chore(deps)!: take the major-version wave, fixing what it exposed by doublegate · Pull Request #8 · doublegate/ProRT-IP

doublegate · 2026-08-29T00:45:36Z

Supersedes #3 and #4. Those are the same wave from two directions — the "fuzz-deps" group turned out to rewrite the root workspace manifest, because fuzz/ depends on the workspace by path. Neither was safe to merge as proposed.

Two of the riskiest majors needed deleting, not migrating

etherparse 0.15 → 0.21 and socket2 0.5 → 0.6 were the two large API rewrites in the group. Both are declared in three manifests each and imported by zero source files — the same pattern as quick-xml and tui-input. Removed.

The real migration was `rand`

0.8 → 0.9: 24 thread_rng, 17 gen/gen::<T>, 16 gen_range sites across 10 files → rng, random, random_range. All were deprecations, not removals, so cargo build stayed green and only -D warnings surfaced them.

`sysinfo` 0.38, not 0.39

0.39 declares rust-version = "1.95" — above this workspace's 1.88 floor, and directly reachable from prtip-core. A local toolchain newer than the MSRV compiles it happily; only checking resolved metadata against the declared floor catches it. Verified after: 0 of 502 resolved packages require > 1.88.

Also colored 3.1, dirs 6.0, governor 0.10, ipnetwork 0.21 (serde moved behind a feature — now explicit), mlua 0.12, nix 0.31, rlimit 0.11, thiserror 2.0, toml 1.1, windows 0.62, x509-parser 0.18. dirs/toml/ipnetwork had crate-local pins outside the workspace table and were each built twice.

What the wave exposed

The ipnetwork bump broke one test. Following it down found three real defects unrelated to dependency versions.

Unbounded expansion — the serious one

expand_hosts() was network.iter().collect() with no limit, so prtip -sT -p 80 0.0.0.0/0 — eight characters of user input — tried to allocate 4.3 billion addresses, ~68 GB.

Measured before and after:

	behaviour
before	banner printed, then hung until SIGKILL (exit 137)
after	exit 1 immediately: `Invalid target: 0.0.0.0/0 contains 4294967296 addresses, above the 16777216 limit for host expansion`

Earlier releases were protected only by accident — expanding a /0 overflowed an integer and panicked fast. ipnetwork 0.21 removed the overflow, turning a loud failure into a silent hang. The bound is the check that panic was standing in for, and is what the test's own note — "Future enhancement: should validate CIDR size before expansion" — had been asking for. expand_hosts() now returns Result; 12 call sites updated.

`host_count()` returned 0 for a `/31`

RFC 3021 makes both addresses of a point-to-point /31 usable, but the code subtracted network and broadcast unconditionally below /32. Latent while nothing trusted the number — it became Scanner error: Result queue full (0/0) the moment the scheduler sized its queue from it.

`host_count()` and `expand_hosts()` measured different things

Usable hosts vs. every address — so a /30 counted 2 and scanned 4. Split into address_count() (what expansion yields; what queues size from) and host_count() (usable hosts; what a user sees), with a test pinning address_count() to the actual expansion length across five prefixes. My own new test caught this — the /31 fix alone would have left the /30 case broken.

Two optimizations from the same reading

scheduler.rs computed targets.iter().map(|t| t.expand_hosts().len()).sum() — materialising every address purely to count it. Now O(1).
ScanTarget::first_host() added. Four decoy_scanner.rs sites expanded a whole target to read hosts[0], so a /8 allocated 268 MB for one address. It's iter().next(), and works even on targets too large to expand.

Test harness

The CLI helper picked the release binary with release_path.exists() — true for a directory. A docker run bind-mount of target/release/prtip with no release build present makes Docker create the mount source as a root-owned directory, after which all 52 CLI integration tests failed with a baffling PermissionDenied instead of falling back to the working debug binary. Now is_file().

Verification

cargo test --workspace       2,587 passed, 0 failed, 121 ignored (exit 0)
cargo clippy --workspace --all-targets --locked -- -D warnings   exit 0
cargo fmt --all -- --check   exit 0
cargo audit                  exit 0
cargo deny check             exit 0  advisories, licenses, bans, sources
both generators --check      exit 0
resolved MSRV                0 of 502 packages require > 1.88

2,581 → 2,587; the +6 are regression tests for the defects above, including one asserting /0 is refused and one that first_host() works on a target expand_hosts() rejects.

