| c++11 containers
|
| sorted_vector |
use when doing lots of unsorted insertions and maintaining constant sort would be expensive; vector is good for a big pile of things that only occasionally needs a sorted lookup
|
| map |
sorted binary search tree; always sorted by key; you can walk through in sorted order (choose unordered if not needed!)
|
| multimap |
same as map but allows dupe keys (not as common)
|
| unordered_map |
hashmap; always sorted by key; additional bucket required for hash collisions; no defined order when walking through
|
| unordered_multimap |
same as map but allows dupe keys; dupes are obviously in the same bucket, and you can walk just the dupes if needed
|
set
multiset
unordered_set
unordered_multiset |
sets are just like maps, except the key is embedded in the object, nice for encapsulation.
Items must be const (!) since they are the key - sounds bad, but this is mitigated by the mutable keyword.
You can use mutable on the variables that are not part of the key to remove the const.
This changes the constness of the object from binary (completely const) to logical (constness is defined by the developer).
So... set is a good way to achieve both encapsulation and logical const - make const work for you, not against! :-)
|
| set (etc.) of pointers |
sets of pointers are the pinnacle of object stores
The entire object can be dereferenced and accessed then without const issues.
A pointer functor can be provided that does a sort by dereferencing the pointer to the object.
Two requirements: you must make sure yourself that you do not change the key values - you can mark them const, provided in constructor;
you must create sort/equal/hash functors that dereference the pointers to use object contents
(the default will be by pointer address).
The arguably biggest advantage, as a result, is that you can create multiple sets
to reference the same group of objects with different sort funtors to create multiple indices.
You just have to manage the keys carefully, so that they don't change (which would invalidate the sorting).
The primary container can manage object allocation; using a heap-based unique_ptr allocation
map vs key redux
use a key in the set, derive a class from it with the contents
+ small key
+ encapsulation
- requires mutable to solve the const problem
use a key in the set, key includes a mutable object
+ encapsulation
- weird bc everything uses a const object but we have const functions like save() that change the mutable subobject
use a map
+ small key
- no encapsulation, have to deal with a pair instead of an object
can we just put a ref to key in the value? sure why not - err, bc we don't have access to it
+ solves const problem bc value is totally mutable by design
+ we can have multiple keys - and the value can have multiple refs to them
+ simpler equal and hash functions
map:
create an object with internal key(s)
create map index(es) with duplicate key values outside the object - dupe data is the downside
use set(s) with one static key for find():
create an object with internal key(s)
create set index(es) with specific hash/equals functor(s)
when finding, use one static key object (even across indexes!) so there isn't a big construction each time; just set the necessary key values
that proves difficult when dealing with member vars that are references
but to solve it, just set up a structure of dummy static key objects that use each other; then provide a function to setKey(Object& keyref) { keyref_ = keyref; }
nope, can't reassign refs
the solution: use pointers not references
yes that's right
just do it
apparently there was a reason i was anti-reference for all those years
two reasons to use pointers:
dynamically allocated
reassignment required
there ya go. simple. get it done.
when accessing find results from the set, use a const_cast on the object!
WARNING: a separate base class with the key sounds good... but fails when you have more than one index on the object. just use a static key object for them all!
|
| c++11 example for large groups of objects with frequent crud AND search
|
Best solution is an unordered set of pointers:
typedef boost::unordered_set<MajorObject*> MajorObjects;
|
| c++11 example for large groups of objects with infrequent crud and frequent search
|
Best solution is a vector of pointers sorted on demand (sorted_vector):
TODO
|
| c++11 example to associate two complex objects (one the map key, one the map value)
|
Use unordered_map with a custom object as key. You must add hash and equals functions. Boost makes it easy:
static bool operator==(MyKeyObject const& m1, MyKeyObject const& m2)
{
return
m1.id_0 == m2.id_0
&& m1.id_1 == m2.id_1;
}
static std::size_t hash_value(MyKeyObject const& mko)
{
std::size_t seed = 0;
boost::hash_combine(seed, mko.id_0);
boost::hash_combine(seed, mko.id_1);
return seed;
}
typedef boost::unordered_map<MyKeyObject, MyValueObject*> MyMap;
Note that you can extend this to use a pointer to a key object, whoop.
|
| c++11 example for multiple unordered_set indexes into one group of objects
|
Objects will be dynamically created. One set should include them all and be responsible for memory allocation cleanup:
TODO
|
| c++11 example for set with specific sorting
|
Use set with a specific sort functor. You can create as many of these indexes as you want!
struct customers_set_sort_functor
{
bool operator()(const MyObject* l, const MyObject* r) const
{
// the id is the key
return l->id_ < r->id_;
}
};
typedef set<MyObject*,myobject_sort_by_id_functor> MyObjectsById;
|
| c++11 loop through vector to erase some items
|
| Note that other containers' iterators may not be invalidated so you can just erase() as needed...
