Skip to content

chore(deps): update dependency fallow to v3.42.0 - #980

Merged
fengmk2 merged 1 commit into
masterfrom
renovate/fallow-3.x-lockfile
Nov 13, 2026
Merged

fengmk2 merged 1 commit into
masterfrom
renovate/fallow-3.x-lockfile

Conversation

@renovate

@renovate renovate Bot commented Nov 13, 2026

Copy link
Copy Markdown
Contributor

This PR contains the following updates:

Package Change Age Confidence
fallow (source) 3.41.0 → 3.42.0 age confidence

Release Notes

fallow-rs/fallow (fallow)

v3.42.0

Compare Source

Added
  • fallow coverage upload-inventory --with-callers now finds the entry routes of Elysia and TanStack Router projects. A TanStack Router route file takes its pattern from the string argument of createFileRoute("/repo/$repo"), with method: null. An Elysia route comes from a call such as app.get("/repos/:repo", handler). The pattern joins the prefix of the instance, the .group() prefixes and the prefixes of the apps that mount the plugin with .use(). The file of the handler is the route entry. A route with a path or a prefix that is not a string literal is not uploaded. The route caps and the routeLimits header do not change. (Closes #​4271)

  • The GitHub Action and the GitLab template can run fallow mutation. Set the input mutation: true (GitLab: FALLOW_MUTATION: "true"). A Resolve change scope step (GitLab: an inline block) writes one root-relative diff first. Then the mutation run uses that diff, before the analysis, because the run refuses a dirty worktree. The default scope is the changed lines of the pull request, with a 10m time budget. The inputs mutation-scope, mutation-time-budget, mutation-max-mutants and mutation-min-score (GitLab: FALLOW_MUTATION_SCOPE, FALLOW_MUTATION_TIME_BUDGET, FALLOW_MUTATION_MAX_MUTANTS, FALLOW_MUTATION_MIN_SCORE) set the run. fallow report renders the results (see the next entry). A failed mutation-min-score fails the job with exit code 1, and a tool failure of the run (exit code 2) fails it with exit code 2, after the comment. The step runs only for a job that cannot run the code of a fork with the secrets of the base project: on GitHub, the pull_request, push, schedule, workflow_dispatch and merge_group events; on GitLab, not a fork pipeline in the target project. It removes the CI tokens and the GITHUB_* file commands from the environment of the tests, and it never passes --allow-dirty. The result cache goes through actions/cache and the GitLab job cache, with a key that covers the lockfiles and the runner config. A fork pull request never saves the cache. With a fallow that cannot render the results, the gate still fails the job, and the job summary names the version that renders them. fallow ci-template gitlab --vendor also writes ci/scripts/mutation.sh and ci/scripts/mutation_common.sh.

  • fallow mutation makes more mutants with the existing operators. The new rules are:

    • an if or ? : test that is a comparison or a logical expression now becomes true and false. A loop test of this kind becomes false only, and for (;;) gets the test false. A constant that the test always has, such as true for items.length >= 0 on an array, gives no mutant;
    • an operand of && that is a comparison or a !x becomes true, and such an operand of || becomes false;
    • an equality also becomes its other constant: a === b becomes true as well as false. A == or === loop test gets no true constant;
    • regex-literal removes a +, * or ? quantifier, shortens {n,m} to {n}, and removes an anchor at the start or end of a group or an alternative. A removal that would join two tokens, as in \1?2, puts (?:) in the place of the quantifier, and a mutated literal that does not parse is dropped;
    • return-value changes the concise body of an arrow function to undefined;
    • arithmetic-operator swaps *= and /=, <<= and >>=, &= and |=, and changes %= to *=. logical-operator swaps &&= and ||=, and changes ??= to &&=;
    • unary-operator swaps -x and +x, and removes ~;
    • string-literal empties a template literal and fills an empty one. A tagged template gives no mutant.

    The operator names do not change. An existing mutant keeps its id. A mutant that two rules make is one mutant.

  • fallow mutation --typecheck now adds an equivalence hint to an optional-chaining survivor. The mutants a?.b to a.b, a?.() to a() and a?.[k] to a[k] change nothing when a is never nullish at run time. When the declared type of a excludes null and undefined, the survivor gets equivalence_hint: "type-non-nullish". The hint is advisory. The mutant stays survived and stays in the score. The human report and fallow mutation tasks show a one-line note, and a verdict file (--verdicts) records the mutant as equivalent after review. typecheck.hinted counts the hinted survivors. The hint needs strictNullChecks and noUncheckedIndexedAccess. A verdict for a hinted survivor removes the hint. The hint can be wrong, for example in a method that loses its this.

