Skip to content

AI uses for records

Dependency update impact dataset: what 162 compatible bumps changed

By SourceX Editorial · Updated

10 sections

Short answer

A dependency update impact dataset links each version bump to the records that show what else moved: the lockfile graph, the dependency's declared metadata, its changelog and the follow-ups. Of 162 Dependabot bumps in nushell from 2023 to 2025 that Cargo treats as compatible, 78 changed something beyond the version: other locked crates (51), declared dependencies or minimum Rust version (48), or a stated compatibility change (11).

System export

Dependency update impact

This page covers dependency update impact dataset: what 162 compatible bumps changed.Illustrative

Key takeaways

  • Read the lockfile, not the title: 101 of 287 nushell Dependabot bumps changed other locked crates, about a third in every version class.
  • Compatible is not unchanged: 18 of 105 compatible bumps whose versions both declared a minimum Rust version raised it, one in a patch release.
  • Changelogs said more than version numbers: 11 of 67 compatible bumps with a pasted excerpt stated a compatibility change.
  • Follow-ups are rare but specific: one revert, over duplicated dependencies and compile cost, and 33 re-bumps within 30 days.
  • Version labels mislead both ways: 124 of 125 bumps classed as breaking merged without touching any file beyond manifests and the lockfile.

What does a dependency update change besides the version number?

A dependency update can change the rest of the lockfile, the dependency's own declared requirements, its minimum supported toolchain and its behavior, while the pull request title shows only two version numbers. A dependency update impact dataset links each bump to the records that reveal those changes: the lockfile diff, the registry's metadata for both versions, the changelog entries in range, the review and test outcome, and whatever happened in the weeks after.

Counted in one active Rust project, the gap is large. Of 162 Dependabot bumps in nushell between 2023 and 2025 that Cargo's own rules treat as compatible, 78 (48.1 percent) changed something beyond the version: other locked crates, declared dependencies, the declared minimum Rust version, or a compatibility change the maintainers wrote into their changelog.

That matters to anyone training a coding agent to propose, review or merge updates. SourceX's guide to software dependency upgrades covers upgrades that broke the build and needed human fixes as coding tasks. This page covers the far larger set that looked safe, merged without code changes, and still moved more than the label said.

Why a compatible version label is a promise, not a measurement

A compatible version number states the publisher's intent; it does not measure what changed. Cargo's documentation defines the rule its default requirements follow: "Versions are considered compatible if their left-most non-zero major/minor/patch component is the same." Under that rule an update from 1.9.6 to 1.10.2 is compatible, and so is 0.2.0 to 0.2.1.

The same documentation allows real changes inside that boundary. On raising the minimum Rust version, it says "It is generally recommended to treat this as a minor change, rather than as a major change," and it warns that a newly deprecated item triggers a deprecated lint that can fail a build whose owners deny warnings. Mistakes happen too: the maintainers of cargo-semver-checks reported 3,062 verified semver violations in non-major releases of the 1,000 most-downloaded library crates, and said more than one in six of those crates had accidentally violated semver.

Dependabot's own documentation puts the burden on review: you "check that your tests pass, review the changelog and release notes included in the pull request summary, and then merge it." The records below show what that review would have had to catch.

What 287 Dependabot bumps in nushell changed

Roughly a third of nushell's Dependabot bumps changed other crates in Cargo.lock, whatever their version label. The count covers 287 single-crate bumps authored by Dependabot and merged to nushell's main branch from January 1, 2023 to December 31, 2025; Cargo's rule classes 28 as patch-level, 134 as compatible minor updates and 125 as breaking.

Declared metadata moved too. Comparing each bump's old and new versions in the crates.io index, 68 of 287 changed the list of normal dependencies the crate declares, and among the 143 bumps where both versions declare a minimum Rust version, 29 raised it. Among compatible bumps alone, 18 of 105 raised it (17.1 percent), including a patch release of array-init-cursor that moved its declared minimum from Rust 1.57 to 1.75.

