<div dir="ltr"><div>David,</div><div><br></div><div>Thanks for the detailed explanation, I'm currently reworking most of the patches here. About the optimizations - I know the linear search is a bad optimization, but it adds legibility, that's what I tried to do.</div><div>I tried to emulate the python `if value in vector` that is really easy to read, compared to the current code.</div><div><br></div><div><br></div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Tue, Dec 24, 2019 at 9:40 AM David Faure <<a href="mailto:noreply@phabricator.kde.org">noreply@phabricator.kde.org</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><table><tbody><tr><td>dfaure requested changes to this revision.<br>dfaure added inline comments.
</td><td><a style="text-decoration:none;padding:4px 8px;margin:0px 8px 8px;float:right;color:rgb(70,76,92);font-weight:bold;border-radius:3px;background-color:rgb(247,247,249);background-image:linear-gradient(rgb(255,255,255),rgb(241,240,241));display:inline-block;border:1px solid rgba(71,87,120,0.2)" href="https://phabricator.kde.org/D26126" target="_blank">View Revision</a></td></tr></tbody></table><br><div><strong>INLINE COMMENTS</strong><div><div style="margin:6px 0px 12px"><div style="border:1px solid rgb(199,204,217);border-radius:3px"><div style="padding:0px;background:rgb(247,247,247) none repeat scroll 0% 0%;border-color:rgb(227,228,232);border-style:solid;border-width:0px 0px 1px;margin:0px"><div style="color:rgb(116,119,125);background:rgb(239,242,244) none repeat scroll 0% 0%;padding:6px 8px;overflow:hidden"><a style="float:right;text-decoration:none" href="https://phabricator.kde.org/D26126#inline-147537" target="_blank">View Inline</a><span style="color:rgb(75,77,81);font-weight:bold">kconfig_compiler.cpp:1020</span></div>
<div style="font:11px/15px "Menlo","Consolas","Monaco",monospace;white-space:pre-wrap;clear:both;padding:4px 0px;margin:0px"><div style="padding:0px 8px;margin:0px 4px;background:rgba(251,175,175,0.7) none repeat scroll 0% 0%">    <span></span><span><span>}</span></span><span> </span><span style="color:rgb(170,64,0)"><span>else</span></span><span> </span><span style="color:rgb(170,64,0)"><span>if</span></span><span> </span><span><span>(</span></span><span></span><span><span>type</span></span><span> </span><span style="color:rgb(170,34,17)"><span>==</span></span> <span>QLatin1String</span><span>(</span><span style="color:rgb(118,101,16)">"<span>font"</span></span><span></span><span><span>))</span></span><span> </span><span><span>{</span></span>
</div><div style="padding:0px 8px;margin:0px 4px;background:rgba(251,175,175,0.7) none repeat scroll 0% 0%">        <span></span><span style="color:rgb(170,64,0)"><span>return</span></span><span> </span><span><span>QStringLiteral</span></span><span></span><span><span>(</span></span><span></span><span style="color:rgb(118,101,16)"><span>"const QFont &"</span></span><span></span><span><span>);</span></span>
</div><div style="padding:0px 8px;margin:0px 4px;background:rgba(251,175,175,0.7) none repeat scroll 0% 0%">    <span></span><span><span>}</span></span><span> </span><span style="color:rgb(170,64,0)"><span>else</span></span><span> </span><span style="color:rgb(170,64,0)"><span>if</span></span><span> </span><span><span>(</span></span><span></span><span><span>type</span></span><span> </span><span style="color:rgb(170,34,17)"><span>==</span></span><span> </span><span><span>QLatin1String</span></span><span></span><span><span>(</span></span><span></span><span style="color:rgb(118,101,16)"><span>"rect"</span></span><span></span><span><span>))</span></span><span> </span><span><span>{</span></span>
</div><div style="padding:0px 8px;margin:0px 4px;background:rgba(151,234,151,0.6) none repeat scroll 0% 0%">    <span>       </span> <span>QLatin1String</span><span>(</span><span style="color:rgb(118,101,16)">"<span>double"</span></span><span></span><span><span>)}</span></span>
</div><div style="padding:0px 8px;margin:0px 4px;background:rgba(151,234,151,0.6) none repeat scroll 0% 0%">        <span></span><span><span>).</span></span><span></span><span><span>contains</span></span><span></span><span><span>(</span></span><span></span><span><span>type</span></span><span></span><span><span>))</span></span><span> </span><span><span>{</span></span>
</div><div style="padding:0px 8px;margin:0px 4px;background:rgba(151,234,151,0.6) none repeat scroll 0% 0%">    <span>    </span><span style="color:rgb(170,64,0)"><span>return</span></span><span> </span><span><span>type</span></span><span></span><span><span>;</span></span>
</div></div></div>
<div style="margin:8px 0px;padding:0px 12px"><p style="padding:0px;margin:8px">This linear search doesn't look like an optimization to me. It would be better to incorporate this into the map, so that a single search is performed, rather than two.</p></div></div><br><div style="border:1px solid rgb(199,204,217);border-radius:3px"><div style="padding:0px;background:rgb(247,247,247) none repeat scroll 0% 0%;border-color:rgb(227,228,232);border-style:solid;border-width:0px 0px 1px;margin:0px"><div style="color:rgb(116,119,125);background:rgb(239,242,244) none repeat scroll 0% 0%;padding:6px 8px;overflow:hidden"><a style="float:right;text-decoration:none" href="https://phabricator.kde.org/D26126#inline-147473" target="_blank">View Inline</a><span style="color:rgb(75,77,81);font-weight:bold">ervin</span> wrote in <span style="color:rgb(75,77,81);font-weight:bold">kconfig_compiler.cpp:1024</span></div>
<div style="margin:8px 0px;padding:0px 12px;color:rgb(116,119,125)"><p style="padding:0px;margin:8px">std::map looks like a bad idea here, either use QHash (preferred in massively based Qt code) or std::unordered_map.</p>

<p style="padding:0px;margin:8px">Also for both temporaries you pay the price of their allocation and deallocation at each call. Even a mutex used at each call would do better. ;-)</p></div></div>
<div style="margin:8px 0px;padding:0px 12px"><p style="padding:0px;margin:8px">I'm not 100% sure about std::map vs QHash given the number of elements, this would need benchmarking.</p>

<p style="padding:0px;margin:8px">But I agree 100% that compiler-generated thread safety around a static is NOTHING compared to amount of nodes allocated to fill in a map or hash.</p>

<p style="padding:0px;margin:8px"><a href="https://phabricator.kde.org/p/tcanabrava/" style="color:rgb(25,85,141);background-color:rgb(241,247,255);border:1px solid transparent;border-radius:3px;font-weight:bold;padding:0px 4px" target="_blank">@tcanabrava</a> Please have a look at <a href="https://gist.github.com/jboner/2841832" rel="noreferrer" target="_blank">https://gist.github.com/jboner/2841832</a><br>
Locking an available mutex is 4 times faster than even fetching data from main memory (i.e. data which isn't yet in a CPU cache). This is many orders of magnitude faster than creating a filling a map or a hash full of QStrings. On top of that, compilers don't even generate a full-blown mutex for threadsafe statics, they generate a three-state atomic (a bit like Q_GLOBAL_STATIC does) (3 because not created, being created, already created).</p>

<p style="padding:0px;margin:8px">The most performant solution is to turn the input string into a QByteArray and then perform a binary search in a C array of const char* (no allocations, even the very first time).<br>
The most readable (but still good, unlike the current code in git) solution is a static map.</p></div></div></div></div></div><br><div><strong>REPOSITORY</strong><div><div>R237 KConfig</div></div></div><br><div><strong>REVISION DETAIL</strong><div><a href="https://phabricator.kde.org/D26126" target="_blank">https://phabricator.kde.org/D26126</a></div></div><br><div><strong>To: </strong>tcanabrava, ervin, dfaure<br><strong>Cc: </strong>dfaure, ervin, GB_2, apol, kde-frameworks-devel, LeGast00n, michaelh, ngraham, bruns<br></div></blockquote></div>