  • fallow mutation --time-budget now runs the cheapest mutants first. With a time budget, the first pass starts the mutants with the lowest estimated test time first. The estimate is the sum of the dry-run times of the covering tests. The budget now settles more mutants. Without a budget, the order does not change.

  • fallow mutation --mutants-per-line <n> limits the mutants on each source line. The config key is mutation.mutantsPerLine. The plan keeps the most productive mutants of a line, in a fixed operator order, and ranks equal operators by a stable hash of the mutant id. The other mutants are pending with the reason sampled-out, and the run does not execute them. The summary and the plan counts give the number as sampled_out. The completion of such a run is partial with the reason sampled-out. A consumer that switches on completion.reasons must accept this new value. The MCP mutation_plan tool takes the same limit as mutants_per_line.

  • fallow mutation now warns about skipped and todo tests. A test file can leave a test out of the run, for example with it.skip. Such a test checks nothing. When the changed-code scope changes a runner test file, a scan of the syntax tree finds these tests. The run reports them in errors with the code skipped-tests, with file, line and name, and the human output shows a warning. A change to the tests alone starts no runner. A run with --all or paths gives no such warning. The completion does not change.

  • fallow mutation --function-scope widens the changed lines to whole functions. The config key is mutation.functionScope. A change to a parameter or a signature line holds no mutant, so the default changed-code scope can select nothing for it. With the option, each added line selects all mutants of the function that holds it. The option is off by default.

  • fallow mutation export writes a saved report in the open mutation-testing-report-schema format. Mutation dashboards and HTML report viewers read this format, so a team can keep its dashboard. The command runs no code. It maps each Fallow status to a schema status: flaky, pruned and equivalent become Ignored with a reason, pending stays Pending, and a survivor of a partial run gets a reason that says the result is not conclusive. Columns count UTF-16 code units. The export refuses a report whose mutated files changed after the run.

  • Built-in gulp plugin. When a package depends on gulp or gulp-cli, fallow makes its gulpfile an entry point. This includes gulpfile.{js,mjs,cjs,ts,mts,cts,babel.js,esm.js} and the index file of a gulpfile.js directory. The task files that a gulpfile imports are no longer unused files.

  • Built-in n8n plugin. n8n loads the nodes and credentials of a node package from the n8n.nodes and n8n.credentials arrays in package.json. Fallow now makes these files entry points, so node files, credential files and the files that they import are no longer unused files. Each path names build output, such as dist/nodes/Foo/Foo.node.js. Fallow maps it to the source file through the tsconfig outDir and rootDir, or to the source file beside the output directory when no rootDir is set. The members that n8n reads from a class that implements INodeType or ICredentialType, such as description, webhookMethods and documentationUrl, are no longer unused class members.

  • Built-in unplugin-icons plugin. An icon import such as import IconX from '~icons/lucide/x' or virtual:icons/mdi/arrow-left is no longer an unlisted dependency ~icons. The import also credits the icon set package (@iconify-json/lucide), the full @iconify/json collection and unplugin-icons, when the project declares them. Before, fallow reported these packages as unused dev dependencies.

  • Built-in vite-plugin-pages plugin. Fallow now reads the Pages(...) call in vite.config.* and makes the page files under each dirs entry entry points. A dirs entry can be a string or an object with dir, and can point into another workspace package ("../common/src/pages"). Without dirs, the page directory is src/pages. The extensions option sets the page file extensions. Before, fallow reported the page files and the files that only pages import as unused files.

  • Built-in SolidStart plugin. The solid-start plugin activates from the @solidjs/start dependency. It makes <appRoot>/app.{tsx,jsx}, <appRoot>/entry-client.{tsx,jsx}, <appRoot>/entry-server.{tsx,jsx} and each module under <appRoot>/<routeDir> entry points. Before, fallow reported these files and all the modules that they import as unused files. Route files get the default, route and HTTP method exports. The plugin reads appRoot, routeDir, extensions and middleware from the solidStart() call in vite.config.* and from defineConfig() in the version 1 app.config.*. The defaults are src and routes.

  • Built-in dumi plugin. The dumi plugin activates from the dumi dependency. It makes each demo file that a <code src="..."> tag of a Markdown file names an entry point, and credits its default export. Before, fallow reported these demo files as unused files. A relative src is read from the directory of the Markdown file, and a src that starts with / from the package root. Fallow now also analyzes the hidden .dumi directory: the theme builtins, slots, layouts and plugin, the pages routes and the global, app and loading files are entry points. The plugin reads .dumirc.ts and .dumirc.js: it credits the imported packages and the plugins and presets packages, and it reads alias as path aliases.

