El dijous, 27 de setembre de 2018, a les 21:01:13 CEST, Friedrich W. H. Kossebau va escriure: > Am Donnerstag, 27. September 2018, 20:08:54 CEST schrieb Andreas Hartmetz: > > Today I fixed the third or so crash in KDE software due to the following > > pattern: > > > > const auto str = someString + anotherString; > > > > What happens is that the type of str ends up being QStringBuilder > > instead of QString. The QStringBuilder is destroyed after the semicolon, > > and all access to "str" produces undefined behavior. > > The QStringBuilder actually survives as long as str is in scope, but the > references it potentially takes from someString or anotherString (if e.g. > some > temporary QString object as returned from QStringLiteral or other QString- > returning methods) will be no longer valid after the scope of the expression > is left. > So if str is finally resolved to a real QString, those references are > dangling > and non-funny things happen. > > > While I'm not a particularly big fan of using auto to hide variable > > types anyway, this kind of usage is just wrong and must be avoided. > > Please take care. > > Care can be done e.g. by deploying clazy with auto-unexpected-qstringbuilder: > > clazy-standalone \ > -checks=auto-unexpected-qstringbuilder -p <buildroot> <projectroot> > > See > https://phabricator.kde.org/source/clazy/browse/master/docs/checks/README-auto-unexpected-qstringbuilder.md?as=remarkup > https://phabricator.kde.org/source/clazy/browse/master/README.md?as=remarkup > > One would recommend to run clazy over your code at least before releases, to > catch all kind of potential issues :)
Or since this is a crasher, just run your app and it'll crash? Or even better, add autotests that exercise the code and they'll crash too? Cheers, Albert > > Cheers > Friedrich > > >
