Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Bespoke, Artisanal Swift Static Analysis

JP Simard
September 10, 2018

Bespoke, Artisanal Swift Static Analysis

Have you ever fixed a bug in your code and wondered if you made that mistake elsewhere?
Wouldn't it be great if you could teach the Swift compiler to catch that bug next time you wrote it?
In this talk, we'll learn how you can build custom static analysis tools for Swift to make sure you never write the same bug twice.

JP Simard

September 10, 2018
Tweet

More Decks by JP Simard

Other Decks in Programming

Transcript

  1. Static analysis Static program analysis is the art of reasoning

    about the behavior of computer programs without actually running them. Anders Møller and Michael I. Schwartzbach h!ps://cs.au.dk/~amoeller/spa/spa.pdf
  2. How hard can it be? — Interprocedural Control Flow Graphs

    — Lattice Theory — Monotonicity — Constant Propagation Analysis — Very Busy Expressions — Transfer Functions — Closure Analysis for the λ-calculus — ! , " , # , $$$$$
  3. As its name implies, the Clang Static Analyzer is built

    on top of Clang and LLVM. https://clang-analyzer.llvm.org
  4. What would a Swi! static analyzer look like? — Similar

    to how the Clang Static Analyzer is built on Clang & LLVM — Built alongside the Swift compiler & LLVM — Require deep knowledge of the compiler internal architecture — Written in C++ — Require advanced static analysis concepts like CFG's
  5. Can we build Swi! static analysis that is: 1. Simple

    to extend? 2. Fast to build? 3. Easy to run? 4. Doesn't require deep knowledge of C++ or the Swift compiler internals?
  6. What are the ingredients? 1. A way to run over

    source files 2. A way to perform an action on the source 3. A way to get detailed information about the source 4. A way to extract interesting results from that information 5. A way to report these results to the user 6. A way to act on these results whenever possible
  7. Swi!Lint Already Provides 1. A way to run over source

    files 2. A way to perform an action on the source 3. A way to get detailed information about the source 4. A way to extract interesting results from that information 5. A way to report these results to the user 6. A way to act on these results whenever possible
  8. All that's le! is to find out how to get

    detailed information and extract interesting results
  9. { "key.substructure" : [ { "key.kind" : "source.lang.swift.decl.struct", "key.offset" :

    0, "key.nameoffset" : 7, "key.namelength" : 1, "key.bodyoffset" : 10, "key.bodylength" : 13, "key.length" : 24, "key.substructure" : [ { "key.kind" : "source.lang.swift.decl.function.method.instance", "key.offset" : 11, "key.nameoffset" : 16, "key.namelength" : 3, "key.bodyoffset" : 21, "key.bodylength" : 0, "key.length" : 11, "key.substructure" : [ ], "key.name" : "b()" } ], "key.name" : "A" } ], "key.offset" : 0, "key.diagnostic_stage" : "source.diagnostic.stage.swift.parse", "key.length" : 24 }
  10. [ { "key.column" : 8, "key.entities" : [ { "key.column"

    : 10, "key.kind" : "source.lang.swift.decl.function.method.instance", "key.line" : 2, "key.name" : "b()", "key.usr" : "s:4main1AV1byyF" } ], "key.kind" : "source.lang.swift.decl.struct", "key.line" : 1, "key.name" : "A", "key.usr" : "s:4main1AV" } ]
  11. { "key.accessibility" : "source.lang.swift.accessibility.internal", "key.annotated_decl" : "<Declaration>func b()<\/Declaration>", "key.filepath" :

    "\/path\/to\/main.swift", "key.fully_annotated_decl" : "<...see below...>", "key.kind" : "source.lang.swift.decl.function.method.instance", "key.length" : 3, "key.name" : "b()", "key.offset" : 32, "key.typename" : "(A) -> () -> ()", "key.typeusr" : "_T0yycD", "key.usr" : "s:4main1AV1byyF" } <decl.function.method.instance> <syntaxtype.keyword>func</syntaxtype.keyword> <decl.name>b</decl.name>() </decl.function.method.instance>
  12. Finding Dead Code ☠ struct Fika { func eat() {}