  • fallow report renders mutation results for CI. The new flags --mutation-report <PATH> and --mutation-not-run <REASON> work with the pr-comment, review, github-summary, github-annotations and codeclimate formats. Survived mutants on changed lines appear as findings with the rule fallow/mutation-survived in the new Tests category: rows in the sticky comment, annotations, inline review comments and Code Quality entries, 20 at most for each render. The comment, the review summary and the job summary get one "Mutation testing" section, and the check run gets a mutation row and a mutation-min-score row. A survivor in a file that changed after the run is not shown. Without the flags, the output of fallow report does not change.

  • fallow mutation writes its score gate into the report. A run with --min-score or mutation.minScore adds gate_outcomes["mutation-min-score"] to the mutation report, and exit code 1 reads that entry. The gates-* outputs of the Action and the FALLOW_GATES_* lists of GitLab name the gate. The Action has the new output mutation-report-rendered.

  • fallow mutation has nine opt-in operators. They run only when --operators, mutation.operators or the MCP operators parameter names them:

    • block-statement removes the statements of a block or function body without declarations;
    • array-literal changes [x] to [] and [] to ["fallow"];
    • object-literal changes a non-empty object literal to {};
    • method-expression swaps method pairs such as startsWith and endsWith, and removes calls such as .trim() and .filter(...);
    • switch-case removes the statements of a clause;
    • call-expression changes a call in an expression position to (void 0);
    • throw-statement removes throw new X(...);
    • regex-constructor applies the regex changes to a new RegExp("...") pattern;
    • regex-assertion negates a lookaround and swaps \b and \B.

    The default set does not change. docs/mutation-testing.md lists both sets.

Changed
  • The warning for a config glob that matches fewer files since 3.34.0 now names the files and gives two fixes. Since 3.34.0, a single * or ? in a config path glob matches in one folder level only. The stderr warning for a glob that lost files gave only a count and a ** rewrite. Now it names up to 3 of the lost files, then "and N more". It also tells you how to keep the one-level meaning: write the names in place of *, for example {name1,name2}. This config change stops the warning. The ** rewrite stays as the second choice. For entry and dynamicallyLoaded, a lost file that another entry source declares, for example a plugin or package.json, does not count, because it is still an entry point. When no lost file remains, fallow prints no warning. Only the commands that find entry points, for example fallow dead-code and fallow health, check these two fields. The warning stays on stderr, --quiet hides it, and JSON output does not change.

  • fallow mutation stops an endless loop through the active mutant in under a second. Before, a synchronous endless loop blocked the Vitest worker until the watchdog fired after timeout_ms (5 seconds or more). Then Fallow started a new companion and a new dry run, and the retry waited twice as long. Now the mutant switch checks a loop budget of 1.5 times the covering-test time plus 500 ms, and the Vitest adapter reports the test as a timeout in the same companion. On three public Vitest libraries, a full run with 2 workers took 1.7 to 6 times less wall time, and the score did not change. A mutant that only makes a test slow, for example under host load, can now get timeout in place of killed or survived. The retry with a doubled budget lowers this risk. Only the Vitest adapter checks the loop budget.
    (Part of #​4341)

  • fallow mutation stops more endless loops in the companion, and gives a loop timeout of short tests no retry. The loop check now counts each evaluation of an instrumented site in a mutant run, not only the evaluations of the active mutant. So a mutant that makes a loop of other source code run forever, for example a return that gives undefined to the condition of a caller loop, also ends at the loop budget. The loop budget of a mutant run now counts the CPU time of the test thread, not the wall time. A test that waits uses no budget, and load from other processes on the host does not count. A loop timeout of covering tests that take 200 ms or less in the dry run is final: the budget is then at least four times the work of the tests. Before, each such mutant ran a second time with twice the budget.
    (Part of #​4341)

  • fallow mutation runs the test files of one Vitest mutant run in one worker. Vitest starts a new worker for each test file of a run when module isolation is on. After the dry run, the Vitest adapter now runs the test files once more in one shared worker. When each test gives the same result, the mutant runs of the companion use one worker for all of their test files. Each run still starts a new worker, so the modules stay fresh for each mutant. A test that needs a fresh worker for each file makes the two dry runs differ, and the runs then stay isolated. A kill in a later file of a shared worker counts only when the file also detects the mutant alone. On a public Vitest library, where a mutant has covering tests in almost three files, this cut the worker processes by a quarter, and no verdict changed.
    (Part of #​4341)

