Repository navigation
Kotlin scope functions (run, let, apply, also, with) bind to unrelated Java methods #155
Description
What is wrong
Kotlin's scope functions are standard-library calls, but ripwire resolves calls by name. Once Kotlin
and Java share one call graph, x.run { … } binds to a same-named Java run() whenever the Kotlin
side defines no run of its own. apply, let, also and with do the same. Removing these
false edges is what makes --callers, --impact and the ranking trustworthy in mixed Java/Kotlin
codebases — Android above all.
The evidence
On square/retrofit, --callers=run goes from 2 on main to 5 with the Kotlin support PR
(#126). Two of the three new rows are .run { User(name) } lambdas in Kotlin serialization tests —
false edges. The third is a real block.run() on a Runnable, which the fix must keep.
The kit records why the obvious levers cannot fix this on their own: arity, receivers,
own-language-first, and the external-name veto are each examined and ruled out, with three
reproduction trees and their recorded output.
Size
Medium. One extracted per-call-site fact carried through the reference cache record (with parser
and cache version bumps, their mirrors and pins), a small sorted name table in
src/externalnames.h, a candidate filter beside keepOwnJvmLanguageCandidates, a new section in
test/kotlincheck.sh, and measurements on retrofit, nowinandroid and ktor.
Prerequisites
- PR feat(lang): Kotlin — vendored grammar, JVM-bridged call graph #126 (Kotlin support), merged 2026-09-11. The prompt describes the code feat(lang): Kotlin — vendored grammar, JVM-bridged call graph #126 adds and how to confirm
the integrated head. The arms go intotest/kotlincheck.sh, which feat(lang): Kotlin — vendored grammar, JVM-bridged call graph #126 adds. - PR fix(resolver): a declined call no longer reads as "no caller exists" — declined=, declined_calls=, and every call accounted for #136 (declined-call disclosure), merged into
main. The disposition arm relies on its
conservation line.
Where to start
prompts/help-wanted/kotlin-scope-functions.md is a self-contained prompt for a coding agent. It
covers the retrofit rows, how Kotlin calls are captured and resolved, the three reproductions, the
design space (the trailing-lambda fact, the name table, the rule, where refused sites go, and what
stays out of scope), the non-negotiables, red-first acceptance arms, the known traps, and what the
fix PR's description must contain. Like every prompt in prompts/, it ends by writing a plan and
stopping — a maintainer agrees the plan before any code is written.
Comment here to claim it.
Activity
joyful-ii-V-I commentedon Sep 16, 2026
Related, same failure class in TypeScript: #59 (built-in member calls binding unrelated functions) and its first step #163.
llvm-x86 commentedon Oct 24, 2026
I want to take this issue. I need a go-ahead on the plan first, as the kit asks.
I reproduced the false edge on current main.
--callers=runon the kit's tree gives two callers.build(kt/Use.kt:3, via=name) is false.kickis true.Plan:
RawRef,Referenceand the cache record.kCacheVersion,kParserVerand their mirrors. Re-pintest/qschemetrip.hashwith a RE-PIN LOG entry.run,let,apply,alsoandwithinsrc/externalnames.h.keepOwnJvmLanguageCandidatesinsrc/graph.h. A trailing-lambda call to one of these names does not bind to a Java method. Refused sites get a disposition under the fix(resolver): a declined call no longer reads as "no caller exists" — declined=, declined_calls=, and every call accounted for #136 conservation line.test/kotlincheck.sh.Questions:
I will write no code until a maintainer agrees.
joyful-ii-V-I commentedon Oct 24, 2026
Thanks @llvm-x86. It's yours, and after #423 we're glad to have you on another one. The plan is right in substance: the trailing-lambda fact, a name table and a site-keyed rule, with name-only vetoes rejected. Go ahead, with the updates below. The kit was written on 09-11, and main has moved since.