    } struct Bagel { func eat() {} // <- never used } let fika = Fika() fika.eat() print(Bagel())
  13. Finding Dead Code ☠☠ let allCursorInfo = file.allCursorInfo(compilerArguments: compilerArguments) let

    declaredUSRs = findDeclaredUSRs(allCursorInfo: allCursorInfo) // s:4main4FikaV, s:4main4FikaV3eatyyF, s:4main5BagelV, s:4main5BagelV3eatyyF, // s:4main4fikaAA4FikaVvp let referencedUSRs = findReferencedUSRs(allCursorInfo: allCursorInfo) // s:4main4FikaV, s:4main4fikaAA4FikaVvp, s:4main4FikaV3eatyyF, // s:s5printyypd_SS9separatorSS10terminatortF, s:4main5BagelV let unusedDeclarations = declaredUSRs.filter { !referencedUSRs.contains($0.usr) } // s:4main5BagelV3eatyyF
  14. Watch out! Not all code that isn't directly accessed is

    unused! enum WeekDay: String { // All these are used but not directly accessed case monday, tuesday, wednesday, thursday, friday } extension String { var isWeekday: Bool { return WeekDay(rawValue: self) != nil } }
  15. Dead code traps Turns out, there's a lot of valid

    Swift code that's never directly accessed. 1. Enum cases 2. Dynamic code: @IBOutlet, @IBAction, @objc 3. Protocol conformance code
  16. Unused Imports ⬇ import Dispatch // <- never used struct

    A { static func dispatchMain() {} } A.dispatchMain() // Would use `Dispatch` module: // // dispatchMain() // or // Dispatch.dispatchMain()
  17. Unused Imports ⬇⬇ 1. Find all imported modules import Dispatch

    // ^ cursor info request { "key.is_system" : true, "key.kind" : "source.lang.swift.ref.module", "key.modulename" : "Dispatch", "key.name" : "Dispatch" }
  18. Unused Imports ⬇⬇ 2. Find all referenced modules dispatchMain() //^

    cursor info request { "key.annotated_decl" : "<Declaration>func dispatchMain()...<\/Declaration>", "key.doc.full_as_xml" : "<Function>...<\/Function>", "key.filepath" : "...\/usr\/include\/Dispatch\/queue.h", "key.fully_annotated_decl" : "<decl.function.free>...<\/decl.function.free>", "key.is_system" : true, "key.kind" : "source.lang.swift.ref.function.free", "key.length" : 13, "key.modulename" : "Dispatch", "key.name" : "dispatchMain()", "key.offset" : 35595, "key.typename" : "() -> Never", "key.typeusr" : "_T0s5NeverOycD", "key.usr" : "c:@F@dispatch_main" }
  19. Configure # .swiftlint.yml included: - Source analyzer_rules: - explicit_self -

    unused_import - unused_private_declaration Run $ swiftlint analyze --autocorrect --compiler-log-path xcodebuild.log
  20. — Lyft's driver & passenger apps are built from over

    4,000 Swift files — 1,358 imports removed — 70 direct dependencies removed
  21. Rule Ideas — Inefficient code patterns (e.g. filter(...).first instead of

    first(where:)) — Conversion of iteration to functional code — Enforcing code architecture or style policies — Avoiding or encouraging certain function calls or types — Linting: explicit/implicit self, superfluous type annotations, etc.
  22. Final Thoughts — ! You don't need to be a

    compiler wizard to build Swift tools.
  23. Final Thoughts — ! You don't need to be a

    compiler wizard to build Swift tools. — Tools are just apps.
  24. Final Thoughts — ! You don't need to be a

    compiler wizard to build Swift tools. — Tools are just apps. — Xcode may be closed source, but most of the building blocks you need for custom Swift tooling is open source.
  25. Final Thoughts — ! You don't need to be a

    compiler wizard to build Swift tools. — Tools are just apps. — Xcode may be closed source, but most of the building blocks you need for custom Swift tooling is open source. — Other language communities build all their own tools. The Swift community should do the same!