The breakdown by Cargo's version classes:

What 287 Dependabot bumps in nushell changed
Cargo version classBumpsOther locked crates changedDeclared dependencies changedMinimum Rust raised (both declare)
Patch289 (32.1%)2 (7.1%)1 of 17 (5.9%)
Compatible minor13442 (31.3%)31 (23.1%)17 of 88 (19.3%)
Breaking by Cargo rule12550 (40.0%)35 (28.0%)11 of 38 (28.9%)
All bumps287101 (35.2%)68 (23.7%)29 of 143 (20.3%)
An iceberg chart: above the waterline, 162 compatible version bumps in nushell; below it, bars for what else changed, 51 moved other locked crates, 33 changed declared dependencies, 18 raised the minimum Rust version, 11 stated a compatibility change and 6 added a duplicate crate, totaling 78 bumps.
What 162 Cargo-compatible Dependabot bumps in nushell changed beyond the version number, 2023-2025. Source: nushell git history and the crates.io index, counted October 27, 2026.

What the changelogs said, where they were pasted in

Where Dependabot's pasted changelog survived in the merged commit, about one compatible bump in six carried a compatibility change in the maintainers' own words. Of the 162 compatible bumps, 67 kept a changelog or release-notes excerpt in nushell's history; a keyword screen found 20 candidates, and reading each confirmed 11.

Five raised the minimum Rust version, two removed public implementations, three deprecated APIs or features, and one moved to a new major version of an underlying HTTP library under a heading marked Breaking. tempfile 3.5.0, a compatible minor update, said: "BREAKING: Remove the implementation of Write for &NamedTempFile<F> where &F: Write. Unfortunately, this can cause compile issues in unrelated code." The same bump added second copies of three crates to nushell's lockfile.

Some changes were transitive even in the changelog. plist 1.8.0 noted "Bump MSRV to v1.81 as required by time-core v0.1.6": the dependency raised its own floor because one of its dependencies had. A keyword screen alone would have misled in both directions; nine of the 20 candidates mentioned the word for unrelated reasons, such as a CI setting or a test fix, so the count is a lower bound from excerpts that are often truncated.

Follow-ups: one revert, quick re-bumps and withdrawn releases

Follow-up records were rare but telling. Only one of the 287 bumps was reverted: regex 1.9.6 to 1.10.2, a compatible minor update that added second copies of regex-automata and regex-syntax to the lockfile and raised regex's declared minimum Rust version from 1.60.0 to 1.65. It was reverted about eight hours after merging, with the reason that the old version was shared with many other dependencies, and that before the bump nushell did not duplicate big dependencies of regex "that affect binary size and compile time."

Faster signals came from upstream. Dependabot bumped the same crate again within 30 days after 33 of the 287 bumps (19 of the 162 compatible ones), which often means the maintainers shipped a quick follow-up release. And as of October 27, 2026, four of the versions nushell moved to had been yanked by their publishers on crates.io, a status Cargo's index documentation says may change at any time, so it has to be read with a date.

None of the 162 compatible bumps changed any file besides Cargo.toml files and Cargo.lock in the merged commit. Whatever they changed, the project absorbed it without touching its own code, which is exactly why the change is invisible in a diff of the application.

The dependency change trace: what each bump record should carry

Each bump record needs seven fields, each from a named source, so a buyer can check what a dataset can and cannot show before licensing it. The right-hand column gives nushell's public coverage, a rough guide to what a raw company export will hold.

The lockfile and the registry carry most of the signal, and both are cheap to keep; changelog text and follow-up links are where public data runs thin:

The dependency change trace: what each bump record should carry
FieldSource recordWhat it revealsnushell coverage
Declared changePull request title, manifest diffOld and new version, version class under Cargo's ruleAll 287
Lockfile deltaCargo.lock before and afterCrates added, removed, re-versioned or duplicatedAll 287; 101 moved other crates
Declared metadata deltaRegistry index entries for both versionsMinimum Rust version, normal dependencies287 in index; 143 declare both minimums
Changelog entries in rangePull request body, upstream changelogStated breaking changes, deprecations, removals112 merged commits kept an excerpt
Review and test outcomeCI runs, review commentsWhat was checked before mergeNot in git history; needs the hosting platform
Follow-upsReverts, pins, re-bumps, fixes citing the bumpWhether the change had to be undone or adjusted1 revert, 33 re-bumps within 30 days
Upstream statusRegistry yank flag, advisories, datedWhether the publisher withdrew the version4 target versions yanked by October 27, 2026

