<html>
<head>
<meta content="text/html; charset=utf-8" http-equiv="Content-Type">
</head>
<body bgcolor="#FFFFFF" text="#000000">
On 6/15/2016 1:15 PM, Eric Fiselier wrote:<br>
<blockquote
cite="mid:CAB=TDAW+h_JNjzyQ8uzxH+SvAgv=p+yvYc6Pj_xGXWNytfGUdg@mail.gmail.com"
type="cite">
<div dir="ltr">On Wed, Jun 15, 2016 at 11:45 AM, Craig, Ben via
cfe-commits <span dir="ltr"><<a moz-do-not-send="true"
href="mailto:cfe-commits@lists.llvm.org" target="_blank">cfe-commits@lists.llvm.org</a>></span>
wrote:<br>
<div class="gmail_extra">
<div class="gmail_quote">
<blockquote class="gmail_quote" style="margin:0 0 0
.8ex;border-left:1px #ccc solid;padding-left:1ex">
<div bgcolor="#FFFFFF" text="#000000">
<p>Does this change (and the paper) permit declarations
like the following?</p>
<p> lock_guard<> guard();</p>
<p>If that syntax is allowed, then this is also likely
allowed...</p>
<p> lock_guard<>(guard);</p>
<p>I would really like the prior two examples to not
compile. Here is a common bug that I see in the
wild...</p>
<p> unique_guard<mutex>(some_member_mutex);</p>
<p>That defines a new, default constructed unique_guard
named "some_member_mutex", that likely shadows the
member variable some_member_mutex. It is almost never
what users want.<br>
</p>
</div>
</blockquote>
<div><br>
</div>
<div>I had no idea that syntax did that. I would have
assumed it created an unnamed temporary. I can see how
that would cause bugs.</div>
</div>
</div>
</div>
</blockquote>
It's also strong rationale for deduced constructor templates.
(<a class="moz-txt-link-freetext" href="http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2015/p0091r0.html">http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2015/p0091r0.html</a>)<br>
auto guard = unique_guard(some_member_mutex); <br>
You don't need to repeat types there, and it's very difficult to
forget to name the guard variable.<br>
<br>
<blockquote
cite="mid:CAB=TDAW+h_JNjzyQ8uzxH+SvAgv=p+yvYc6Pj_xGXWNytfGUdg@mail.gmail.com"
type="cite">
<div dir="ltr">
<div class="gmail_extra">
<div class="gmail_quote">
<div> </div>
<blockquote class="gmail_quote" style="margin:0 0 0
.8ex;border-left:1px #ccc solid;padding-left:1ex">
<div bgcolor="#FFFFFF" text="#000000">
<p> </p>
<p>Is it possible to have the empty template remain
undefined, and let the one element lock_guard be the
base case of the recursion? Does that help any with
the mangling?<br>
</p>
</div>
</blockquote>
<div>Nothing in the spec says the empty template should be
undefined. The default constructor on the empty template
is technically implementing "lock_guard(MutexTypes...)"
for an empty pack.</div>
<div>However your example provides ample motivation to make
it undefined. I'll go ahead and make that change and I'll
file a LWG defect to change the standard.</div>
<div><br>
</div>
<div>There is actually no recursion in the variadic
lock_guard implementation, so the change is trivial.</div>
<div><br>
</div>
<div>As for mangling I'm not sure what you mean? It
definitely doesn't change the fact that this change is ABI
breaking. (Note this change is not enabled by default for
that reason).</div>
</div>
</div>
</div>
</blockquote>
My thought regarding the mangling was that you could still provide a
one argument lock_guard, as well as a variadic lock_guard. The one
argument lock_guard would have the same mangling as before. I think
some of your other comments have convinced me that that won't work,
as I think the variadic lock_guard has to be made the primary
template, and I think the primary template dictates the mangling.<br>
<br>
I'm also going to guess that throwing inline namespaces at the
problem won't help, as that would probably cause compile-time
ambiguity.<br>
<br>
If I'm not mistaken, this only breaks ABI for those foolish enough
to pass a lock_guard reference or pointer as a parameter across a
libcxx version boundary. Does that sound accurate?
<pre class="moz-signature" cols="72">--
Employee of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project
</pre>
</body>
</html>