🤖 Generated with Claude Code

https://claude.ai/code/session_01NBDCJvLbq7nuor7RtT57rD

Supersedes Dependabot #3 and #4. Those are the same wave from two directions: the "fuzz-deps" group turned out to rewrite the root workspace manifest, because `fuzz/` depends on the workspace by path. Neither was safe to merge as proposed. ## Two of the riskiest majors needed deleting, not migrating `etherparse` 0.15 -> 0.21 and `socket2` 0.5 -> 0.6 were the two large API rewrites in the group. Both crates are declared in three manifests each and imported by **zero** source files -- the same pattern as `quick-xml` and `tui-input` earlier on this branch. Removed. ## The real migration was rand `rand` 0.8 -> 0.9: 24 `thread_rng`, 17 `gen`/`gen::<T>`, and 16 `gen_range` call sites across 10 files, renamed to `rng`, `random`, `random_range`. Every one was a deprecation rather than a removal, so `cargo build` stayed green throughout and only `-D warnings` surfaced them. ## sysinfo 0.38, not 0.39 0.39 declares `rust-version = "1.95"`, well above this workspace's 1.88 floor, and it is directly reachable from `prtip-core`. A local toolchain newer than the MSRV compiles it without complaint; only checking resolved metadata against the declared floor catches it. 0.38.4 is the newest line that fits and carries the same CPU API. `refresh_cpu()`/`global_cpu_info()` -> `refresh_cpu_usage()`/ `global_cpu_usage()`. Verified after: 0 of 502 resolved packages require more than 1.88. Also `colored` 3.1, `dirs` 6.0, `governor` 0.10, `ipnetwork` 0.21 (serde moved behind a feature; `ScanTarget` derives over `IpNetwork`, so it is now explicit), `mlua` 0.12, `nix` 0.31, `rlimit` 0.11, `thiserror` 2.0, `toml` 1.1, `windows` 0.62, `x509-parser` 0.18. `dirs`, `toml` and `ipnetwork` had crate-local pins outside the workspace table and were each being built twice. ## What the wave exposed The `ipnetwork` bump broke one test, and following it down found three real defects that had nothing to do with dependency versions. **Unbounded expansion (the serious one).** `expand_hosts()` was `network.iter().collect()` with no limit, so `prtip -sT -p 80 0.0.0.0/0` tried to allocate 4.3 billion addresses -- about 68 GB -- from eight characters of user input. Measured before and after: the old build printed its banner and hung until SIGKILL (exit 137); it now exits 1 immediately with Invalid target: 0.0.0.0/0 contains 4294967296 addresses, above the 16777216 limit for host expansion. Earlier releases were protected only by accident -- expanding a `/0` overflowed an integer and panicked fast. `ipnetwork` 0.21 removed the overflow and turned a loud failure into a silent hang. The bound is the check that panic was standing in for, and is exactly what the test's own note ("Future enhancement: should validate CIDR size before expansion") had been asking for. `expand_hosts()` now returns `Result`; 12 call sites updated. **`host_count()` returned 0 for a /31.** RFC 3021 makes both addresses of a point-to-point /31 usable, but the code subtracted network and broadcast unconditionally below /32. Latent while nothing trusted the number; it became `Scanner error: Result queue full (0/0)` the moment the scheduler sized its queue from it. **`host_count()` and `expand_hosts()` measured different things.** Usable hosts versus every address -- so a /30 counted 2 and scanned 4. Split into `address_count()` (what expansion yields, what queues size from) and `host_count()` (usable hosts, what a user is shown), with a test pinning `address_count()` to the actual expansion length across five prefixes. My own test caught this; the /31 fix alone would have left the /30 case broken. ## Two optimizations from the same reading - `scheduler.rs` computed `targets.iter().map(|t| t.expand_hosts().len()).sum()` -- materialising every address purely to count it. Now O(1) arithmetic. - `ScanTarget::first_host()` added. Four call sites in `decoy_scanner.rs` expanded an entire target to read `hosts[0]`, so a /8 allocated 268 MB to obtain one address. It is `iter().next()`, and works even on targets too large to expand. ## Test harness The CLI test helper selected the release binary with `release_path.exists()`, true for a *directory*. `docker run -v "$PWD/target/release/prtip:/prtip"` with no release build present makes Docker create the mount source as a root-owned directory, after which all 52 CLI integration tests failed with a baffling `PermissionDenied` rather than falling back to the working debug binary. Now `is_file()`. ## Verification cargo test --workspace 2,587 passed, 0 failed, 121 ignored (exit 0) cargo clippy --workspace --all-targets --locked -- -D warnings exit 0 cargo fmt --all -- --check exit 0 cargo audit exit 0 cargo deny check exit 0 advisories, licenses, bans, sources both generators --check exit 0 resolved MSRV 0 of 502 packages require > 1.88 2,581 -> 2,587; the +6 are regression tests for the defects above, including one asserting `/0` is refused and one that `first_host()` works on a target `expand_hosts()` rejects. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NBDCJvLbq7nuor7RtT57rD

github-actions · 2026-08-29T00:45:44Z

🤖 Hi @doublegate, I've received your request, and I'm working on it now! You can track my progress in the logs for more details.

coderabbitai · 2026-08-29T00:45:45Z

Important

🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: cf006575-636f-4417-b3ed-6ecbd6e7d587

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

_{Comment @coderabbitai help to get the list of available commands.}

github-actions · 2026-08-29T00:46:16Z

🤖 I'm sorry @doublegate, but I was unable to process your request. Please see the logs for more details.