  • fallow mutation keeps one Vitest worker process for many mutant runs. Vitest starts a new worker process for each run, and that start cost more CPU than most mutant runs. When the dry run gives the same results in one reused process, also when each module loads a second time, the Vitest adapter now keeps the worker process between mutant runs. Each test file of a reused run loads the source modules again with the current mutant. A run whose tests change state outside the modules (a global, a prototype, process.env, a listener, a timer) ends the process, and a kill in such a run counts only when the test file also detects the mutant in a new worker. Dry runs, control runs and --kill-matrix runs still start a new worker. On three public Vitest libraries, the CPU time of a full run went down by a factor of 1.6 to 3.0, and no verdict changed.
    (Part of #​4341)

  • arithmetic-operator skips *, / and % with a string or template operand, as it already did for + and -.

Fixed
  • fallow dead-code credits a member on a method chain with arguments on a typed parameter. In function toWorld(local: Point) { return local.rotate(1).translate(1, 1); }, where rotate(angle: number): Point, fallow now credits Point.translate. A new Point(...) local and a chain without arguments already gave this credit. Thanks @​Smrtnyk for the report.
    (Closes #​4311)

  • fallow dead-code credits a member on the return value of an inherited method in a class that is not exported. In class Viewer extends Floor, where Floor declares find(): Table | undefined in another file, this.find()?.setVip() now credits Table.setVip also when Viewer is not exported. A deeper chain of unexported classes also gives the credit. A class on the chain that declares find() itself gives no credit to the base. Thanks @​Smrtnyk for the report.
    (Closes #​4312)

  • fallow dead-code credits a member on an element of an array that a local function returns. With function all(): Table[] in the same file, fallow now credits Table.setVip for for (const table of all()), all().forEach((table) => table.setVip()), all()[0].setVip(), and const tables = all(); for (const table of tables). The function can come before or after the reads. A call of a nested declaration or a catch parameter named all, and a call in a namespace body, gives no credit. The extraction cache version is 422. Thanks @​Smrtnyk for the report.
    (Closes #​4313)

  • fallow mutation reports tests that share a test ID. Two tests of one file that ran, with the same full name (for example test.each with a title that has no placeholder), get the IDs name and name (2) in report order, so the evidence of these IDs can belong to the other test. The dry run now reports them in a duplicate-test-ids entry of errors (a warning in the human output), and the run goes on. A skipped or todo twin makes no duplicate. --kill-matrix-for with a test ID target that names a duplicate ID exits 2 with FALLOW_MUTATION_TARGET_DUPLICATE_TEST_ID. A test file target drops the duplicate IDs and keeps its other tests.
    (Closes #​4170)

  • fallow audit keeps a css-duplicate-block finding inherited when only the lines of a later occurrence move. Before, a script or template edit that shifted the copy of a block in the second file of a group made the finding introduced, although the group and its count did not change. The audit key of a duplicate block finding now uses the sorted paths of all occurrences of the group, not the path of the occurrence that the changed-file scope names first. A new file that joins the group still gives an introduced finding. Thanks @​Smrtnyk for the report. (Closes #​4344)

  • fallow dead-code uses the element class of a for...of loop variable only in the loop body. After for (const t of tables) with tables: Table[], a parameter t of another type in another function no longer credits Table members. This applies also to a loop over a call, such as for (const t of all()).
    (Closes #​4352)

  • fallow dead-code no longer credits a member through this in a function with its own this. In a nested function or an object-literal method inside a class method, this.find().setVip() no longer credits find or setVip. An arrow function keeps the this of the method and still gives the credit. A function that the code gives the instance also keeps the credit: .bind(this), .call(this) or .apply(this) on the function or its name, a call that also passes this (such as items.forEach(function () {}, this)), a method call on this (such as the listener in this.on("ready", function () {})), an assignment to a this.<field> or a class field, and a this parameter with a named type. The extraction cache version is 423.
    (Closes #​4351)

  • fallow mutation now selects a test that asserts only on the output of a Node child process. With the Vitest and node:test runners, a Node process that a test starts (spawn, exec, execFile or fork) now gives its coverage hits to that test in the dry run. Before, such a test had no hits, so the run did not select it for a mutant. When no other test file detected the mutant, the report showed a false survivor. A child process that a signal ends gives no hits.
    (Closes #​4171)

  • A crafted TOML file no longer crashes fallow. A fallow.toml, an external plugin file, or a TOML file that a framework plugin reads (bunfig.toml, mise.toml, wrangler.toml, .prettierrc.toml, GraphQL config) could crash any command with a stack overflow. Fallow now rejects TOML that nests arrays and inline tables deeper than 128 levels before it parses the file. A rejected config file gives a parse error. A rejected plugin file is skipped, as with other invalid TOML.

