[dynamic] Initialization issue

[sdebionne]

Actual behavior

boost::gil::any_image<
  boost::mp11::mp_list<
    boost::gil::gray8_image_t
  >
> img = boost::gil::gray8_image_t();

Fails to compile with

error: conversion from 'boost::gil::gray8_image_t {aka boost::gil::image<boost::gil::pixel<unsigned char, boost::gil::layout<boost::mp11::mp_list<boost::gil::gray_color_t> > >, false, std::allocator<unsigned char> >}' to non-scalar type 'boost::gil::any_image<boost::mp11::mp_list<boost::gil::image<boost::gil::pixel<unsigned char, boost::gil::layout<boost::mp11::mp_list<boost::gil::gray_color_t>, boost::mp11::mp_list<std::integral_constant<int, 0> > > >, false, std::allocator<unsigned char> > > >' requested

while

boost::gil::any_image<
  boost::mp11::mp_list<
    boost::gil::gray8_image_t
  >
> img(boost::gil::gray8_image_t());

compiles fine.

Expected behavior

It would be nice if both initializations work the same.

C++ Minimal Working Example

#include <boost/gil.hpp>
#include <boost/gil/extension/dynamic_image/any_image.hpp>
namespace gil = boost::gil;
int main
{
  any_image<boost::mp11::mp_list<gray8_image_t>> img = gray8_image_t();
}

Environment

• Compiler version: GCC 7.3
• Build settings: -std=gnu++1z
• Version (Git ref or <boost/version.hpp>): 1.75

[mloskot]

@sdebionne yes, it is desired. I have only applied very minimal C++11 modernisation to the dynamic_image extension, it needs more.

By the way, we are likely removing some of apparently dead code from the dynamic_image extension (see #281) and we will also be looking into adding move semantics to image classes (see #438).

[sdebionne]

@mloskot I can have a shot at #438 (I think this issue is related) in nobody is working on it.

[mloskot]

@sdebionne I discovered that we might have lost the boost::gil::any_image<boost::mp11::mp_list<...>> completely, what is somewhat related to this issue.

There have been number of changes to any_image and apply_operation recently and earlier, and I might have got lost.
Could you update me on what is currently supported form of specialising any_image?

Here is what I'm testing using Boost 1.76, https://godbolt.org/z/4Ta31Tf3P, and below:

#include <boost/gil.hpp>
#include <boost/gil/extension/dynamic_image/any_image.hpp>
#include <boost/mp11.hpp>
namespace gil = boost::gil;

int main()
{
    // OK
    {
        gil::any_image
        <
            gil::gray8_image_t,
            gil::rgb8_image_t
        > image;
        image.dimensions();
    }
    // not OK
    {
        gil::any_image
        <
            boost::mp11::mp_list
            <
                gil::gray8_image_t,
                gil::rgb8_image_t
            >
        > image;
        image.dimensions();
    }
}

[sdebionne]

The first form is the only one supported for any_image and any_image_view. It's indeed a breaking change that was maybe not advertised enough. But I believe the doc is up-to-date.
I'm a bit surprised that I reported this issue with boost::mp11::mp_list<, that probably means it predates the final variant2 refactoring. The use of inheriting constructors b8f48bc should fix #453.

[mloskot]

@sdebionne Thank for the update. Yes, it does predate the variant2.

I will have to catch up the IO tests with those changes.
I missed that because of the still convoluted way of running the IO tests, what tests are run by our CI, what #define-s are set, etc. so it needs clearing to try avoid any conditional running of tests.

[sdebionne]

I thought I had updated all the tests, but since I am usually not running the tests locally with IO tests, I might have left some behind... sorry about that. Indeed, it would be great to have the CI catch these oversights.

[mloskot]

@sdebionne No problem, not your fault, but the tests configuration - it should be no brainer w/o any manual tweaks required.