Copilot

🟡 Changes recommended

IPv6 address_count() can overflow for /64 and potentially bypass expansion limits, and scheduler sizing should validate/saturate counts to avoid overflow/panic before rejecting oversized targets.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

This PR consolidates a large dependency-major upgrade wave across the Rust workspace and addresses multiple correctness and reliability issues that surfaced as a result (notably around target expansion/counting and test harness behavior), while updating call sites to rand/sysinfo API changes.

Changes:

Upgrade and de-duplicate workspace dependencies (e.g., rand 0.9, ipnetwork 0.21 w/ serde feature, sysinfo 0.38, toml 1.1), removing unused majors (socket2, etherparse) and folding crate-local pins into workspace versions.
Make ScanTarget counting/expansion safer and more explicit (address_count, corrected host_count semantics, bounded+fallible expand_hosts, and first_host to avoid full materialization).
Update scanners/tests to the new APIs and improved target semantics; fix CLI integration-test binary selection (is_file()).

File summaries

File	Description
crates/prtip-tui/Cargo.toml	Switch `ipnetwork` to workspace dependency to avoid duplicate versions.
crates/prtip-scanner/tests/integration_scanner.rs	Update target expansion test to handle fallible `expand_hosts()`.
crates/prtip-scanner/tests/common/error_injection.rs	Migrate probabilistic failure injection to `rand` 0.9 APIs.
crates/prtip-scanner/src/udp_scanner.rs	Update source-port randomization to `rand` 0.9 APIs.
crates/prtip-scanner/src/timing.rs	Update jitter generation to `rand` 0.9 APIs.
crates/prtip-scanner/src/tcp_connect.rs	Propagate `expand_hosts()` failures via `?` instead of assuming infallible expansion.
crates/prtip-scanner/src/syn_scanner.rs	Update RNG usage (`rng/random/random_range`) and sequence generation.
crates/prtip-scanner/src/stealth_scanner.rs	Update RNG usage for source ports / sequence numbers.
crates/prtip-scanner/src/scheduler.rs	Stop allocating expansions purely to count; use `address_count()` for sizing.
crates/prtip-scanner/src/idle/idle_scanner.rs	Update spoofed SYN randomization to `rand` 0.9 APIs.
crates/prtip-scanner/src/decoy_scanner.rs	Avoid full expansion when only first address is needed; update RNG APIs.
crates/prtip-scanner/Cargo.toml	Remove unused deps (`socket2`, `etherparse`); bump `x509-parser`; use workspace `toml`.
crates/prtip-network/src/packet_builder.rs	Update packet builder random defaults to `rand` 0.9 APIs.
crates/prtip-network/src/ipv6_packet.rs	Update IPv6 fragment ID generation to `rand` 0.9 APIs.
crates/prtip-network/src/fragmentation.rs	Update fragment ID generation to `rand` 0.9 APIs.
crates/prtip-network/Cargo.toml	Remove unused deps (`socket2`, `etherparse`).
crates/prtip-core/tests/integration.rs	Update expansion test to handle fallible `expand_hosts()`.
crates/prtip-core/src/types.rs	Add `address_count()`, fix `/31` host counting, add bounded/fallible `expand_hosts()`, add `first_host()`, and add regression tests.
crates/prtip-core/src/retry.rs	Update retry jitter to `rand` 0.9 APIs.
crates/prtip-core/src/resource_monitor.rs	Update CPU refresh/reading to `sysinfo` 0.38 APIs.
crates/prtip-core/Cargo.toml	Switch `dirs` to workspace dependency.
crates/prtip-cli/tests/test_edge_cases.rs	Update `/0` behavior test commentary to reflect bounded expansion behavior.
crates/prtip-cli/tests/common/mod.rs	Fix binary selection to require an actual file (`is_file()`) for release/debug paths.
crates/prtip-cli/Cargo.toml	Switch `dirs` to workspace dependency.
CHANGELOG.md	Document the dependency wave and the exposed/fixed defects and safeguards.
Cargo.toml	Update workspace dependency versions/features; remove unused deps; pin `sysinfo` below MSRV-breaking line.
Cargo.lock	Refresh lockfile for upgraded majors, removals, and dependency graph changes.

Review details

Suppressed comments (3)

crates/prtip-core/src/types.rs:97

address_count() can overflow for IPv6 prefixes like /64 (2^64), which may wrap/panic and then let expand_hosts() attempt to collect an effectively unbounded iterator. Use checked_pow and clamp to u64::MAX (or otherwise signal “too large”) to keep the count monotonic and prevent bypassing the expansion limit.

