Bug 153626 – Web Inspector: Possible unexpected tail call optimization issue, sourceURL lost from eval() source

• Status: NEW
• Component: JavaScriptCore

[Joseph Pecoraro]

* SUMMARY
Possible unexpected tail call optimization issue, sourceURL lost from eval() source

* STEPS TO REPRODUCE
1. Attach the attached patch containing (LayoutTests/inspector/sampling-profiler/eval-source-url-2.html)
2. shell> run-webkit-tests LayoutTests/inspector/sampling-profiler/eval-source-url-2.html
  => FAIL

* NOTES
- Compare inspector/sampling-profiler/eval-source-url-2.html to inspector/sampling-profiler/eval-source-url.html
- The only difference is bar():

inspector/sampling-profiler/eval-source-url-2.html:
> function bar() {
>     for (let i = 0; i < 2; i++)
>         eval("//# sourceURL=eval.js\nfoo();");
> }

inspector/sampling-profiler/eval-source-url.html:
> function bar() {
>     eval("//# sourceURL=eval.js\nfoo();");
> }

For some reason, the Sampling Profiler doesn't see the "eval.js" sourceURL for the eval executable.

This showed up after no longer enabling the legacy profiling. One of the things that did was flip this bit:

    bytecompiler/BytecodeGenerator.cpp
    // FIXME: We should be able to have tail call elimination with the profiler
    // enabled. This is currently not possible because the profiler expects
    // op_will_call / op_did_call pairs before and after a call, which are not
    // compatible with tail calls (we have no way of emitting op_did_call).
    // https://bugs.webkit.org/show_bug.cgi?id=148819
    , m_inTailPosition(Options::useTailCalls() && !isConstructor() && constructorKind() == ConstructorKind::None && isStrictMode() && !m_shouldEmitProfileHooks)

Toggling that did in fact make the test pass again.

[Joseph Pecoraro]

Created attachment 270167 [details]
[ISSUE] Patch with new test that fails

[Radar WebKit Bug Importer]

<rdar://problem/25540228>