  • A crafted stylesheet can no longer make fallow dead-code or fallow health --css hang. The CSS parser parsed the content of some unclosed color functions such as light-dark( twice at each level, so the parse time doubled with each nested function. Fallow now skips the structural CSS parse of a stylesheet that nests more than 8 color functions. CSS Modules classes in such a file come from the plain class scan, and the file gets no CSS analytics. Real stylesheets are not affected.

  • A crafted TypeScript, JSX or MDX file can no longer make fallow hang. Some sources with many nested < characters made the TypeScript parser try each nested part again and again, so one small file could keep fallow dead-code, fallow dupes or MDX extraction busy for minutes. Fallow now counts this nesting before it parses a file. A file past the cap is reported as a failed parse, the same as a file with too-deep bracket nesting. Real code stays well below the cap, so its results do not change.

  • fallow health --css keeps a .scss or .less file with an absolute url() in css_analytics. Before, the Sass and Less lowering read the // of url(//cdn.example.com/a.png), url(https://...) or url("//...") as a line comment, so the file lowered to nothing and dropped out of the analytics. As in Sass, a // inside a string or an unquoted url(), or after a :, is now part of the value. A // or a quote inside a /* */ comment no longer hides the end of that comment or a later line comment. Thanks @​rezamuliawijaya for the fix. (Closes #​4324)

  • The LSP offers the remove-export quick fix for a name with a combining mark. JavaScript allows a combining mark after the first character of a name, for example café written as an e followed by the character U+0301. The quick fix read the name only up to the mark, so it gave no action for such an export. It could also give the action for an export cafe on that line. The quick fix now reads a name with the identifier rule of the parser. The "Dismiss optional prop review" action uses the same rule, so it no longer adds a comment for a longer name that only starts with the prop name.

  • LSP diagnostics on the first line of a file with a UTF-8 byte order mark start at the correct column. The editor buffer does not hold the byte order mark, but the LSP counted the mark when it read the file from disk. Thus diagnostics on line 1 started 2 columns too far to the left, and a config diagnostic on line 1 started 1 column too far to the right. Also, an open file with a byte order mark never matched its file on disk, so the LSP did not publish its diagnostics. The LSP now drops the leading byte order mark when it reads a file.

  • Duplication results count lines and columns from the end of the byte order mark. The duplication detector read a file with its UTF-8 byte order mark, so a clone on line 1 started 1 column too far to the right in the CLI, JSON and LSP output. The detector now drops the mark, the same as extraction. The duplication token cache version changes, so a warm cache does not give the old columns.

  • A file that starts with two U+FEFF characters gets correct spans. Only the first U+FEFF of a file is the UTF-8 byte order mark. The second U+FEFF is source text. Extraction removed both, so each span of the file was 3 bytes too early. For \u{feff}\u{feff}export const a0 = 1;, the span of a0 held st. Extraction now removes the mark one time only. The extraction cache version and the graph cache version change, so a warm cache does not give the old spans.

  • A Glimmer argument that only a <template> block reads counts as used in a file with a byte order mark. In a .gts or .gjs file with a UTF-8 byte order mark, the template ranges counted the mark and the parser spans did not. A template at the end of its class then fell outside the class, so its argument reads did not count.

  • CodeClimate clone fingerprints no longer change with CRLF line endings. The CodeClimate fingerprint of a clone instance hashes the first 64 characters of the clone fragment. Before, these characters kept each \r, so a Windows checkout with core.autocrlf gave other fingerprints than a Linux checkout of the same code. GitLab Code Quality then reported each clone as new and as fixed. The review-github, review-gitlab, pr-comment-github and pr-comment-gitlab formats merge and sort by this fingerprint, so their output changed too, in fallow dupes and in fallow audit. The hash now leaves out \r. Files with LF line endings keep their fingerprints. A project with CRLF files gets new fingerprints one time, and they then match the fingerprints of an LF checkout. When two clone groups share one dup: hash, the order of their -r1 and -r2 suffixes also no longer changes with the line endings. The fallow audit clone-group key also leaves out \r, and the audit base snapshot cache moves to version 21.