For vectors, you have to play with iterators to get it right - watch for proper ++ pre/postfix!
for (it = numbers.begin(); it != numbers.end(); ) // NOTE we increment below, only if we don't erase
{
if (*it.no_good())
{
numbers.erase(it++); // NOTE that we ERASE THEN INCREMENT here.
}
else
{
++it;
}
}
I thought I had always looped backwards to do this, I *think* that's ok too, but I don't see it used in my code, I think I'll avoid. :-)
|
| c++11 range based for loop, jacked with boost index if needed
|
No iterator usage at all. Nice at times, not enough at others. Make SURE to always use a reference or you will be working on a COPY. Make it const if you aren't changing the object.
for (auto& mc : my_container)
mc.setString("default");
for (const auto& cmc : my_container)
cout << cmc.asString();
boost index can give you the index if you need it, sweet:
#include <boost/range/adaptor/indexed.hpp>
...
for (const auto &element: boost::adaptors::index(mah_container))
cout << element.value() << element.index();
|
| c++11 for loop using lambda
|
This C++11 for loop is clean and elegant and a perfect way to check if your compiler is ready for c++11:
vector<int> v;
for_each( v.begin(), v.end(), [] (int val)
{
cout << val;
} );
This is using a lambda function, we should switch from iterators and functors to those - but not quite yet, since we're writing cross-platform code. Do not touch this until we can be sure that all platforms provide compatible C++11 handling.
|
| c++11 integer types
|
| I really like the "fast" C++11 types, that give best performance for a guaranteed minimum bit width.
Use them when you know a variable will not exceed the maximum value of that bit width, but does not have to be a precise bit width in memory or elsewhere.
Pick specific-width fields whenever data is shared with other processes and components and you want a guarantee of its bit width.
And when using pointer size and array indices you should use types defined for those specific situations.
FAST types:
int_fast8_t
int_fast16_t fastest signed integer type with width of
int_fast32_t at least 8, 16, 32 and 64 bits respectively
int_fast64_t
uint_fast8_t
uint_fast16_t fastest unsigned integer type with width of
uint_fast32_t at least 8, 16, 32 and 64 bits respectively
uint_fast64_t
SMALL types:
int_least8_t
int_least16_t smallest signed integer type with width of
int_least32_t at least 8, 16, 32 and 64 bits respectively
int_least64_t
uint_least8_t
uint_least16_t smallest unsigned integer type with width of
uint_least32_t at least 8, 16, 32 and 64 bits respectively
uint_least64_t
EXACT types:
int8_t signed integer type with width of
int16_t exactly 8, 16, 32 and 64 bits respectively
int32_t with no padding bits and using 2's complement for negative values
int64_t (provided only if the implementation directly supports the type)
uint8_t unsigned integer type with width of
uint16_t exactly 8, 16, 32 and 64 bits respectively
uint32_t (provided only if the implementation directly supports the type)
uint64_t
SPECIFIC-USE types:
intptr_t integer type capable of holding a pointer
uintptr_t unsigned integer type capable of holding a pointer
size_t unsigned integer type capable of holding an array index (same size as uintptr_t)
|
| C++11 scoped enumeration
|
C++11 has scoped enumeration, which lets you specify the SPECIFIC VARIABLE TYPE for the enum. Perfect, let's use uint_fast32_t.
enum class STRING_PREF_INDEX int_fast32_t: { ... };
Unfortunately, gcc gives me a scary warning, and stuff fails. For some reason, it does not know about the provided type, although it is definitely defined. Revisit this later if you have time.
warning: elaborated-type-specifier for a scoped enum must not use the ‘class’ keyword
Old skool is still cool:
typedef enum
{
// assert( SP_COUNT == 2 );
SP_FIRST = 0 ,
SP_SOME_PREF = SP_FIRST ,
SP_ANOTHA ,
SP_COUNT
} STRING_PREF_INDEX;
|
| c++ in-memory storage of "major" objects
|
OBSERVATION ONE
Consider An Important Qt Design: QObjects cannot normally be copied
their copy constructors and assignment operators are private
why? A Qt Object...
might have a unique QObject::objectName(). If we copy a Qt Object, what name should we give the copy?
has a location in an object hierarchy. If we copy a Qt Object, where should the copy be located?
