never passed across the engine boundary; the workspace compiles and all tests pass), but it means the memory-safety fix is only adopted for this crate's own RNG use, not the engine's. Fully retiring rand 0.8 needs a coordinated bump: `truco-engine` and `truco-server` deliberately pin `rand < 0.9` in their Dependabot configs ("must move with bots/server consumers to keep one RNG trait ABI"), and the rand PRs in engine/server/bots were all closed unmerged. That multi-repo migration is out of scope here and should be tracked separately. ## bincode 1.3.3 → 2.0.1 bincode 2 removed the free functions (`serialize`, `serialize_into`, `deserialize`, `deserialize_from`) and its new default config is varint-encoded — a naive port would silently orphan every existing artifact (GCS checkpoints, treepacks, teacher sidecars, match-value tables). Instead, all bincode traffic now goes through a new `bincode_v1` module that wraps `bincode::serde` with `bincode::config::legacy()` (little-endian, fixed-width integers, no limit) — bincode 1's exact default wire format. Call sites are unchanged apart from the module path. ## Compatibility verification - **Scratch parity harness** (bincode 1.3.3 and 2.0.1 side by side): byte-identical output for representative shapes (data-carrying enums, tuples, `Option`, `String`, `Vec<f64>/<f32>`, fixed arrays, smallvec, sequential multi-value streams), cross-decode in both directions, trailing-bytes tolerance (the legacy-layout fallback decodes rely on it). - **Golden bytes in-tree**: `bincode_v1` tests assert against a byte vector produced by real bincode 1.3.3 pre-migration, so any future format drift fails CI. - **Real GCS artifacts** decoded with the migrated code: - `10x10-full-20260709/match_values.bin`: decoded and re-encoded **byte-identically** (3,042 bytes). - `10x10-full-20260709/.../tc0.d0.ckpt.bin` (6.7 GB, range-read prefix): `CheckpointMeta` decodes correctly (score (10,10), SyncCFR+, iter 910, 39,508,752 info sets, dealer_filter Some(0)) and 48,991 streamed entries decode with consistent action/regret/strategy lengths, all values finite. - `deep-probe-0x0-tc0-d0-20260721/checkpoint/deep-0x0-d0.ckpt` (15.7 GiB resume asset, range-read header): leading scalars decode to the known run config (0x0 tc0 d0, round 3 of 3, 140,118 deals, 3,135 subgames). `make check` passes (locks, fmt, clippy -D warnings, full workspace tests, autoresearch). 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Elvis Sikora <11654926+elvis-sik@users.noreply.github.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
CHANGELOG.mdentrySummary
Backport #1790 to rand 0.9.5.
Motivation
This changes address an unsoundness issue: code like the following could yield a
Distributionobject yielding invalidcharvalues when sampled (unsound):