<html>
<head>
<meta content="text/html; charset=ISO-8859-1"
http-equiv="Content-Type">
</head>
<body text="#000000" bgcolor="#FFFFFF">
<div class="moz-cite-prefix">On 08.03.2013 23:25, Jordan Rose wrote:<br>
</div>
<blockquote
cite="mid:6FEFB13F-E234-4057-83A1-976596FA2A10@apple.com"
type="cite">
<meta http-equiv="Content-Type" content="text/html;
charset=ISO-8859-1">
<br>
<div>
<div>On Mar 8, 2013, at 11:22 , Anna Zaks <<a
moz-do-not-send="true" href="mailto:ganna@apple.com">ganna@apple.com</a>>
wrote:</div>
<br class="Apple-interchange-newline">
<blockquote type="cite">
<meta http-equiv="Content-Type" content="text/html;
charset=ISO-8859-1">
<div style="word-wrap: break-word; -webkit-nbsp-mode: space;
-webkit-line-break: after-white-space;">Anton,
<div><br>
</div>
<div>We've briefly discussed this with Jordan. Below are the
cases we should handle:</div>
<div><br>
</div>
<div>Case 1: The selector starts with "<span style="color:
rgb(209, 47, 27); font-family: Menlo;">dataWithBytesNoCopy</span>"
or "<span style="color: rgb(209, 47, 27); font-family:
Menlo;">initWithBytesNoCopy</span>" or "<span
style="color: rgb(209, 47, 27); font-family: Menlo;">initWithCharactersNoCopy</span>"
and "freeWhenDone" is not set to "0".</div>
<div>We should assume that the call preforms hold action
from malloc family. (So the pointer should not escape and
we should model this during ObjCMsgCall processing)</div>
</div>
</blockquote>
<br>
To explain the rationalization here, Anna pointed out that while
it's <i>likely</i> that methods with a "freeWhenDone:" or
"...NoCopy:" selector piece all behave as ownership-holders, we
can't actually prove it. In particular, if/when MallocChecker
gains the ability to reason about custom allocators, someone
could very well use "...NoCopy" to mean "I will free this using
my custom deallocator", and we don't want to produce a bogus
allocator mismatch bug in that case.</div>
<div><br>
</div>
<div>By the way, by "the selector starts with __", we mean the
existing logic of "the first selector piece is __", not "the
first selector piece starts with __".</div>
<div><br>
</div>
<div>Thanks for bearing with us.</div>
<div>Jordan</div>
</blockquote>
Aaaa, got it!<br>
Updated the patch.<br>
<br>
<div>> Case 1: The selector starts with "<span style="color:
rgb(209, 47, 27); font-family: Menlo;">dataWithBytesNoCopy</span>"
or "<span style="color: rgb(209, 47, 27); font-family: Menlo;">initWithBytesNoCopy</span>"
or "<span style="color: rgb(209, 47, 27); font-family: Menlo;">initWithCharactersNoCopy</span>"
and "freeWhenDone" is not set to "0".</div>
<div>> We should assume that the call preforms hold action from
malloc family. (So the pointer should not escape and we should
model this during ObjCMsgCall processing)</div>
<div>> </div>
<div>> Case 2: The selector contains "freeWhenDone", which is set
to '0'. </div>
<div>> The call results in no escape of the pointer.</div>
<div>> </div>
<div>> Case 3: Otherwise (this is not Case 1 nor Case 3) and the
selector ends with "NoCopy"</div>
<div>> The pointer should escape.</div>
<div>> </div>
<div>> Please, work off your previous patch, but change the logic
to handle these cases and include the tests. Also, I think it's <br>
> important to include a commit message with your patch so that
there would be no ambiguity in what the patch is trying to
accomplish.<br>
> Usually, the commit message and the tests would resolve that.<br>
I'll put this cases in a commit message.<br>
</div>
<pre class="moz-signature" cols="72">--
Anton</pre>
</body>
</html>