boostorg / multi_array Goto Github PK

Using multi_array, MSVC (VS 2019, VS 2022) emits the following warnings (using Boost release 1.82.0, but current code in the repo includes the same global variable and argument name):

`C:\lo\build\workdir\UnpackedTarball\boost\boost/multi_array/multi_array_ref.hpp(615): warning C4459: declaration of 'extents' hides global declaration

C:\lo\build\workdir\UnpackedTarball\boost\boost/multi_array/base.hpp(69): note: see declaration of 'boost::anonymous-namespace'::extents'

This prevents warning-free builds, or requires pollution of the code with pragmas. Would it be possible to rename the constructor argument name?

When I use boost::multi_array with boost::fortran_storage_order(), and I want to .resize by only the latest (rightmost) dimension, the current implementation is a nested N function call (std::copy), one by dimension.
The same functionality can be reached with one std::copy on the storage.

This minimal example shows the problem:
https://godbolt.org/z/GWbxcM14h

Hello, this is a copy of my email I sent to the boost-users mailing list, as it might improve visibility.

I'd love to be able to use multi_array with classes that do not have a default constructor. At the moment, this is impossible, as both the constructor and resize functions only take extents as arguments (and possibly some other options), but no value_type, as for example std::vector does. Thus, trying to store objects without a default constructor in a multi_array results in hard compiler errors, with no workaround that I know of.

This happens to me fairly often, as I use multi_array to organize sets of complex objects. If there is a technical reason why this is not possible, I would also be happy to learn it.

... or at least make its members protected to allow a subclass to do so.

swap would be substantially the same as the current resize method.

Background: I use a subclass of multi_array in order to resize during assignment to get around the most dagnerous multi_array pitfall. This works, but I'd also like to fix another longstanding inefficiency ( #24 ) while I'm at it. Unfortunately the private data members make it impossible to fix in my subclass (barring a type-punning hack).

The following program:

#include "boost/multi_array.hpp"
using Emax = boost::multi_array<double, 7>; 

int main() {
    Emax emax(boost::extents[40][30][50][5][5][5][5]);
    return 0;
}

when run under valgrind gives the following warnings:

==20786== Memcheck, a memory error detector
==20786== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al.
==20786== Using Valgrind-3.15.0 and LibVEX; rerun with -h for copyright info
==20786== Command: ./test_arr
==20786==
==20786== Warning: set address range perms: large range [0x5171040, 0x16f8b340) (undefined)
==20786== Warning: set address range perms: large range [0x5171028, 0x16f8b358) (noaccess)
==20786==
==20786== HEAP SUMMARY:
==20786==     in use at exit: 0 bytes in 0 blocks
==20786==   total heap usage: 2 allocs, 2 frees, 300,072,704 bytes allocated
==20786==
==20786== All heap blocks were freed -- no leaks are possible
==20786==
==20786== For lists of detected and suppressed errors, rerun with: -s
==20786== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)

versions are:

gcc version 8.3.1 20190223 (Red Hat 8.3.1-2) (GCC)
valgrind-3.15.0

The bug described in https://svn.boost.org/trac10/ticket/6554 seems to return. I tried the example from this bug in godbolt: https://godbolt.org/z/4vMf8oo4q and it fails to compile. The code I used is:

#include <boost/multi_array.hpp>

struct data
{
	int	value;
};
typedef boost::multi_array<data, 2> array_type;

int main() {
    array_type	a(boost::extents[4][5]);

    // ERROR: Cannot compile this line
    a.begin()->begin()->value = 7;

    // Compiles successfully
    (*a.begin()->begin()).value = 5;
    return 0;
}

All versions of boost on godbolt (1.64 to 1.79) seem to be affected.

