Make generator more msvc-friendly

[JesseTG]

This PR makes generator able to compile in Visual C++ 2015 by doing the following:

• Changing all inheritance from private to public I removed that commit
• Adding some msvc-specific information to CMakeLists.txt

[ilmola]

Why did you change private inheritance to public? Private inheritance is an implementation detail. Public inheritance is part of the interface.

[JesseTG]

MSVC was choking on the private inheritance for some reason (saying it couldn't access the members specified with the using declarations), while gcc and clang took it just fine. I'm not sure which is the correct behavior; I think this might be up to different interpretations of the C++ standard. What do you want me to do?

[ilmola]

Private inheritance could be changed to private composition. Here is a branch created by search replacing inheritance to composition:

https://github.com/ilmola/generator/tree/inheritance_to_composition

Does it compile with MSVC?

[JesseTG]

I'll get back to you on that tonight (for you, tomorrow). It's probably the better thing to do anyway, because now there's not as much indirection or other virtual fuckery.

[JesseTG]

@ilmola Yes it does, actually. Haven't tried it with gcc or clang, though. After I do, assuming it checks out do you want me to cherry-pick the relevant commits onto this PR?

Also, can I see the script you wrote to process the code base? I'm curious.

[ilmola]

I am not familiar with git best practices, but it seems that all you need to do, is to merge your chances to the cmake files to the version generated by the script, and then you can submit a version that compiles with MSVC and does not introduce a massive API change.

The script was just this quick and dirty php search and replace. It converts all but 2 lines that need to fixed by hand.

function convertFile($file, $path, $pathNew) {
    $content = file_get_contents($path."include/generator/".$file);

    $srcFile = str_replace("hpp", "cpp", $file);
    $srcPath = $path."src/".$srcFile;
    $srcContent = file_exists($srcPath) ? file_get_contents($srcPath) : "";

    $matches = [];
    preg_match_all("/class\s+(.+)\s+(:\s+private\s([^{]+)){/", $content, $matches);

    for ($i = 0; $i < count($matches[0]); $i++) {
        $parent = trim($matches[3][$i]);
        $shortParent = explode("<", $parent)[0];
        $member = lcfirst($shortParent)."_";

        $newClass = "class ".$matches[1][$i]."\n{\nprivate:\n\n\tusing Impl = ".$parent.";\n\tImpl ".$member.";\n"; 
        $content = str_replace($matches[0][$i], $newClass, $content);

        foreach (["triangles", "edges", "vertices"] as $t) {
            $T = ucfirst($t);
            $newFn = "using ".$T." = typename Impl::".$T.";\n\n\t".$T." ".$t."() const noexcept { return ".$member.".".$t."(); }";

            $content = str_replace("using ".$shortParent."::".$t.";", $newFn, $content);
            $content = str_replace("using ".$parent."::".$t.";", $newFn, $content);
        }

        foreach (["mesh", "path", "shape"] as $name) {
            $cast = "static_cast<const ".$parent."&>(".$name.")";
            $content = str_replace($cast, $name.".".$member, $content);
        }

        $content = str_replace($parent."{", $member."{", $content);
        $srcContent = str_replace($parent."{", $member."{", $srcContent);
        $srcContent = str_replace($shortParent."{", $member."{", $srcContent);  
    }

    file_put_contents($pathNew."include/generator/".$file, $content);

    if ($srcContent != "") file_put_contents($pathNew."src/".$srcFile, $srcContent);
}

function convertDir($path, $pathNew) {
    foreach (glob($path."include/generator/*") as $filePath)
        convertFile(basename($filePath), $path, $pathNew);
}

convertDir($argv[1], $argv[2]);

[JesseTG]

I'll just cherry-pick the commit, then. However, I still want to ensure that this won't give gcc or clang a hard time. I also have a linker problem and I need to figure out if it's related to this or not (probably not but I don't want to merge bad code into master).

I don't want to create a whole new branch or PR because not all the changes in this one are about the code itself (the stuff in CMakeLists.txt, for instance, is vitally important).

[JesseTG]

That branch is good to go with gcc. Can't try it with clang right now because of an unrelated problem. I'll cherry-pick the relevant commits onto this branch.