On Tue, Sep 4, 2012 at 3:40 PM, Martin J. Evans <martin.ev...@easysoft.com> wrote: > I recently received an rt (https://rt.cpan.org/Ticket/Display.html?id=79190) > saying some code in DBD::ODBC was wrong and a patch to fix it. The code was: > > while (*cp != '\0') { > *cp++ = toupper(*cp); > } > > and the fix (which shouldn't be required as the above code is fine) was: > > while (*cp != '\0') { > *cp = toupper(*cp); > cp++; > }
The order of execution is undefined if you read a variable and assign to it in the same (sub)statement. The compiler is allowed to read the *cp on the right both before the cp++, as the only sequence point is the semicolon at the end. The patch as offered does the correct and safe thing of splitting it up in two statement. > The test code and his results are below. I can obviously avoid this issue > with a simple change to the code but how many other places might this occur? > Do I ignore this (as the reporter is happy to use a newer working compiler) > or do I somehow check for a broken compiler and abort the Makefile? Has > anyone else come across something like this and what did they do? You fix it. GCC will actually want you against this if you enable -Wall/-Wsequence-point Leon