I am trying to use multi_arrays in a CUDA GPU project and NVCC is unable to compile it.
NVCC version is 12.3 (error also occurs with other versions)
Boost version is 1.84.0 (error also occurs with other versions)

I used the following command to compile:
nvcc -o C:/_repro/output.exe C:/_repro/main.cu -IC:/boost_1_84_0

I am using the following code (main.cu):

#include <boost/multi_array.hpp>

int main()
{
	boost::multi_array<int, 3> my_arr;
	return 0;
}

I get the following errors:

C:/boost_1_84_0\boost/multi_array/base.hpp(408): error C2988: unrecognizable template declaration/definition
C:/boost_1_84_0\boost/multi_array/base.hpp(408): error C2143: syntax error: missing ')' before '<'
C:/boost_1_84_0\boost/multi_array/base.hpp(408): error C2143: syntax error: missing ';' before '<'
C:/boost_1_84_0\boost/multi_array/base.hpp(408): error C2238: unexpected token(s) preceding ';'
C:/boost_1_84_0\boost/multi_array/base.hpp(408): error C2059: syntax error: '<'
C:/boost_1_84_0\boost/multi_array/base.hpp(414): error C2059: syntax error: ')'
C:/boost_1_84_0\boost/multi_array/base.hpp(494): error C2143: syntax error: missing ';' before '}'
C:/boost_1_84_0\boost/multi_array/base.hpp(494): error C2238: unexpected token(s) preceding ';'

Several more errors occur when I attempt to use further multi_array functionality such as extents and indexing, this is just a minimal example. Additionally, the code compiles just fine when you change the file extention to .cpp, but I assume this is just NVCC falling back on the default cl.exe compiler. I specifically need multi_arrays to compile within a .cu file along with CUDA device code.

Any help is much appreciated, thank you.

The following macros are missing a BOOST_ prefix, which is against Boost library guidelines:

./libs/multi_array/test/generative_tests.hpp:#ifndef GENERATIVE_TESTS_RG072001_HPP
./libs/multi_array/test/generative_tests.hpp:#define GENERATIVE_TESTS_RG072001_HPP
./libs/multi_array/test/generative_tests.hpp:#endif // GENERATIVE_TESTS_RG072001_HPP
./boost/multi_array/base.hpp:#ifndef BASE_RG071801_HPP
./boost/multi_array/base.hpp:#define BASE_RG071801_HPP
./boost/multi_array/base.hpp:#endif // BASE_RG071801_HPP
./boost/multi_array/collection_concept.hpp:#ifndef COLLECTION_CONCEPT_RG103101_HPP
./boost/multi_array/collection_concept.hpp:#define COLLECTION_CONCEPT_RG103101_HPP
./boost/multi_array/collection_concept.hpp:#endif // COLLECTION_CONCEPT_RG103101_HPP
./boost/multi_array/range_list.hpp:#ifndef RANGE_LIST_RG072501_HPP
./boost/multi_array/range_list.hpp:#define RANGE_LIST_RG072501_HPP
./boost/multi_array/range_list.hpp:#endif // RANGE_LIST_RG072501_HPP
./boost/multi_array/iterator.hpp:#ifndef ITERATOR_RG071801_HPP
./boost/multi_array/iterator.hpp:#define ITERATOR_RG071801_HPP
./boost/multi_array/iterator.hpp:#endif // ITERATOR_RG071801_HPP
./boost/multi_array/copy_array.hpp:#ifndef COPY_ARRAY_RG092101_HPP
./boost/multi_array/copy_array.hpp:#define COPY_ARRAY_RG092101_HPP
./boost/multi_array/copy_array.hpp:#endif // COPY_ARRAY_RG092101_HPP
./boost/multi_array/subarray.hpp:#ifndef SUBARRAY_RG071801_HPP
./boost/multi_array/subarray.hpp:#define SUBARRAY_RG071801_HPP
./boost/multi_array/subarray.hpp:#endif // SUBARRAY_RG071801_HPP

Hello!

I think it would be great if multi_array had move semantics, at the moment unfortunately it is inefficient to return multi_arrays by value and the workaround is not great:

https://stackoverflow.com/questions/42698451/how-to-stdmove-a-boostmulti-array#comment72520077_42698451

I think having move construction / assignment would be great

While not a C++ N-D array, NumPy provides an N-D array for Python. Might be nice to include this in the related work section of the docs.