  • A test runner or a bundler in the root package.json now activates its plugin in a workspace. A monorepo often declares a tool such as Mocha or Vite only in the root package.json. The package manager hoists it, and a workspace script such as mocha --exit or vite build calls its binary. Before, the plugin of a workspace became active only for a dependency of that workspace. The test files of the workspace and the files that its index.html links were then reported as unused files. A root dependency now activates its plugin in a workspace when a script of that workspace calls the binary of the dependency. A root dependency that no workspace script calls does not activate a plugin in the workspace.

  • fallow dead-code credits a member on the result of an awaited async static factory. With static async create(): Promise<Service> in another file, const service = await Service.create(); service.getConflicts() now credits Service.getConflicts, as Service.build() with static build(): Service already did. The factory can build the instance in a local before it returns it.

  • fallow dead-code credits a member on an element of a typed array field of a class in another file. With days: Day[] on Month, fallow now credits Day members for m.days[0].getFirst(), m.days.some((day) => day.flag) and for (const day of m.days), where m is a parameter typed Month. The same is true for this.days[0]?.getFirst() in Month, and for a getter or a method that returns an array (get sorted(): Day[]). The extraction cache version is 427.

  • fallow dead-code credits a member of a page object that a Playwright fixture reaches through a property. With the fixture app: AppPage and readonly details: DetailsPage on AppPage, app.details.getParameter() in a test now credits DetailsPage.getParameter. A longer chain (app.details.input.select()) also gives the credit when each property has a declared type or a new initializer. The chain also credits each property that it reads: details on AppPage and input on DetailsPage.

  • fallow dead-code credits a member on an element of an array that a function in another file returns. With export const make = (): Sheet[] => ..., fallow now credits Sheet members for make().forEach((sheet) => sheet.member), for (const sheet of make()) and make()[0].member, and for the same reads through a local const sheets = make(). A function that returns one instance gives no credit for a read on an element of its result.

  • React Router route files that an imported module declares are now entry points. A routes.ts file can import route arrays from other local modules and spread or merge them, for example [...coreRoutes] or mergeRoutes(coreRoutes, extendedRoutes). The React Router plugin read only the expressions in routes.ts itself. fallow dead-code then reported the route files of the imported modules as unused files, and also each file that only these route files import. The plugin now follows spread elements, named and default imports from relative specifiers, export { a as b }, export { a } from and export * from. A .js specifier that names a TypeScript source also resolves. Route file paths in an imported module stay relative to the directory of routes.ts, the same as React Router resolves them.

  • React Router route files with brackets in their path are now entry points. The plugin used each route file path as a glob pattern. A segment such as [workspaceSlug] then matched one character and not the directory with that name, so fallow dead-code reported the route file as unused. The plugin now escapes the path before it makes the pattern.

  • A warm run keeps its cache hits when the extraction cache is near its size limit. All entries that one run sees get the same seen stamp. When a save went past 80% of the limit, eviction removed entries down to 60% of the limit, also entries of the latest run. The next run parsed those files again, and the save removed the same entries again. On a project with a working set a little over 80% of the limit, each warm run parsed about 40% of the files again. Eviction now removes the entries that the latest run did not see down to 60% of the limit. It removes entries of the latest run only down to the limit itself. A working set that fits in the limit now stays in the cache, so .fallow/cache.bin can stay near the full limit.

  • Clone locations are correct on a line with non-ASCII text. The duplication detector gave start_col and end_col as character counts, but the LSP read them as byte columns. Thus a non-ASCII character before a clone on the same line moved the clone diagnostic and hover in the editor. The detector now gives byte columns, the same as the other analyses. A consumer that read these JSON fields as character columns must change. (Closes #​4378)

  • SARIF region columns now count UTF-16 code units. A SARIF run without columnKind counts startColumn in UTF-16 code units. Fallow gave the byte column plus 1, so on a line with non-ASCII text before a finding, a code scanning view showed the finding too far to the right. All SARIF output now converts the column with the source line: dead code, dupes, health, security, flags, flag retirement and fallow report --from. A byte order mark on line 1 does not count. The fingerprints keep the byte column, so existing alerts keep their identity. JSON column fields do not change. (Closes #​4405)

  • The changed-code scope of fallow mutation detects renamed and copied files. Before, the scope counted each line of a copied or moved file as new code, so unchanged copied lines got mutants. Now, when Git detects the source of a copy, only the edited lines of the copy are in scope. A new file stays in scope. A diff that you supply keeps its literal meaning. (Closes #​4364)

  • fallow mutation keeps the scope of a diff that you supply. Before, when the paths of an explicit patch did not match the project, the run used a generated Git diff in its place. Now the CLI exits with code 2 and a structured error before any test runs. A patch on stdin and --diff-file - keep their hunks, also when the scope is empty.

