<html><body><div style="color:#000; background-color:#fff; font-family:arial, helvetica, sans-serif;font-size:10pt"><div><span>Sure, I started a previous thread about this:</span></div><div><span><br></span></div><div><span><a href="http://lists.cs.uiuc.edu/pipermail/cfe-dev/2011-September/016941.html">http://lists.cs.uiuc.edu/pipermail/cfe-dev/2011-September/016941.html</a><br></span></div><div><a href="http://lists.cs.uiuc.edu/pipermail/cfe-dev/2011-September/016942.html">http://lists.cs.uiuc.edu/pipermail/cfe-dev/2011-September/016942.html</a><br></div><div><a href="http://lists.cs.uiuc.edu/pipermail/cfe-dev/2011-September/016944.html">http://lists.cs.uiuc.edu/pipermail/cfe-dev/2011-September/016944.html</a><br></div><div><a href="http://lists.cs.uiuc.edu/pipermail/cfe-dev/2011-September/016970.html">http://lists.cs.uiuc.edu/pipermail/cfe-dev/2011-September/016970.html</a><br></div><div><br></div><div>It boils down to two problems I've
had:</div><div><br></div><div>* If there is a double anywhere in a struct, MSVC pads the vftbl pointer.</div><div>* The <span class="Apple-style-span" style="font-family: monospace; white-space: pre; font-size: 16px; ">IsPODForThePurposeOfLayout</span> check is invalid for MS targets and so you get a different layout depending upon whether your type has a struct or not.</div><div><br></div><div>I suggested a simple patch in my last post that fixes these. My test cases so far are here:</div><div><br></div><div><a href="https://bitbucket.org/dwilliamson/clreflect/src/tip/src/clReflectTest/TestOffsets.cpp">https://bitbucket.org/dwilliamson/clreflect/src/tip/src/clReflectTest/TestOffsets.cpp</a><br></div><div><br></div><div>my patch works for all of them but unfortunately I haven't had time to sync to the lastest build of clang so I'm not in a position to submit a patch.</div><div><br></div><div>It would be nice if we could figure something out
that solves the biggest number of cases :)</div><div><br></div><div>Cheers,</div><div>- Don</div><div><br></div><div style="font-size: 10pt; font-family: arial, helvetica, sans-serif; "><div style="font-size: 12pt; font-family: 'times new roman', 'new york', times, serif; "><font size="2" face="Arial"><hr size="1"><b><span style="font-weight:bold;">From:</span></b> r4start <r4start@gmail.com><br><b><span style="font-weight: bold;">To:</span></b> Don Williamson <don.williamson@yahoo.com><br><b><span style="font-weight: bold;">Cc:</span></b> cfe-dev@cs.uiuc.edu<br><b><span style="font-weight: bold;">Sent:</span></b> Thursday, September 15, 2011 12:06 PM<br><b><span style="font-weight: bold;">Subject:</span></b> Re: [cfe-dev] Implementation of MSRecordLayoutBuilder.<br></font><br><meta http-equiv="x-dns-prefetch-control" content="off"><div id="yiv787442229">
On 15/09/2011 14:52, Don Williamson wrote:
<blockquote type="cite">
<div style="color: rgb(0, 0, 0); background-color: rgb(255, 255, 255); font-size: 10pt; font-family: arial, helvetica, sans-serif; ">
<div><span>Hi Dmitry,</span></div>
<div><span><br>
</span></div>
<div><span>I'm really interested in the original test cases you
have that failed with the original implementation of the
