weird static cast
in [C++]
|
Prev: Global Variables
Next: std::ws
From: Ulrich Eckhardt on 22 Jun 2006 06:53 Hi! Following code, paraphrased from Boost[1] puzzles me a bit: #if BOOST_HAS_LONG_LONG size_t const npos = static_cast<size_t>(~0ULL); #else size_t const npos = static_cast<size_t>(~0UL); #endif What I wonder is if one couldn't simply write this: size_t const npos = ~static_cast<size_t>(0); I'm aware of the fact that using size_t const npos = ~0; might not work because '0' is an int and that might be smaller than a size_t. However, truncating the largest possible integer to a size_t is a waste on 32 bit platforms with long long support. Also, I think size_t is unsigned (correct me if I'm wrong!) and the almost idiomatic way to get the largest value of an unsigned integer is size_t const npos = size_t(-1); or is there a reason not to do this? How about using numeric_limits, btw? So I wonder if there's some corner case I didn't consider and above code is just an undocumented workaround... thanks Uli [1] Boost 1.33.1, in libs/regex/src/cregex.cpp [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ]
From: dan2online on 23 Jun 2006 06:12 Ulrich Eckhardt wrote: > Hi! > > Following code, paraphrased from Boost[1] puzzles me a bit: > > #if BOOST_HAS_LONG_LONG > size_t const npos = static_cast<size_t>(~0ULL); > #else > size_t const npos = static_cast<size_t>(~0UL); > #endif It is portable for all 32/64 bit data models. > > What I wonder is if one couldn't simply write this: > > size_t const npos = ~static_cast<size_t>(0); I think it is acceptable unless your compiler complains it. > > I'm aware of the fact that using > > size_t const npos = ~0; > > might not work because '0' is an int and that might be smaller than a > size_t. However, truncating the largest possible integer to a size_t is a > waste on 32 bit platforms with long long support. Also, I think size_t is On 64-bit systems, size_t may be a different data type to int. > unsigned (correct me if I'm wrong!) and the almost idiomatic way to get the > largest value of an unsigned integer is > > size_t const npos = size_t(-1); > > or is there a reason not to do this? How about using numeric_limits, btw? > You assume 2' complement. [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ]
From: Vladimir Marko on 24 Jun 2006 05:40 dan2online wrote: > Ulrich Eckhardt wrote: > > Hi! > > > > Following code, paraphrased from Boost[1] puzzles me a bit: > > > > #if BOOST_HAS_LONG_LONG > > size_t const npos = static_cast<size_t>(~0ULL); > > #else > > size_t const npos = static_cast<size_t>(~0UL); > > #endif > [...] > > Also, I think size_t is > > unsigned (correct me if I'm wrong!) and the almost idiomatic way to get the > > largest value of an unsigned integer is > > > > size_t const npos = size_t(-1); > > > > or is there a reason not to do this? How about using numeric_limits, btw? > > > > You assume 2' complement. There is no such assumption. 4.7/2 defines the result of the conversion to an unsigned integer type as congruent to the source value modulo 2^n where n is the number of bits of the _destination_ type. Since size_t is unsigned (by inclusion from the C standard), size_t(-1) shall be the largest value of type size_t (unless the size_t in question is not the std::size_t). Vladimir Marko [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ]
From: dan2online on 26 Jun 2006 04:51 Vladimir Marko wrote: > There is no such assumption. 4.7/2 defines the result of the conversion > to an unsigned integer type as congruent to the source value modulo > 2^n where n is the number of bits of the _destination_ type. Since > size_t is unsigned (by inclusion from the C standard), size_t(-1) shall > be the largest value of type size_t (unless the size_t in question is There is no standard for data type size_t, and the actual size of size_t is architecture-dependent. The assumption of size_t as an unsigned integer will lead to programming errors for 64bit architecture. Only 2 complement and size_t defined as unsigned, size_t(-1) is the largest value of type size_t. > not the std::size_t). size_t is not explicit std::size_t in the original post. size_t also doesn't mean unsigned int. [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ]
From: Martin Bonner on 27 Jun 2006 07:17
dan2online wrote: > Vladimir Marko wrote: > > > There is no such assumption. 4.7/2 defines the result of the conversion > > to an unsigned integer type as congruent to the source value modulo > > 2^n where n is the number of bits of the _destination_ type. Since > > size_t is unsigned (by inclusion from the C standard), size_t(-1) shall > > be the largest value of type size_t (unless the size_t in question is > > There is no standard for data type size_t, No, but the standard imposes some constraints. Specifically, it is unsigned, and it is integral. > and the actual size of > size_t is architecture-dependent. True. > The assumption of size_t as an > unsigned integer will lead to programming errors for 64bit > architecture. Assuming that size_t is <code>unsigned int</code> will lead to errors. However, the standard assures us that it will be an integer, and that it will be unsigned. > > Only 2 complement and size_t defined as unsigned, size_t(-1) is the > largest value of type size_t. Not so (I assume you meant "Only in that case that ..., will size_t(-1) ....". As explained above, the standard defines the conversion from a signed int (like -1) to /any/ unsigned integer type. On a two-complement machine that degenerates to a no-op if the length is unchanged, but on a sign-and-magnitude machine the compiler will generate the appropriate code. > > not the std::size_t). > > size_t is not explicit std::size_t in the original post. size_t also > doesn't mean unsigned int. But it does mean unsigned something. [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] |