Your three questions
kCacheVersion+kIngestCacheVersionMirror+kMinRefRecordBytesmove together, and the extraction change moveskParserVer+kIngestParserVerMirror.kParserVer226 andkCacheVersion40, and train 35 has reserved values after that. So start atkParserVer228 andkCacheVersion41, and put theqschemetripre-pin in its own commit so it's cheap to redo.What changed on main since the kit
kMinRefRecordBytesis now 60 (3×u32 + 12×u8 + 9×str), not 39. Your fact is one moreu8, written and read besidememberCall/forwardedinwriteRef/readRef, withRawRef→Referencecopied iningest_model.h.externalnames.hnow has a builtin-method family:kPythonBuiltinMethodNames,kJsBuiltinMethodNames,kRubyBuiltinMethodNames,kCppStdMemberNamesandkRustStdMemberNames. Each table is generated by one command written in its provenance comment, and each has a sortednessstatic_assert. Give the Kotlin table the same treatment: cite the Kotlin stdlib docs page and version for the five scope functions, so the list can be regenerated and diffed, not hand-tuned.RawRef::memberCall(FE-A) now records "written on a receiver" for Go/JS/TS/Rust. Kotlin isn't in it. You don't need it for this fix, but it's the natural home for the receiver gap you'll disclose (tree C).Where the rule goes: fit it into an existing family, not a new standalone rule
docs/ARCHITECTURE.md(a seeds → hops → walks section lands with train 34) and CONTRIBUTING's feature-family map both push toward one mechanism per family, with languages differing by data. On main, three neighbours already cover "a call by name that cannot reach the in-tree target":FalseEdgeRules(graph.h, gatetest/falseedgecheck.sh). It removes candidates the call cannot reach by the language's own rules. When nothing survives, the site isexternal=with proof (oneC externalcensus row), andunresolved=without proof. It never declines and never retargets. Its STATED FLOORS header lists Kotlin as untouched today.BuiltinMethodGate(Python/JS/TS/Ruby): it declines a builtin-named member call unless the caller's file gives evidence for the target.StdMemberGate(C++/Rust): one row per language. Its comment says why Java/Kotlin get no name row, because in-tree types implementadd/sizeon purpose. That is the same reason your plan rejects a name-only veto.Your rule is FE-A-shaped. The trailing lambda is a call-site proof that this is the stdlib scope function, not a guess from the name. So please implement it as a Kotlin arm of
FalseEdgeRules:go);external=with proof, which also satisfies fix(resolver): a declined call no longer reads as "no caller exists" — declined=, declined_calls=, and every call accounted for #136's conservation line with no new disposition.Then update FE-A's STATED FLOORS header and the
#kotlin-extractionparagraph ofdocs/ARCHITECTURE.mdwith the floors you leave: receiver typing, SAM conversion, and parenthesised lambdas if you exclude them. If FE-A's structure fights you, say so in the plan reply before you write code. Keeping it besidekeepOwnJvmLanguageCandidatesis acceptable, but then the PR should say why it isn't an FE-A arm.One more thing to keep in mind: #59/#163 is the same false-edge class for TypeScript. If the table hangs off a per-language row (the
StdMemberRowpattern) keyed on a call-site fact, TypeScript can reuse it later as data, not as a second rule.Two coordination notes
test/kotlincheck.shis also being edited in feat(resolvers): order every multi-definition name answer with one tower; disclose splits in --lsp (RIPWIRE-RUBY C) #432 (a--lego=Markertie-break fix). Take the next free § when you open the PR, and expect a small rebase.src/model.h,src/ingest_cache.handsrc/quality.h). Starting now on current main is fine too; just merge main before you open the PR.The acceptance arms and measurements in the kit stand as written, red on a named main commit, plus
test/falseedgecheck.shif you take the FE-A route. Looking forward to the plan's numbers on retrofit, nowinandroid and ktor.fix(kotlin): a trailing-lambda scope function never binds a Java meth…
test(qschemetrip): re-pin for the redhat-et#155 placeholder versions
test(qschemetrip): re-pin for the redhat-et#155 placeholder versions
fix(kotlin): a trailing-lambda scope function never binds a Java meth…
test(qschemetrip): re-pin for the redhat-et#155 placeholder versions
test(qschemetrip): re-pin for the redhat-et#155 placeholder versions
fix(kotlin): keep a Kotlin receiver's proof for trailing-lambda scope…
refactor(kotlin): reuse isStrictlySortedTable and firstChildOfKind (r…
chore(docs): drop the regenerated COMMANDS.md and showcase capture (r…
Merge PR redhat-et#438 (@llvm-x86, Fixes redhat-et#155): Kotlin scope…
Rust: re-parse a file with its precise-capturing use<..> bounds blanked
Merge lane/queue-hop (Python queue hand-off hops) into integration/tr…