An outcome rubric for dependency bumps

The rubric turns the trace into labels a model can learn from or be graded against. Labels can overlap; a bump that moved the lockfile and raised the minimum Rust version carries both. Counts are for nushell's 162 compatible bumps:

An outcome rubric for dependency bumps
LabelRuleCount of 162Use in training or evaluation
Version onlyNo other locked crate, declared metadata or stated change moved84 (51.9%)Negatives for a safe-to-merge classifier
Graph movedOther crates added, removed, re-versioned or duplicated in the lockfile51 (31.5%)Teach review agents to read the lockfile, not the title
Contract movedDeclared minimum Rust version raised or normal dependencies changed48 (29.6%)Toolchain and supply-chain checks
Stated changeChangelog in range states a compatibility change (manually confirmed)11 of 67 with excerptsGround truth for changelog reading
Rolled backLater commit reverts the bump1Rare positive; keep the stated reason
Superseded fastSame crate bumped again within 30 days19Signal that the upstream release needed a follow-up

The strongest objection: the tests passed and nothing broke

The strongest objection is that the tests passed, the bumps merged, and only one in 287 was undone, so these changes did not matter. That is partly right. Most extra changes caused no visible trouble in nushell's history, and a dataset that labels every moved lockfile as risky would teach a model to cry wolf.

But passing tests answer a narrower question than the business one. The regex revert was not about a failing test; it was about binary size and compile time, costs no unit test reports. A raised minimum Rust version does not fail a project that already builds on a newer toolchain, yet it can strand a downstream user or a packager on an older one. And 124 of the 125 bumps that Cargo's rule classes as breaking merged without changing any file besides manifests and the lockfile, so the label misleads in the other direction too.

The fair conclusion is that version labels are poor labels in both directions, and that the records to replace them exist. Whether a moved graph mattered is a question for each company's own builds, binaries and customers, which open-source history can only hint at.

How buyers would use dependency impact records, and what private repositories add

A buyer would use these records to train and test the judgment behind an update, not the edit. Given the trace's pre-merge fields, a model predicts the outcome label, flags a raised toolchain floor or a new duplicate before merge, and writes the review note a careful maintainer would write. Held-out periods, never random splits, keep the same crate's releases from leaking across sets.

Private repositories add what public history lacks. They hold internal packages and private registries, the CI logs for each bump, the update pull requests that were closed without merging, which never reach the main branch, and incidents traced back to an upgrade. They also add other ecosystems, with lockfiles from npm, Maven or Python tools, where the same fields apply with different rules.

SourceX sources datasets on request, not from stock. A buyer describes the domain, source systems, volume, format, timeframe, licensing and de-identification needs on its request form; for this task it should name the seven trace fields, the rubric labels it needs, and an acceptance test such as recomputing lockfile deltas from the delivered files.

Methods note: repository, index and counting rules

Data: github.com/nushell/nushell, cloned on October 27, 2026 as a partial clone at commit 9970552 (committed 10:39 CDT that day); Cargo.lock blobs were fetched for the commits read, and no repository code was run. The population is first-parent commits on main whose author name is dependabot[bot], committed from 2023-01-01 to 2025-12-31 UTC, whose subject matches a single-package bump (optional build(deps) or chore(deps) prefix, then bump, a name without a slash, from, a version, to, a version) and whose diff touches Cargo.lock: 287. The script dropped and printed 104 bumps of names containing a slash, which are GitHub Actions.