record layout builder - could you share them?</span></div>
<div><span><br>
</span></div>
<div><span>How does your current patch work with the following
test:</span></div>
<div><span><br>
</span></div>
<div><span>#pragma pack(push, 8)</span></div>
<div><span>class B {</span></div>
<div><span>public:</span></div>
<div> virtual void b(){}</div>
<div> int a;</div>
<div> double b;</div>
<div><span>};</span></div>
<div><span>#pragma pack(pop)</span></div>
<div><span><br>
</span></div>
<div><span>Have you also tried test cases with and without
constructors?</span></div>
<div><span><br>
</span></div>
<div><span>Thanks,</span></div>
<div><span>- Don</span></div>
<div><br>
</div>
<div style="font-size: 10pt; font-family: arial, helvetica, sans-serif; ">
<div style="font-size: 12pt; font-family: times, serif; "><font face="Arial" size="2">
<hr size="1"><b><span style="font-weight:bold;">From:</span></b>
r4start <a rel="nofollow" class="yiv787442229moz-txt-link-rfc2396E" ymailto="mailto:r4start@gmail.com" target="_blank" href="mailto:r4start@gmail.com"><r4start@gmail.com></a><br>
<b><span style="font-weight:bold;">To:</span></b> John
McCall <a rel="nofollow" class="yiv787442229moz-txt-link-rfc2396E" ymailto="mailto:rjmccall@apple.com" target="_blank" href="mailto:rjmccall@apple.com"><rjmccall@apple.com></a><br>
<b><span style="font-weight:bold;">Cc:</span></b>
<a rel="nofollow" class="yiv787442229moz-txt-link-abbreviated" ymailto="mailto:cfe-dev@cs.uiuc.edu" target="_blank" href="mailto:cfe-dev@cs.uiuc.edu">cfe-dev@cs.uiuc.edu</a><br>
<b><span style="font-weight:bold;">Sent:</span></b>
Thursday, September 15, 2011 11:20 AM<br>
<b><span style="font-weight:bold;">Subject:</span></b>
Re: [cfe-dev] Implementation of MSRecordLayoutBuilder.<br>
</font><br>
On 15/09/2011 00:39, r4start wrote:<br>
> On 15/09/2011 00:36, John McCall wrote:<br>
>> On Sep 14, 2011, at 1:33 PM, r4start wrote:<br>
>> <br>
>>> On 15/09/2011 00:24, John McCall wrote:<br>
>>>> On Sep 14, 2011, at 1:23 PM, r4start wrote:<br>
>>>>> On 14/09/2011 21:53, John McCall wrote:<br>
>>>>>> On Sep 14, 2011, at 4:05 AM,
r4start wrote:<br>
>>>>>> <br>
>>>>>>> On 14/09/2011 00:42, r4start
wrote:<br>
>>>>>>>> On 14/09/2011 00:37, Eli
Friedman wrote:<br>
>>>>>>>>> On Mon, Sep 12, 2011 at
11:45 PM, r4start<<a rel="nofollow" ymailto="mailto:r4start@gmail.com" target="_blank" href="mailto:r4start@gmail.com">r4start@gmail.com</a>>
wrote:<br>
>>>>>>>>>> On 12/09/2011
21:57, r4start wrote:<br>
>>>>>>>>>>> On 12/09/2011
21:37, Eli Friedman wrote:<br>
>>>>>>>>>>>> On Mon, Sep
12, 2011 at 8:56 AM, r4start<<a rel="nofollow" ymailto="mailto:r4start@gmail.com" target="_blank" href="mailto:r4start@gmail.com">r4start@gmail.com</a>>
wrote:<br>
>>>>>>>>>>>>> Hi,<br>
>>>>>>>>>>>>> I have
some prototype code for MSRecordLayoutBuilder. I test<br>
>>>>>>>>>>>>> this
code<br>
>>>>>>>>>>>>> with
MSVS 2010 and \Zp8 compiler option. On simple examples it<br>
>>>>>>>>>>>>> seems
to<br>
>>>>>>>>>>>>> work
properly.<br>
>>>>>>>>>>>>> <br>
>>>>>>>>>>>>> Is it
interesting to the clang project?<br>
>>>>>>>>>>>> We
certainly want to fix any cases where clang does not do
struct<br>
