On Monday 26 November 2012, Dominique Leuenberger a.k.a DimStar wrote:
Quoting Ruediger Meier
: On Monday 26 November 2012, Dominique Leuenberger a.k.a DimStar wrote:
Quoting Ruediger Meier
: On Monday 26 November 2012, Dominique Leuenberger a.k.a DimStar
wrote:
Quoting Ilya Chernykh
: On Monday 26 November 2012 13:36:46 Dominique Leuenberger a.k.a
DimStar wrote: > > I have a package that brings about 50 errors > > "no-return-in-nonvoid-function". It is very difficult to > > fix all these cases (not to mention I do not know C++ and > > can only guess where to insert "0" and where "NULL") > > and also I am afraid this can break all the existing and > > future patches to this package. Previously I could disable > > it by setting the badness > > to 0 but it seems this does not work any more. What should > > I do? > > You should contact upstream and fix the issues... > > If you really 'must' disable this (there is no chance for > the package to enter Factory though!) you can add > #!BuildIgnore: brp-check-suse > to the .spec file...
I know I can buildrequire -post-build-checks (was this package renamed?), but this will disable all ckecks. I want other checks in place, but only this one removed (why it was transferred from rpmlint to brp by the way?). Currently I added "-Wno-return-type" to the compiller command line, but this still seems to be too broad. Still looking for more specific method.
The most specific one is: patch the source and fix the underlying issue. The no-return-in-nonvoid has been a brp check for as long as I remember...
Best thing is really: bring it to upstream to get a proper fix to it...
But don't forget that "no-return-in-nonvoid" is just a warning and not an error. In many cases you can safely ignore it.
A warning by the compiler does not mean you don't have to care for it.. It means the programmer is responsible to know what is happening in this code and there is NO guarantee between gcc versions that this is not 'just changing'.
The only valid solution is to fix the code.
See for example this code piece (C++)
### #include <iostream>
int foo() { int a = 5; int b = a + 1; }
int main() { std::cout << foo() << std::endl; } ###
What is the recommended output of this? this example is somewhat simple enough to understand what is going on... BUT the bad thing is, if you ever extend the function foo() with anything AFTER, you migh get a new return value, which is not what you wanted... hence the warning.
Best regards, Dominique (who wishes upstreams would learn about -Werror -Wall)
The example below also gives you "control reaches end of non-void function" and obviously the compiler (gcc 4.5.1) is wrong. A valid solution here is to ignore the warning or to fix gcc.
int main() { int i = 1; if (i == 1) { return 1; } }
(This is a stupid trivial example. You can imagine that there are also real world examples where the programmer knows for sure that the warning is invalid.)
Right.. so both our examples show: it's not just 'sane to ignore' the warning but the programmer better knows what he does and is conscious about it... I'd come across MANY of those errors were the programmer simply was not aware of side effects... often it is expected that the return value be 0 in this case (which, by the way, is a valid assumption for int main() {})
btw, another common case for this error is:
#include
int foo(int i) { if (i > 0) return 1; assert(0) }
Now, arguably, you can say the end is never reached... and the compiler is just annoying... yt, the compiler can't know...
but if I use gcc -DNDEBUG, then the compiler is right...
I do not agree. The fact that the programmer placed an assert here means that he carefully thought about it. It would be a bug somewhere else in case it would be violated nevertheless. But see a real world example in package devel:libraries:c_c++/hdf5 Most of the fixes in hdf5-non_void_return.patch are non-sense (just to avoid the false warning). Upstream would not accept this patch for sure. Note that probably the hdf5 guys don't even use gcc as primary target compiler and other compilers may be more smart with their warnings. cu, Rudi -- To unsubscribe, e-mail: opensuse-packaging+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-packaging+owner@opensuse.org