Name anonymous export default functions and classes "default" in stack traces and inspect - #38517

Problem

• A stack frame inside an anonymous export default class { ... } renders as at new starDefault (p.mjs:1:55) until something reads the class's .name; after that the same frame renders at new default (...), which is also what node prints. D.name itself is always "default", so the frame name depends on evaluation order.
• Same for export default class extends Base {} (synthesized constructor), export default () => ... and export default (function () { ... }), and on every surface that shows a frame name: error.stack, CallSite#getFunctionName() under Error.prepareStackTrace, and the frames Bun's own uncaught error / unhandled rejection printer writes to stderr.
• Bun.inspect / console.log of the class or function prints [class starDefault], [class starDefault extends Base], [Function: starDefault], and a JSX element whose type is such a class prints <starDefault /> (node: [class default], [Function: default]). These stay wrong even after .name is read.
• Cause: JSC binds an anonymous default export to the private *default* identifier (vm.propertyNames->starDefaultPrivateName, whose string is "starDefault") and only substitutes "default" when it reifies the name property (JSFunction::reifyName). Bun's name lookups use an already reified own name property when there is one and otherwise take the name from the executable, where it is still the raw identifier:
  • src/jsc/bindings/ErrorStackTrace.cpp: functionName(vm, codeBlock) and the FinalizerSafety::MustNotTriggerGC branch read ecmaName() directly; functionName(vm, globalObject, object) gets it through JSC::getCalculatedDisplayName, whose last fallback is ecmaName() too.
  • src/jsc/bindings/bindings.cpp: JSC__JSValue__getName (inspect of functions and classes, describe(Class) names) through JSC::getCalculatedDisplayName, and JSC__JSValue__getNameProperty (JSX tag names) through jsExecutable()->name(). That fallback is only reached when JSFunction::name() returned "", which it does precisely for *default*, so it could only ever produce "" or "starDefault".

Fix

• Zig::functionNameForDisplay(vm, name) returns "default" when name is the *default* identifier's string and returns name unchanged otherwise. The five fallbacks above pass their result through it; nothing else about any lookup changes, so every other name comes out exactly as before (verified below for the builtin-module functions that the first revision of this PR got wrong).
• The check is name.impl() == starDefaultPrivateName.impl(), the same comparison Identifier::operator== performs in JSFunction::reifyName. The private identifier's string is its own unique StringImpl, so a function that is really named starDefault is left alone.
• Why this is correct: "default" is the name JSC itself reifies for these functions (JSFunction::reifyName and JSFunction::originalName apply the same substitution), so the lookups now return the same string before and after .name is materialized, and they match node for every shape above.
• The fuller shape of this fix is in JSC: JSFunction::calculatedDisplayName / getCalculatedDisplayName (and the callers of ecmaName() in StackFrame, SamplingProfiler and the inspector) have the same missing substitution, which is also why console.log(new D()) still prints starDefault {} (JSObject::calculatedClassName, tracked in inspect: print instances of an anonymous export default class as "default" #38530) and why --cpu-prof and the debugger show the same name. Fixing it in the fork would cover all of those and reduce this PR to its tests; that needs a WebKit PR and a version bump, and the two direct ecmaName() reads in ErrorStackTrace.cpp are Bun's own code either way. This PR fixes the surfaces Bun formats itself now; when the fork applies the substitution, functionNameForDisplay becomes an identity function and can be deleted along with its five call sites.
• Also fixed by the same change today, without a dedicated test here: the Received function starDefault suffix of native ERR_INVALID_ARG_TYPE messages, since determineSpecificType uses functionName(vm, globalObject, object); node errors: render a callable's name in determineSpecificType like node #38473 independently reworks that call site to read .name.
• Verified:
  • test/js/bun/test/stack.test.ts ("anonymous export default is named 'default'"): error.stack for all four shapes plus a control function really named starDefault, CallSite#getFunctionName(), before and after .name is read, and the frames printed for an uncaught throw and an unhandled rejection (those go through the MustNotTriggerGC branch; confirmed by reverting only that line, which fails exactly those two cases). All three fail on the released bun with starDefault and pass with this change.
  • test/js/bun/util/inspect.test.js ("anonymous export default class and function are named 'default'"): [class default], [class default extends Base], [Function: default], <default />. Fails on the released bun, passes with this change.
  • test/js/bun/util/inspect.test.js ("functions from built-in modules do not inspect as their internal variable name"): Bun.inspect(zlib.gzip) stays [Function] and Bun.inspect(fs.promises.readdir) stays [AsyncFunction]. Passes on the released bun and with this change; fails on the first revision of this PR ([Function: fn], [AsyncFunction: wrapped]).
  • Bun.inspect of zlib.gzip, zlib.gunzipSync, fs.promises.readFile / writeFile, Readable.prototype.map, util.promisify, events.once, Bun.serve, Array.prototype.map, new Promise resolvers, functions with an empty or non-empty displayName, inferred names and bound functions is byte-identical between the released bun and this change.
  • test/js/node/v8/capture-stack-trace.test.js, test/js/bun/util/inspect-error.test.js, reportError.test.ts, error-name-preservation.test.ts, test/js/bun/test/describe.test.ts, test/js/node/util/{bun-inspect,custom-inspect}.test.*, the two prepare-stack-trace / bindings-stack-trace regression tests and the rest of inspect.test.js pass (the two "minified file" cases in inspect-error.test.js fail identically on main under debug builds, where BUN_DEBUG enables showPrivateScriptsInStackTraces). error-gc-test.test.js and inspect-error-leak.test.js, which stress these lookups under Bun.gc, pass under the debug ASAN build with their timeouts raised.

Background

• *default*: the spec's name for the module-internal binding created by export default <anonymous declaration or expression>. JSC represents it with a private symbol whose description is starDefault; that symbol is also the StringImpl behind the identifier's string, which is what the identity comparison relies on.
• ecmaName(): the name JSC's parser records on a function's executable for the purposes of the ECMAScript name property (for example f in const f = () => {}). JSFunction::name() returns the binding name instead, and returns "" for *default*.
• Lazy name reification: JSC does not create a function's own name property until it is first looked up. Bun's lookups read that property when it exists (so Object.defineProperty(fn, "name", ...) wins) and otherwise fall back to the executable, which is why the output used to depend on whether .name had been read.
• FinalizerSafety::MustNotTriggerGC: Bun formats some stack traces while JSC is finalizing an error, or when printing an error that still owns its raw frames, where running getters or allocating on the JS heap is not allowed. That branch reads structure slots directly; this PR only wraps its existing ecmaName() return, which allocates nothing (defaultKeyword.string() is a VM-lifetime string).