  • A wildcard exports target in build output now maps to its source files. A package with "./*": "./esm/*.mjs" or "./*": "./dist/*.js" exports each file under src/, but fallow matched the target only against files on disk. Thus the public subpath files were unused files, and their exports were unused exports. A wildcard target now follows the rule of a single package entry: the last dist, build, out, esm or cjs directory becomes src, and a JavaScript output extension becomes each source extension. A lib target maps into src only when it matches no file. A target at the package root such as ./*.js maps into src. When src has a match, the target still matches the files directly in the package root, but not the files in other directories such as scripts/. The root package of a project now also expands its wildcard exports targets, and a match skips the files of nested packages that have a name.

  • Plugin entry points apply again in a monorepo with 200 or more workspaces. All plugin entry globs compiled into one glob set, with a copy for each workspace package. With 200 or more workspaces, the set went past the size limit of the glob engine and did not compile. Then no plugin entry point applied, and fallow dead-code reported test files, pages and config files as unused. Fallow now groups the globs by their literal directory prefix and compiles each group on its own.

  • An import into the unbuilt output folder of a workspace package now resolves to the source file. A deep import such as lib/dist/util, and a require of a package root whose main names unbuilt output in a custom outDir, were unresolved. The imported source file was then reported as unused. Fallow now maps these paths with the same output-to-source rules as a main or exports target. The tsconfig output map also applies each parent of an array extends in order, and matches a package extends subpath against * patterns in exports.

  • Workspace entry point paths no longer keep a ./ segment. fallow list --entry-points showed packages/a/./index.js for the entry ./index.js of a workspace package. The path is now packages/a/index.js, the same as in the other reports.

  • fallow mutation inspect and fallow mutation tasks show the code of a mutant in a file with a UTF-8 byte order mark. The run instruments a file without its byte order mark, but inspect and tasks instrumented the file with the mark. Thus each mutant of such a file was stale with mutant-not-reproduced. They now drop the mark first.

  • return-value now mutates a concise arrow body that is a call, such as () => save(). Before, such a body got no mutant.

  • A branch or operand constant now gives no equivalent mutant through !. !(items.length >= 0) || flag on an array no longer gives false || flag.

  • A mutant at the start of an expression statement now keeps the grouping of the rest of the statement. The guard against a join with the line before is now void 0,, not void. Before, !x ? a() : b(); became void (x) ? a() : b() in the report and always ran b(), and f() ** 2; gave a syntax error. The replacement and the mutant_id of such a mutant change. A verdict for its old id becomes unmatched, so review it again.

  • fallow cloud login states the owner rule once when a member approves the code. Before, the message printed the reason from Fallow Cloud and then the same rule again from the CLI. The message now says once that only an organization owner can approve a key, and it names where to create one.

  • The API keys links of fallow cloud login and fallow cloud logout follow the cloud that the login uses. Before, they always pointed to fallow.cloud, also with a custom --api-url. The link now uses the dashboard address that the cloud returns for the login, when that address is on the host of the API. Otherwise it uses <host> for an API at api.<host>, or names the Settings page without a link. The CLI never prints a link with userinfo, a backslash, or plain http outside loopback, so a cloud cannot make the link open another site. The login line also drops control characters from the address and the code that the cloud returns.

  • fallow mutation suggests --concurrency 1 when a restarted companion fails in a parallel phase. Tests or a runner config that share files on disk can fail when workers run in parallel. The hint applies to init and dry-run failures and to missing covering tests. The original error and the timeout guidance stay. A serial phase keeps its diagnostics. (Closes #​4363)

Security
  • The MCP server requires rmcp 3.9 or later. rmcp 3.8.0 has the advisory RUSTSEC-2026-0462, unbounded response buffering in the Streamable HTTP clients. The Fallow MCP server uses stdio and does not enable these client transports, so it was not exposed. The update clears the dependency audit.