This issue also appears on line 167 of the same file.

            IpNetwork::V6(net) => {
                let prefix = net.prefix();
                if prefix >= 64 {
                    2u64.pow((128 - prefix) as u32)
                } else {

crates/prtip-core/src/types.rs:168

The expand_hosts() error message says “a /8 is the widest accepted”, which is only meaningful for IPv4; for IPv6 the widest accepted prefix would be much narrower. Consider making the guidance version-agnostic to avoid misleading users.

                "{} contains {} addresses, above the {} limit for host expansion. \
                 Narrow the prefix (a /8 is the widest accepted) or split the scan.",

crates/prtip-core/src/types.rs:1085

test_expand_hosts_limit_boundary uses host_count() to test the expansion cap, but the cap is defined in terms of allocated addresses (what expand_hosts() yields). Using host_count() makes the boundary check off by 2 for IPv4 networks (network/broadcast), weakening the regression coverage.

        // A /8 is exactly at the cap and must still be permitted; a /7 is over.
        let at_limit = ScanTarget::parse("10.0.0.0/8").unwrap();
        assert!(at_limit.host_count() <= ScanTarget::MAX_EXPANDABLE_HOSTS);

        let over_limit = ScanTarget::parse("10.0.0.0/7").unwrap();
        assert!(over_limit.host_count() > ScanTarget::MAX_EXPANDABLE_HOSTS);

Files reviewed: 26/27 changed files
Comments generated: 2
Review effort level: Lite

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

+        // These two must never disagree: the scheduler sizes its result queue
+        // from host_count() and then scans expand_hosts(). They diverged for
+        // /31, where host_count() returned 0 while expansion produced 2.


+        // address_count(), not host_count(): this sizes a queue for what
+        // expand_hosts() actually yields, which includes the network and
+        // broadcast addresses. host_count() reports usable hosts and would
+        // undersize the queue by 2 per target.
+        let estimated_hosts: usize = targets.iter().map(|t| t.address_count() as usize).sum();


`Test (windows-latest)` failed on this branch while the Linux suite was green: the `windows` 0.52 -> 0.62 bump moved `BOOL` out of `Win32::Foundation`. error[E0432]: unresolved import `windows::Win32::Foundation::BOOL` --> crates/prtip-network/src/privilege.rs:160:9 | no `BOOL` in `Win32::Foundation` Reproduced locally with `cargo check --target x86_64-pc-windows-gnu -p prtip-network` rather than waiting on CI; `prtip-network` now checks clean for that target. Both call sites -- `privilege.rs` and the privilege test -- import from `windows::core` instead. A whole-workspace Windows check still stops at `aws-lc-sys`, which needs a Windows C toolchain this machine does not have. That is a local cross-compilation limit, not a defect: CI builds natively on windows-latest with MSVC. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NBDCJvLbq7nuor7RtT57rD

Copilot AI lite review requested due to automatic review settings August 29, 2026 00:45

This was referenced Aug 29, 2026

chore(fuzz-deps): bump the fuzz-deps group in /fuzz with 13 updates #3

Closed

chore(deps): bump the cargo-major group with 17 updates #4

Closed

ci(deps): bump the actions group across 1 directory with 15 updates #5

Closed

Copilot started reviewing on behalf of doublegate August 29, 2026 00:46 View session

Copilot AI reviewed Aug 29, 2026

View reviewed changes

doublegate merged commit 3e0d196 into main Aug 29, 2026
14 checks passed

Navigation Menu

chore(deps)!: take the major-version wave, fixing what it exposed - #8

chore(deps)!: take the major-version wave, fixing what it exposed#8
doublegate merged 2 commits into
mainfrom
chore/dependency-majors

doublegate commented Aug 29, 2026

Uh oh!

github-actions Bot commented Aug 29, 2026

Uh oh!

coderabbitai Bot commented Aug 29, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Aug 29, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

doublegate commented Aug 29, 2026

Two of the riskiest majors needed deleting, not migrating

The real migration was rand

sysinfo 0.38, not 0.39

What the wave exposed

Unbounded expansion — the serious one

host_count() returned 0 for a /31

host_count() and expand_hosts() measured different things

Two optimizations from the same reading

Test harness

Verification

Uh oh!

github-actions Bot commented Aug 29, 2026

Uh oh!

coderabbitai Bot commented Aug 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Aug 29, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Suppressed comments (3)

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

The real migration was `rand`

`sysinfo` 0.38, not 0.39

`host_count()` returned 0 for a `/31`

`host_count()` and `expand_hosts()` measured different things

coderabbitai Bot commented Aug 29, 2026 •

edited

Loading