can be connected to other Qt Objects to emit signals to them or to receive signals emitted by them. If we copy a Qt Object, how should we transfer these connections to the copy?
can have new properties added to it at runtime that are not declared in the C++ class. If we copy a Qt Object, should the copy include the properties that were added to the original?
in other words, a QObject is a pretty serious object that has the ability to be tied to other objects and resources in ways that make copying dangerous
isn't this true of all serious objects? pretty much
OBSERVATION TWO
if you have a vector of objects, you often want to track them individually outside the vector
if you use a vector of pointers, you can move the object around much more cheaply, and not worry about costly large vector reallocations
a vector of objects (not pointers) only makes sense if the number of objects is initially known and does not change over time
OBSERVATION THREE
STL vectors can store your pointers, iterate thru them, etc.
for a vector of any substantial size, you want to keep objects sorted so you can find them quickly
that's what my sorted_vector class is for; it simply bolts vector together with sort calls and a b_sorted status
following STL practices, to get sorting, you have to provide operator< for whatever is in your vector
BUT... you are not allowed to do operator<(const MyObjectPtr* right) because it would require a reference to a pointer which is not allowed
BUT... you can provide a FUNCTOR to do the job, then provide it when sorting/searching
a functor is basically a structure with a bool operator()(const MyObjectPtr* left, const MyObjectPtr* right)
OBSERVATION FOUR
unordered_set works even better when combining frequent CRUD with frequent lookups
SUMMARY
Dealing with tons of objects is par for the course in any significant app.
Finding a needle in the haystack of those objects is also standard fare.
Having multiple indices into those objects is also essential.
Using unordered_set with object pointers and is very powerful.
|
| c++ stl reverse iterator skeleton
|
From SGI...
reverse_iterator rfirst(V.end());
reverse_iterator rlast(V.begin());
while (rfirst != rlast)
{
cout << *rfirst << endl;
...
rfirst++;
}
|
| c++ stl reading a binary file into a string
|
std::ifstream in("my.zip",std::ios::binary);
if (!in)
{
std::cout << "problem with file open" << std::endl;
return 0;
}
in.seekg(0,std::ios::end);
unsigned long length = in.tellg();
in.seekg(0,std::ios::beg);
string str(length,0);
std::copy(
std::istreambuf_iterator< char >(in) ,
std::istreambuf_iterator< char >() ,
str.begin()
);
For more, see c++ stl reading a binary file
|
| C/C++ best-in-class tool selection
|
I need to have easy setup of debug-level tool support for portable C++11 code. And I need to decide and stick to it to be efficient.
- Compiler selection
- linux and mac: gcc
- windows: Visual Studio
- IDE selection
- linux and mac: eclipse
- windows: eclipse OR Visual Studio OR Qt Creator
- Debugger selection
- linux and mac: eclipse using gdb OR ddd
- windows: eclipse OR Visual Studio OR Qt Creator
|
| c/c++ gdb debugging
|
(gdb) help break
Set breakpoint at specified line or function.
Argument may be line number, function name, or "*" and an address.
If line number is specified, break at start of code for that line.
If function is specified, break at start of code for that function.
If an address is specified, break at that exact address.
With no arg, uses current execution address of selected stack frame.
This is useful for breaking on return to a stack frame.
Multiple breakpoints at one place are permitted, and useful if conditional.
Do "help breakpoints" for info on other commands dealing with breakpoints.
|
| ddd gives you a front end. I need to use it more, compare to other options
|
| C - Create a portable command line C project in Visual Studio
|
Visual Studio: File -> New -> project
Visual C++ -> Win32 -> Win32 Console Application
name: oms_with_emap
next -> click OFF precompiled header checkbox (even tho it didn't seem to respect it)
you'll get a _tmain(..., TCHAR*...)
change it to main(..., char*...)
change the project to explicitly say "Not using precompiled header"
remove the f'in stdafx.h
recompile! should be clean
vs will recognize C files and compile accordingly
|
| c++ Create a portable C++ project in Visual Studio
|
|
If you don't have existing code, it's probably best to create a project_name.cpp file with your main() function.
int main( int argc, char * argv[] )
{
return 0;
}
Then in Visual Studio...