  • migrate, rule-pack init, coverage setup and ci-template gitlab --vendor no longer write through a symlink that leaves the project. A cloned repository could commit a file or directory that one of these commands writes as a symlink to a path outside the project. fallow migrate then created the outside file with the migrated config when .fallowrc.json, .fallowrc.jsonc or fallow.toml was a dangling link. rule-pack init did the same for rule-packs/team-policy.jsonc or a linked rule-packs directory, and it added the rulePacks entry to an outside file that the discovered config linked to, also when that config sits above --root in the same repository. coverage setup wrote setup.lock and setup.json through a link at .fallow or at the file itself, and it wrote docs/collect-coverage.md through a link, so it could overwrite an outside file. ci-template gitlab --vendor <dir> wrote ci/gitlab-ci.yml and ci/scripts/*.sh through a link below <dir>, and with --force it also overwrote an existing outside file. These commands now stop with an error when the target, or a directory on the way to it, is a symlink that resolves outside the project or a dangling link. migrate, rule-pack init, coverage setup and ci-template exit with code 2. When only the config file of rule-pack init links outside the project, the command writes the pack, leaves the config unchanged and prints the rulePacks entry to add by hand. This is the check that init and agent install already use. A link that stays inside the project still works, and migrate --dry-run is not affected.

  • The analysis caches no longer write through a symlink that leaves the project. A cloned repository could commit .fallow, a directory in it, or a cache file such as .fallow/.gitignore as a symlink to a path outside the project. An analysis run then wrote through that link. It created or overwrote the outside file with the content * (the cache .gitignore), or it wrote cache files into the outside directory. This affected the extraction cache, the graph cache, the duplication token cache, the churn cache, the flag age cache, the audit and security base snapshot caches, the walkthrough state of audit, and the default snapshot path of health --save-snapshot. Each of these writers now checks the path before it writes. When the cache directory, a directory below it, or the file is a symlink that resolves outside the project, or a dangling link, the writer skips the cache save and the analysis continues. health --save-snapshot stops with an error in that case, and audit-cache prune does not delete through such a link. A link that stays inside the project still works. When cache.dir or FALLOW_CACHE_DIR points outside the project, for example ../shared-cache, the check starts at that directory, so a link at the directory itself stays allowed.

  • The GitHub Action no longer writes its artifacts through a symlink. The Action now rejects a symlink or another file that is not a regular file at the path of an analysis, comment, review or scoped-result artifact, before it writes the artifact. The shared mutation helpers check their output paths before cleanup and publication. A regular artifact file from an earlier run is still replaced. The checks apply to a static checkout. They do not stop a path that another process replaces during the run.

  • A rejected dependency version in the installation diagnostics of the Action can no longer start a workflow command. The Action now escapes the version text, so a line break in it does not start another workflow command.
    (Related to #​4259)


Configuration

📅 Schedule: (UTC)

  • Branch creation
    • At any time (no schedule defined)
  • Automerge
    • At any time (no schedule defined)

🚦 Automerge: Disabled by config. Please merge this manually once you are satisfied.

♻ Rebasing: Whenever PR becomes conflicted, or you tick the rebase/retry checkbox.

🔕 Ignore: Close this PR and you won't be reminded about this update again.


  • If you want to rebase/retry this PR, check this box

This PR was generated by Mend Renovate. View the repository job log.

@coderabbitai

coderabbitai Bot commented Nov 13, 2026

Copy link
Copy Markdown

Important

Review skipped

Bot user detected.

To trigger a single review, invoke the @coderabbitai review command.

Run configuration

  • Source mode: Auto
  • Repository: Unavailable
  • Branch: Unavailable
  • File: No configuration file from this repository applied
  • Commit: Unavailable
  • Layers: defaults
  • Review profile: CHILL
  • Plan: Advanced
  • Run ID: 51bc4611-998d-49b1-a3be-f467f4b0f26d

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

  • Trigger review
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

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

@socket-security

Copy link
Copy Markdown

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff Package Supply Chain
Security
Vulnerability Quality Maintenance License
Updatedfallow@​3.41.0 ⏵ 3.42.0100 +110010096100

View full report

@pkg-pr-new

pkg-pr-new Bot commented Nov 13, 2026

Copy link
Copy Markdown

Open in StackBlitz

npm i https://pkg.pr.new/node-modules/urllib@980

commit: f30fa72

@codecov

codecov Bot commented Nov 13, 2026

Copy link
Copy Markdown

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 94.77%. Comparing base (610c713) to head (f30fa72).

Additional details and impacted files
@@           Coverage Diff           @@
##           master     #980   +/-   ##
=======================================
  Coverage   94.77%   94.77%           
=======================================
  Files          10       10           
  Lines         746      746           
  Branches      235      235           
=======================================
  Hits          707      707           
  Misses         36       36           
  Partials        3        3           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
  • 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

@fengmk2
fengmk2 merged commit 99741f8 into master Nov 13, 2026
21 checks passed
@fengmk2
fengmk2 deleted the renovate/fallow-3.x-lockfile branch November 13, 2026 10:05
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant