tcanabrava added a comment.
In D26868#602162 <https://phabricator.kde.org/D26868#602162>, @kossebau wrote: > In D26868#602150 <https://phabricator.kde.org/D26868#602150>, @dfaure wrote: > > > There is indeed a QString overload for concatenating QLatin1String, but it will have to be converted char-by-char (from 8 bits to 16 bits), so isn't it faster to concatenate QStringLiterals? > > > I have never seen numbers, but any time I asked someone who might have seen numbers, they told me: QStringLiteral only for final strings not further modified or given to operations. Sadly cannot find references now, but I asked often the recent years. > Looking at the code, the `qt_from_latin1` helper method might be as efficient on most processors as the `memcpy` is perhaps, and making up for any other overhead the QStringLiteral-generated code has, perhaps atomic stuff around the generated QString instance is expensive compared to the copy logic? > > But ready to hear from real Qt experts once more, perhaps with numbers this time :) I checked here with a really big xml file and the difference seems negligible. ( QStringLiteral 0.0s and QLatin1String 0.0s ) - I'd use QStringLiteral just for the sake of consistency across the codebase, tough REPOSITORY R237 KConfig REVISION DETAIL https://phabricator.kde.org/D26868 To: tcanabrava, ervin, dfaure Cc: kossebau, kde-frameworks-devel, LeGast00n, GB_2, michaelh, ngraham, bruns
