Re: [boost] Config and "unused variables"

From: Beman Dawes (bdawes_at_[hidden])
Date: 2001-09-26 14:21:46

At 11:51 AM 9/26/2001, Darin Adler wrote:

 >on 9/26/01 4:13 AM, Beman Dawes at bdawes_at_[hidden] wrote:
 >
 >> But isn't the usual practice in that case to wrap the temporary and 
store
 >> to it in an #ifndef NDEBUG?
 >
 >A good way to see where this comes up is to look at the examples already 
in
 >Boost. In most cases, we just don't declare the variable or don't name it 

 >in
 >the parameter list, but there are a number of files that use one of the
 >techniques we're discussing.
 >
 >The ignore_unused_variable_warning function template is used in 8 
different
 >source files in Boost: boost/concept_check.hpp,
 >boost/graph/graph_concepts.hpp, boost/iterator_adaptors.hpp,
 >boost/property_map.hpp, libs/concept_check/class_concept_check_test.cpp,
 >libs/concept_check/stl_concept_covering.cpp,
 >libs/utility/call_traits_test.cpp (called unused_variable in this file),
 >and
 >libs/tuple/test/tuple_test_bench.cpp (called dummy in this file).
 >
 >The cast to (void) technique is used in 32 different source files in 
Boost:
 >boost/python/detail/extension_class.hpp,
 >boost/regex/detail/regex_format.hpp, boost/regex/detail/regex_match.hpp,
 >boost/type_traits/type_traits_test.hpp,
 >libs/concept_check/class_concept_fail_expected.cpp,
 >libs/python/src/objects.cpp, libs/python/src/types.cpp,
 >libs/python/test/comprehensive.cpp, libs/random/random_speed.cpp,
 >libs/random/random_test.cpp, libs/regex/src/primary_transform.hpp,
 >libs/regex/src/wide_posix_api.cpp, 
libs/utility/counting_iterator_test.cpp,
 >libs/utility/half_open_range_test.cpp, and in many of the tests in the
 >libs/config/test directory.

Interesting.  Note that those uses involve a number of different 
developers, too, not just one person repeating a personal favorite 
technique.

I wouldn't have guessed anywhere near that much use for either 
technique.  Makes a strong argument we should have something in Boost.

 >I find the function template more readable, more elegant, and perhaps
 >slightly more portable, since it relies on the "don't complain about
 >parameters without names" behavior rather than the "mentioning in an
 >expression that is unused is using a variable" behavior.
 >
 >About 1/2 of the uses of the (void) technique also have a comment to
 >explain what the void cast means.
 >
 >There can be overhead caused by the function template technique, though,
 >when the compiler is not doing function inlining. This is not an issue 
for
 >me, but it might be an issue for others.

Pending more discussion, I lean toward the template too.

--Beman