This patch adds a new environment variable XPCOM_MEM_LOG_JS_STACK that
changes XPCOM leakchecking to record a JS stack for all objects, in
addition to a C++ stack. This is useful when a C++ object is being
leaked due to JS. The JS stack will be printed if the object leaks, if
it is used in combination with XPCOM_MEM_BLOAT_LOG=1 and
XPCOM_MEM_LOG_CLASSES=nsFoo, if nsFoo is the class of interest.
This patch moves a few XPConnect functions for recording the stack
into xpcpublic.h so they can be called from nsTraceRefcnt.cpp.
MozReview-Commit-ID: FX2QVCSXz4f
Most subclasses of Runnable don't bother to override AddRef and
Release, so XPCOM leak checking ends up reporting Runnable, which
makes it impossible to know what is actually leaking.
Each subclass of Runnable is already required to pass in the name of
the class, which is stored in the field mName. This patch changes
Runnable to use mName as the class name for XPCOM leak checking, thus
giving each subclass a specific name without needing to change the
hundreds of existing subclasses of Runnable.
The limitation of this approach is the classes that DO use
NS_IMPL_ADDREF/RELEASE_INHERITED end up using the same class name that
is used by the superclass AddRef/Release, but with a different size,
which causes assertions in the leak checker. To work around this, I
change NS_IMPL_ADDREF/RELEASE_INHERITED to not call into
NS_LOG_ADDREF/RELEASE for classes that are a subclass of
Runnable. This needs to use IsConvertible<> and not IsBaseOf<> because
the latter requires the classes involved to be defined, and headers
can use nsISupportsImpl.h without nsThreadUtils.h.
MozReview-Commit-ID: H0pgvwQSZAE
And remove unreachable code after MOZ_CRASH_UNSAFE_OOL().
MOZ_CRASH_UNSAFE_OOL causes data collection because crash strings are annotated to crash-stats and are publicly visible. Firefox data stewards must do data review on usages of this macro. However, all the crash strings this patch collects with MOZ_CRASH_UNSAFE_OOL are already collected with NS_RUNTIMEABORT.
MozReview-Commit-ID: IHmJfuxXSqw
MozStackWalk() is different on Windows to the other platforms. It has two extra
arguments, which can be used to walk the stack of a different thread.
This patch makes those differences clearer. Instead of having a single function
and forbidding those two arguments on non-Windows, it removes those arguments
from MozStackWalk, and splits off MozStackWalkThread() which retains them. This
also allows those arguments to have more appropriate types (HANDLE instead of
uintptr_t; CONTEXT* instead of than void*) and names (aContext instead of
aPlatformData).
The patch also removes unnecessary reinterpret_casts for the aClosure argument
at a couple of MozStackWalk() callsites.
It's silly to use prmem.h within Firefox code given that in our configuration
its functions are just wrappers for malloc() et al. (Indeed, in some places we
mix PR_Malloc() with free(), or malloc() with PR_Free().)
This patch removes all uses, except for the places where we need to use
PR_Free() to free something allocated by another NSPR function; in those cases
I've added a comment explaining which function did the allocation.
With frame pointer omission disabled we should always have usable stacks on Windows. This allows us to remove the MOZ_STACKWALKING define as it will always be enabled.
MozReview-Commit-ID: 54xs3Hf1r4P
CLOSED TREE
Backed out changeset 01cfc71ce542 (bug 1322735)
Backed out changeset 84c729c41230 (bug 1322735)
Backed out changeset b419aaefae95 (bug 1322735)
With frame pointer omission disabled we should always have usable stacks on Windows. This allows us to remove the MOZ_STACKWALKING define as it will always be enabled.
MozReview-Commit-ID: 54xs3Hf1r4P
NS_LogCOMPtrAddRef and NS_LogCOMPtrRelease always pass false to
GetSerialNumber, because they pass in everything they get without
regard to whether it is being logged or not, so they don't want to
create a serial number if none exists. This causes the assertions
added in bug 1309051 to be hit. To work around this, I hoist the
assertion into the other callers of this method. Two of them already
had this check, but it was non-fatal.
This also makes the asserts not happen in release builds, as I decided
it doesn't really matter what happens if somebody tries to use it
there.
Only about:bloat used NEW_STATS, so act like we have ALL_STATS
everywhere. A later patch will remove the mNewStats field.
MozReview-Commit-ID: LUJBpS3iYht
DumpStatistics does not track any objects that are created or
destroyed while it is running, which means that any subsequent calls
to it will produce incorrect results. This can lead to incorrect
positive or negative leaks being reported. See 1271182.
Now that about:bloat has been removed, DumpStatistics should only be
called once, during shutdown.
MozReview-Commit-ID: IjMkExeBRBr
This doesn't add comments for all of them, just some of those spread
out over many lines. The lack of these comments made writing part 1
more annoying.
GetSerialNumber accesses global state through gSerialNumbers. We call
GetSerialNumber under a lock when doing normal object refcount logging.
However, we call GetSerialNumber outside of a lock when we're tracing
individual classes for nsCOMPtr refcount logging, even if we don't
actually care about nsCOMPtr refcount logging. We should call it under
a lock always.
This will ensure we properly parse class names containing spaces.
Note that if a class name somehow ends up containing operator| then this will end up again silently failing.