>>>>>>>>>>>> layout
consistently with MSVC on Windows. It's hard to comment on<br>
>>>>>>>>>>>> your patch
without seeing it.<br>
>>>>>>>>>>>> <br>
>>>>>>>>>>>> -Eli<br>
>>>>>>>>>>> Now I'm not at
work. When I come to work I'll send a patch.<br>
>>>>>>>>>>> <br>
>>>>>>>>>>> - Dmitry<br>
>>>>>>>>>> Here is patch for
MSRecordLayoutBuilder.<br>
>>>>>>>>> I think it would be
better to integrate the checks into the main<br>
>>>>>>>>> RecordLayoutBuilder
class rather than subclassing it.<br>
>>>>>>>>> <br>
>>>>>>>>> Please include tests in
your patch. (See test/CodeGen/ms_struct.c for<br>
>>>>>>>>> an example.)<br>
>>>>>>>>> <br>
>>>>>>>>> I would like someone
more familiar with MSVC to review the actual<br>
>>>>>>>>> logic here (ping me if
nobody does within a few days, though, and I'll<br>
>>>>>>>>> try to review anyway).<br>
>>>>>>>>> <br>
>>>>>>>>> -Eli<br>
>>>>>>>> Ok, thanks for hint.<br>
>>>>>>>> <br>
>>>>>>>> -Dmitry.<br>
>>>>>>> I have a problem when writing a
test. Clang doesn't implement Microsoft<br>
>>>>>>> ABI and he crashes at code
generation stage.<br>
>>>>>>> Clang have dump-record-layouts
option but he prints AST layouts and<br>
>>>>>>> CodeGen layouts.<br>
>>>>>>> <br>
>>>>>>> Is there a way to save only AST
record layout to file and then check<br>
>>>>>>> or maybe my patch must include
CGRecordLayoutBuilder for Microsoft?<br>
>>>>>> It would be easy enough to add a
new testing switch if you'd like to get this in.<br>
>>>>> Can you give me some hint how can I do
this?<br>
>>>> Follow the code for -dump-record-layouts.<br>
>>> You propose to duplicate the output to a file
and then run FileCheck to check output, am I right?<br>
>> It would probably be better to run FileCheck
directly on the output, but<br>
>> I think you get the general idea.<br>
>> <br>
>> Whenever you get CGRecordLayoutBuilder working
adequately, you can rip<br>
>> out the new option and just use
-dump-record-layouts.<br>
>> <br>
>> John.<br>
> Ok, thanks. Tomorrow I'll do it.<br>
> <br>
Here is a test and patch.<br>
<br>
- Dmitry.<br>
<br>
<br>
_______________________________________________<br>
cfe-dev mailing list<br>
<a rel="nofollow" ymailto="mailto:cfe-dev@cs.uiuc.edu" target="_blank" href="mailto:cfe-dev@cs.uiuc.edu">cfe-dev@cs.uiuc.edu</a><br>
<a rel="nofollow" target="_blank" href="http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev">http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev</a><br>
<br>
<br>
</div>
</div>
</div>
</blockquote>
Hi,<br>
what I get from MSVS output :<br>
<br>
class B size(24):<br>
1> +---<br>
1> 0 | {vfptr}<br>
1> 8 | a<br>
1> | <alignment member> (size=4)<br>
1> 16 | c<br>
1> +---<br>
<br>
And this I get from clang with my patch:<br>
0 | class B<br>
0 | (B vtable pointer)<br>
4 | int a<br>
8 | double c<br>
sizeof=16, dsize=16, align=8<br>
nvsize=16, nvalign=8<br>
<br>
I'll try to fix this.<br>
<br>
What do you mean when you say "<span>Have you also tried test cases
with and without constructors?</span>", can you give me some
examples and I check them.<br>
<br>
- Dmitry.<br>
</div><meta http-equiv="x-dns-prefetch-control" content="on"><br><br></div></div></div></body></html>