Re: -Wsign-promo (Re: [BUILD FAILS DEBUG MODE] with test_strings_valuex.cxx)
On Sat, 2013-01-19 at 11:28 +0100, Lubos Lunak wrote: it incidentally triggers when passing bool to SvStream, because it doesn't have any overload for operator(bool), and int is chosen over unsigned char AKA sal_Bool , so Caolan added it in e8bbb76827dd7a0e30d7d1db34a812a84d85f390 - if SvStream gets overload for bool, the warning can be dumped Or am I missing something there? Yeah, IIRC I just wanted to see if there were more cases of bools being streamed out. Should have tweaked the stream class instead. C. ___ LibreOffice mailing list LibreOffice@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/libreoffice
Re: -Wsign-promo (Re: [BUILD FAILS DEBUG MODE] with test_strings_valuex.cxx)
On 01/19/2013 03:12 PM, Michael Stahl wrote: On 19/01/13 11:28, Lubos Lunak wrote: -Wsign-promo is a rather pointless warning these days (the section in the gcc manpage is a funny read and not only because it talks about Cfront). I've Some of the funniness is likely due to http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56067 Removal of -Wsynth from doc didn't remove example. added more overloads to silence it, but I rather wonder why we have this explicitly enabled at all. My hypothesis is like this: - the idea behind the warning is just nonsense (who cares to what integer type the value is promoted) overflow for unsigned integral types is defined by the C and C++ standards, while overflow for signed types is explicitly undefined. some implementations therefore assume that it does not occur and may remove tests on signed integers from the code that could only evaluate to true in case of an overflow in order to improve the all-important SPEC benchmark scores. -Wsign-promo is specifically about overload resolution, i'm not sure if that overflow problem would be relevant in this case but this is C++ so i'm never sure about anything :) No, undefined behavior due to signed overflow is not relevant here. For the integral promotion involved in overload resolution, no overflow can happen. (And the overflow that can happen with integral conversion, say a truncation from long to int, is implementation-defined rather than undefined, btw.) Stephan ___ LibreOffice mailing list LibreOffice@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/libreoffice
Re: -Wsign-promo (Re: [BUILD FAILS DEBUG MODE] with test_strings_valuex.cxx)
On 01/19/2013 11:28 AM, Lubos Lunak wrote: On Friday 18 of January 2013, julien2412 wrote: Hello, On pc Debian x86-64 with master sources after having runned make clean, I've got this: /home/julien/compile-libreoffice/libo/sal/qa/rtl/strings/test_strings_value x.cxx: In instantiation of ‘void testInt() [with T = rtl::OUString]’: /home/julien/compile-libreoffice/libo/sal/qa/rtl/strings/test_strings_value x.cxx:77:28: required from here /home/julien/compile-libreoffice/libo/sal/qa/rtl/strings/test_strings_value x.cxx:48:5: error: passing ‘unsigned char’ chooses ‘int’ over ‘unsigned int’ [-Werror=sign-promo] -Wsign-promo is a rather pointless warning these days (the section in the gcc manpage is a funny read and not only because it talks about Cfront). I've added more overloads to silence it, but I rather wonder why we have this explicitly enabled at all. I removed the explicitly enabled -Wsign-promo with http://cgit.freedesktop.org/libreoffice/core/commit/?id=488823a140217e393298bc83e75084041a85ed45 Remove -Wsign-promo, mainly as a quick way to silence the warnings that started to pop up after http://cgit.freedesktop.org/libreoffice/core/commit/?id=63bcb139b941a2eff1b5ad367046bca067e7d1f8 Replaced O[U]String::valueOf( static_cast ) with O[U]String::number(). That also allowed to revert http://cgit.freedesktop.org/libreoffice/core/commit/?id=1efe9a15f86a7a9dc08b57fd1dd12336522ba515 avoid -Wsign-promo warnings again with http://cgit.freedesktop.org/libreoffice/core/commit/?id=2ae6e77881028a287610f13d10f4c37242ff464b 'Revert avoid -Wsign-promo warnings.' The explicit -Wsign-promo had originally been added with http://cgit.freedesktop.org/libreoffice/core/commit/?id=e8bbb76827dd7a0e30d7d1db34a812a84d85f390 ensure correct export