gecko-dev/js
brendan%mozilla.org 43c254feee Fix 97540, r=rginda, sr=jband, a=asa:
- The most significant fix, to keep JSStackFrame.spbase, the operand stack base pointer for an active frame, null except when there is an operand stack allocated and in use by js_Interpret.  Previously, spbase would point after args and local vars (if any), then advance upon allocation of the (possibly discontiguous) operand stack space.  This made for a fatal ambiguity: js_AllocStack, called by XPConnect, could not tell when there was allocated operand stack space above the frame's sp, which needs to be set to a known (JSVAL_VOID) state for exact GC to work.  Now, the GC doesn't have to mark any operand stack space for a frame whose spbase is null, and js_AllocStack doesn't need to void any unused space for such a frame.

- Fixes to reload the JSRuntime's callHook or executeHook after calling or executing, in case the debugger removes the hook.  In which case, it must clean up any dynamic memory held by hookData, but in any event, in which case the engine must not call the post-call or post-execute hook.

- While debugging with rginda, I was horrified to see his trivial testcase function, expressed as a lambda, fail to be invoked using the "inline_call" machinery in js_Interpret (which avoids js_Interpret recursion through js_Invoke for most JS functions).  The problem was a test of fun->flags == 0 conditioning the /* inline_call: */ code.  Since that test was written, at least one JSFUN_* flag (JSFUN_LAMBDA, used only for pretty-printing or accurate decompilation) has been added.  But all along, that test was an over-optimization (testing against 0 without &'ing certain flags), making for an accident waiting to happen -- which did happen.  The relevant flags are JSFUN_HEAVYWEIGHT (set by the compiler when a function calls eval, uses with, or otherwise needs an activation object for its scope; if lightweight, the compiler can see the function's scope and eliminate it via specialized bytecodes) and JSFUN_BOUND_METHOD (for Java method calls, where |this| binds statically to the instance, not dynamically to the calling expression reference's base object, as in JS).
2001-08-31 21:25:26 +00:00
..
benchmarks
js2 Added fdlibm_ns to build, stopped warnings from formmatter.h 2001-07-18 22:34:09 +00:00
jsd - not built - 2001-08-28 22:03:44 +00:00
jsdj
jsj
macbuild Bug#81373 (mac static build) 2001-08-27 23:45:59 +00:00
ref
rhino Port of performance fixes from Monkey (see bug #85721). 2001-08-27 18:03:19 +00:00
src Fix 97540, r=rginda, sr=jband, a=asa: 2001-08-31 21:25:26 +00:00
tests Corrected testcase so it will work in Rhino as well as SpiderMonkey. 2001-08-31 06:10:13 +00:00
.cvsignore
landbranch.pl
Makefile.in bug 76050, build js/jsd from toplevel makefile so we're sure we've got an xpidl compiler. r=cls 2001-05-15 20:38:26 +00:00
makefile.win