Version class follows Cargo's rule: breaking if the left-most non-zero component differs; patch if major and minor match; otherwise compatible minor. The lockfile delta compares registry and git packages (path packages skipped) at the commit and its parent: other locked crates changed means any crate other than the bumped one added, removed, or with a changed set of versions; a new duplicate is a crate with more versions than before and at least two. Registry metadata comes from the crates.io sparse index, fetched on October 27, 2026: rust_version compared as numbers when both versions declare it, and the sorted list of normal dependencies, duplicates kept (renamed dependencies by package name).

Changelog candidates are excerpts inside Dependabot's Changelog or Release notes blocks, with tags stripped before entities are decoded, that match whole-word breaking, MSRV, minimum supported rust version, rust-version, behavior or behaviour and their plurals, words starting deprecat, or remove and removed; each candidate was read and labeled by the writer, and the labels are saved with the scripts. A revert is a later commit naming the bump's hash, quoting its subject after Revert, or saying Reverts nushell/nushell# and its pull request number. A re-bump is a later Dependabot bump of the same crate within 30 days. Shares are computed from integers and rounded once, half up. Not measured: CI results, review comments, closed pull requests, compile time or binary size, and anything about any company's software. Outputs hold hashes, crate names, versions, dates and counts only.

Frequently asked questions

What is a dependency update impact dataset?

A set of version bumps, each linked to the records that show what else changed: the lockfile diff, registry metadata for both versions, changelog entries in range, test and review results, and follow-ups such as reverts or quick re-bumps. It lets a model learn what a bump did rather than what its title claimed.

Can a patch or minor update raise the minimum Rust version?

Yes. Cargo's documentation recommends treating a raised minimum Rust version as a minor change, and in nushell 18 of 105 compatible bumps where both versions declared a minimum raised it, including one patch release.

Why look at Cargo.lock instead of Cargo.toml?

Cargo.toml shows the requirement you edited; Cargo.lock shows every crate that actually resolved. In nushell, 51 of 162 compatible bumps changed other locked crates, and six added a second version of some crate, none of which the manifest diff shows.

Does Dependabot include changelogs in its pull requests?

GitHub's documentation says reviewers should read the changelog and release notes included in the pull request summary. Whether that text survives depends on how the project merges; in nushell, 112 of 287 merged bump commits kept an excerpt.

Can a company license its dependency update history?

Often, after rights review. Repository history can include vendored third-party code and private registry details, so counsel checks those terms, and access to a repository is not the same as the right to license what is inside it.

Sources

  • nushell's main branch history at commit 9970552 (October 27, 2026) supplies every bump, lockfile and follow-up count in this article. Source
  • The crates.io sparse index lists each published version with its declared rust_version, dependencies and yanked flag; regex 1.9.6 declares Rust 1.60.0 and 1.10.2 declares 1.65. Source
  • Cargo's documentation says versions are compatible if their left-most non-zero major, minor or patch component is the same. Source
  • Cargo's SemVer guide recommends treating a raised minimum Rust version as a minor change and notes that new deprecations trigger lints that can fail builds denying warnings. Source
  • Cargo's registry index documentation says index entries are not modified after publication except the yanked field, which may change at any time. Source
  • GitHub's Dependabot documentation tells maintainers to check that tests pass and review the changelog and release notes in the pull request summary before merging. Source
  • The cargo-semver-checks maintainers found 3,062 verified semver violations in non-major releases of the top 1,000 Rust library crates and said more than one in six crates had violated semver. Source
  • nushell's tempfile 3.4.0 to 3.5.0 bump kept a changelog excerpt marking the removal of a Write implementation as breaking. Source
  • nushell's plist 1.7.0 to 1.8.0 bump kept a changelog excerpt raising the minimum Rust version because a transitive dependency required it. Source
  • nushell reverted its regex 1.9.6 to 1.10.2 bump on October 23, 2023 to avoid duplicating large regex dependencies. Source
  • SourceX's request form asks buyers for domain, source systems, volume, format, timeframe, licensing and de-identification requirements; datasets are sourced on request. Source

Tell us what data you need

Share scope, volume, format, timing and licensing needs. Datasets are sourced on request.

See if your company qualifies