Returning the JSScript from JS::Evaluate is necessary for triggering the encoding of the bytecode. The JSScript is used by the StartIncrementalEncoding and FinishIncrementalEncoding which have to be called by the ScriptLoader in order to save the bytecode used during the execution.

This change reverts the modification added in Bug 852331. It is outdated for or multiple reasons: - We no longer have the bytecode analysis straight after the full parse. - We have Lazy parsing now. - We replaced the analysis by using IonBuilder at the time of the first Baseline compilation which uses a LifoAlloc, and which are discarded right away. (almost as mentioned in Bug 852331 comment 1) - LARGE_SCRIPT_LENGTH is above the HUGE_LENGTH heuristics of the CanCompileOffThread function, which takes another path for executing the code compiled off-main-thread.

Comment on attachment 8866845 [details] [diff] [review] Remove FINISH_LARGE_EVALUATE gc. Review of attachment 8866845 [details] [diff] [review]: ----------------------------------------------------------------- Looks good. There's always a small chance that changes like this will cause some kind of regression though so be prepared. ::: js/public/GCAPI.h @@ -107,5 @@ > D(INTER_SLICE_GC) \ > D(REFRESH_FRAME) \ > D(FULL_GC_TIMER) \ > D(SHUTDOWN_CC) \ > - D(FINISH_LARGE_EVALUATE) \ Please replace this with UNUSED2 so that existing telemetry data retains the same meaning.

For Bug 1364117, I will also need to insert a call to StartIncrementalEncoding between the compilation and the execution. Doing it in the Evaluate function might add much more API changes, and way to ignore them in existing functions. So I will keep the current patch as-is, and split the Evaluate function call into CompileScript & JS_Execute in the dom/base/nsJSUtils.cpp file.

Pushed by npierron@mozilla.com: https://hg.mozilla.org/integration/mozilla-inbound/rev/9414aca02ae0 Remove FINISH_LARGE_EVALUATE gc. r=jonco