File->New->Project from existing code
C++
(then use mostly defaults on this page, once you provide file location and project name)
Project file location: <base>\project_name
Project name: project_name
[x] Add files from these folders
Add subs
[x] <base>\project_name
NEXT
Use Visual Studio
Console application project
No ATL, MFC, CLR
NEXT
NEXT
FINISH
Then add Reusable and boost include and lib paths:
Example if you built boost according to notes below:
Project->Properties->All Configurations->Configuration Properties->VC++ Directories->Include Directories->
$(VC_IncludePath);$(WindowsSDK_IncludePath);..\..\..\Reusable\c++
INCLUDE: C:\Software Development\boost_1_53_0
LIB: C:\Software Development\boost_1_53_0\stage\lib
|
| boost release and debug build for linux
|
# download latest boost, eg: boost_1_59_0
m@wallee:~/development$ 7z x boost_1_59_0.7z
cd boost_1_##_0
build_boost_release_and_debug.sh
# then patch .bashrc as instructed
# eclipse and server/nix/bootstrap.sh are customized to match
To upgrade a project:
cd nix
make distclean # removes nasty .deps folders that link to old boost if you let them
make clean # removes .o files etc
cd ../build-Release && make distclean && make clean
cd ../build-Debug && make distclean && make clean
cd ..
./bootstrap force release
./bootstrap force debug
|
| boost release and debug build for Windows
|
| Open a VS2015 x64 Native Tools Command Prompt.
EITHER: for new installs, you have to run bootstrap.bat first, it will build b2;
OR: for reruns, remove boost dirs: [bin.v2, stage].
Then build 64-bit:
cd "....\boost_1_59_0"
b2 toolset=msvc variant=release,debug link=static address-model=64
rem trying to avoid excessive options, assuming I don't need these: threading=multi
(old stuff)
--toolset=msvc-14.0 address-model=64 --build-type=complete --stagedir=windows_lib\x64 stage
Now open VS2013 x86 Native Tools Command Prompt and build 32-bit:
cd "C:\Michael's Data\development\sixth_column\boost_1_55_0"
bjam --toolset=msvc-12.0 address-model=32 --build-type=complete --stagedir=windows_lib\x86 stage
|
| gcc install multiple versions in ubuntu (4 and 5 in wily, eg)
|
| My code will not compile with gcc 5, the version provided with Ubuntu wily.
It gives warnings like this:
/home/m/development/boost_1_59_0/boost/smart_ptr/shared_ptr.hpp:547:34: warning: ‘template<class> class std::auto_ptr’ is deprecated [-Wdeprecated-declarations]
and outright errors like this:
depbase=`echo AtServer.o | sed 's|[^/]*$|.deps/&|;s|\.o$||'`;\
g++ -DPACKAGE_NAME=\"at_server\" -DPACKAGE_TARNAME=\"at_server\" -DPACKAGE_VERSION=\"1.0\" -DPACKAGE_STRING=\"at_server\ 1.0\" -DPACKAGE_BUGREPORT=\"[email protected]\" -DPACKAGE_URL=\"\" -DPACKAGE=\"at_server\" -DVERSION=\"1.0\" -I. -I../../src -I/home/m/development/Reusable/c++ -I/home/m/development/Reusable/c++/sqlite -std=c++11 -I/home/m/development/boost_1_59_0 -ggdb3 -O0 -std=c++11 -MT AtServer.o -MD -MP -MF $depbase.Tpo -c -o AtServer.o ../../src/AtServer.cpp &&\
mv -f $depbase.Tpo $depbase.Po
In file included from /usr/include/c++/5/bits/stl_algo.h:60:0,
from /usr/include/c++/5/algorithm:62,
from /usr/include/c++/5/ext/slist:47,
from /home/m/development/boost_1_59_0/boost/algorithm/string/std/slist_traits.hpp:16,
from /home/m/development/boost_1_59_0/boost/algorithm/string/std_containers_traits.hpp:23,
from /home/m/development/boost_1_59_0/boost/algorithm/string.hpp:18,
from /home/m/development/Reusable/c++/utilities.hpp:4,
from ../../src/MemoryModel.hpp:11,
from ../../src/SqliteLocalModel.hpp:13,
from ../../src/AtServer.cpp:70:
/usr/include/c++/5/bits/algorithmfwd.h:573:13: error: initializer provided for function
noexcept(__and_<is_nothrow_move_constructible<_Tp>,
^
/usr/include/c++/5/bits/algorithmfwd.h:582:13: error: initializer provided for function
noexcept(noexcept(swap(*__a, *__b)))
You can set up the update-alternatives tool to switch out the symlinks:
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-4.9 60 --slave /usr/bin/g++ g++ /usr/bin/g++-4.9
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-5 20 --slave /usr/bin/g++ g++ /usr/bin/g++-5
BUT that seems BAD to me to switch out the compiler used by the SYSTEM. Instead, we should specify the compiler required by the PROJECT. This is supposed to do it, but it still uses /usr/include/c++/5, which seems wrong, and gives me errors:
CC=gcc-4.9
|
|