size type in stream operation, apparently to detect an SvStream::operator call with a bool argument that would pick the int overload instead of the unsigned char (aka sal_Bool) one (which would have happened to trigger the -Wsign-promo warning by luck, given that sal_Bool is unsigned char and not signed char, say). I assume that https://gerrit.libreoffice.org/#/c/1798/ Detect SvStream::operator calls with bool args w/o using -Wsign-promo would be a better fix for that (at least, a local make check kept succeeding with that patch applied here). Stephan - but it incidentally triggers when passing bool to SvStream, because it doesn't have any overload for operator(bool), and int is chosen over unsigned char AKA sal_Bool , so Caolan added it in e8bbb76827dd7a0e30d7d1db34a812a84d85f390 - if SvStream gets overload for bool, the warning can be dumped Ach, I failed to read Lubos' mail carefully until I now quote it for my reply, and all the stuff I write above has already been said... ___ LibreOffice mailing list LibreOffice@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/libreoffice
-Wsign-promo (Re: [BUILD FAILS DEBUG MODE] with test_strings_valuex.cxx)
On Friday 18 of January 2013, julien2412 wrote: Hello, On pc Debian x86-64 with master sources after having runned make clean, I've got this: /home/julien/compile-libreoffice/libo/sal/qa/rtl/strings/test_strings_value x.cxx: In instantiation of ‘void testInt() [with T = rtl::OUString]’: /home/julien/compile-libreoffice/libo/sal/qa/rtl/strings/test_strings_value x.cxx:77:28: required from here /home/julien/compile-libreoffice/libo/sal/qa/rtl/strings/test_strings_value x.cxx:48:5: error: passing ‘unsigned char’ chooses ‘int’ over ‘unsigned int’ [-Werror=sign-promo] -Wsign-promo is a rather pointless warning these days (the section in the gcc manpage is a funny read and not only because it talks about Cfront). I've added more overloads to silence it, but I rather wonder why we have this explicitly enabled at all. My hypothesis is like this: - the idea behind the warning is just nonsense (who cares to what integer type the value is promoted) - but it incidentally triggers when passing bool to SvStream, because it doesn't have any overload for operator(bool), and int is chosen over unsigned char AKA sal_Bool , so Caolan added it in e8bbb76827dd7a0e30d7d1db34a812a84d85f390 - if SvStream gets overload for bool, the warning can be dumped Or am I missing something there? -- Lubos Lunak l.lu...@suse.cz ___ LibreOffice mailing list LibreOffice@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/libreoffice
Re: -Wsign-promo (Re: [BUILD FAILS DEBUG MODE] with test_strings_valuex.cxx)
On 19/01/13 11:28, Lubos Lunak wrote: -Wsign-promo is a rather pointless warning these days (the section in the gcc manpage is a funny read and not only because it talks about Cfront). I've added more overloads to silence it, but I rather wonder why we have this explicitly enabled at all. My hypothesis is like this: - the idea behind the warning is just nonsense (who cares to what integer type the value is promoted) overflow for unsigned integral types is defined by the C and C++ standards, while overflow for signed types is explicitly undefined. some implementations therefore assume that it does not occur and may remove tests on signed integers from the code that could only evaluate to true in case of an overflow in order to improve the all-important SPEC benchmark scores. -Wsign-promo is specifically about overload resolution, i'm not sure if that overflow problem would be relevant in this case but this is C++ so i'm never sure about anything :) - but it incidentally triggers when passing bool to SvStream, because it doesn't have any overload for operator(bool), and int is chosen over unsigned char AKA sal_Bool , so Caolan added it in e8bbb76827dd7a0e30d7d1db34a812a84d85f390 - if SvStream gets overload for bool, the warning can be dumped Or am I missing something there? maybe there are more overloaded identifiers that simply haven't been called with wrongly promoted parameters yet? ___ LibreOffice mailing list LibreOffice@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/libreoffice
