[cfe-commits] r162833 - /cfe/trunk/docs/AutomaticReferenceCounting.html
John McCall
rjmccall at apple.com
Wed Aug 29 01:32:31 PDT 2012
Author: rjmccall
Date: Wed Aug 29 03:32:30 2012
New Revision: 162833
URL: http://llvm.org/viewvc/llvm-project?rev=162833&view=rev
Log:
Clarify the point at which ARC destroys ivars vis-à-vis
[super dealloc]. rdar://problem/11141872
Modified:
cfe/trunk/docs/AutomaticReferenceCounting.html
Modified: cfe/trunk/docs/AutomaticReferenceCounting.html
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/docs/AutomaticReferenceCounting.html?rev=162833&r1=162832&r2=162833&view=diff
==============================================================================
--- cfe/trunk/docs/AutomaticReferenceCounting.html (original)
+++ cfe/trunk/docs/AutomaticReferenceCounting.html Wed Aug 29 03:32:30 2012
@@ -1611,6 +1611,36 @@
that <tt>NSObject</tt>'s <tt>dealloc</tt> would, which is outside the
scope of this document to describe.</p></div>
+<p>The instance variables for an ARC-compiled class will be destroyed
+at some point after control enters the <tt>dealloc</tt> method for the
+root class of the class. The ordering of the destruction of instance
+variables is unspecified, both within a single class and between
+subclasses and superclasses.</p>
+
+<div class="rationale"><p>Rationale: the traditional, non-ARC pattern
+for destroying instance variables is to destroy them immediately
+before calling <tt>[super dealloc]</tt>. Unfortunately, message
+sends from the superclass are quite capable of reaching methods in the
+subclass, and those methods may well read or write to those instance
+variables. Making such message sends from dealloc is generally
+discouraged, since the subclass may well rely on other invariants that
+were broken during <tt>dealloc</tt>, but it's not so inescapably
+dangerous that we felt comfortable calling it undefined behavior.
+Therefore we chose to delay destroying the instance variables to a
+point at which message sends are clearly disallowed: the point at
+which the root class's deallocation routines take over.</p>
+
+<p>In most code, the difference is not observable. It can, however,
+be observed if an instance variable holds a strong reference to an
+object whose deallocation will trigger a side-effect which must be
+carefully ordered with respect to the destruction of the super class.
+Such code violates the design principle that semantically important
+behavior should be explicit. A simple fix is to clear the instance
+variable manually during <tt>dealloc</tt>; a more holistic solution is
+to move semantically important side-effects out of
+<tt>dealloc</tt> and into a separate teardown phase which can rely on
+working with well-formed objects.</p></div>
+
</div>
</div> <!-- misc.special_methods -->
More information about the cfe-commits
mailing list