(I wrote this paper but forgot to publish it, sorry! I’m still gonna publish it as a bit of a retrospective on what I prepared…!) Got all loaded up decked out for the big paper bonanza; practice talks, draft papers with Double T’s, and more…
Sections marked with 🛡️ means I’ve either been voluntold or volunteered to pick the paper up and defend it. Others are the ones I wrote up. Some papers I originally volunteered for, but eventually bailed on staying with them (time constraints, etc.) so someone else will have to present them in my stead.
Heading for EWG(I)
The Committee has
factionsgroups, as I detailed in a previous blogpost. The Uruk’Hai EWG put a new splinter Study Group to help deal with the unprecedented paper volume, Orcs EWGI, or Evolution Working Group, Incubator. It is apparently pronounced “oogie”, as in “oogie boogie”. EWG is one of the hardest groups to get through, with many papers often dying in Evolution and then staying dead for a good while. As put by someone else: “I can get the hang of every room in the Standards Committee. I generally know what they all want… except for EWG.” To get things through here, your best game has to be on the table; and maybe a little luck, too. EWGI will hopefully be a tad easier…!
This paper should have been in C++17 but the original author lost motivation to continue the work. I picked it up at the behest of someone else. I’ve essentially become its champion and now I need to shepherd it through. The change is fairly simple, but I anticipate WAY more bikeshed than is necessary about the actual suffixes. Some individuals are hellbent on having
ptrdiff_t. That leaves the problem of what
z means (is it like POSIX’S
ssize_t? Which, by the way, is not part of the Standard right now) and what
ut means (
unsigned ptrdiff_t? What is that type? Who knows.) I proposed
z for consistency with the core language, but some people have a nice little hill about
t that they would like to die on.
As is usual, the simpler papers generate the most
awful bikeshed, so I anticipate to spend long minutes justifying why having 2 different identifiers mixed with the Standards consistent
u treatment is just not in scope for this paper: until someone can define in the standard what
ut and what
z mean in the world where we have 2 suffixes, those suffixes and definitions are inconsistent and useless to the Standard because it will just be one more Implementation Defined Handwaved Integral Type That Can Have This Signed And Size Or Totally More Maybe You Know Whatever Works, Friendo. Sometimes those are needed.
This is not one of those times.
Not really much to say here. Everybody, and their cat/dog/guinea pig/mouse/zebra/tiger, wants this. It was supposed to be seen in Rapperswil by EWG but it ended up losing out. Now it’s in EWGI’s hands.
Hopefully we get this into C++.
This paper is actually proposed by Alex Christensen, but I recently leaned he will not be able to attend the meeting to push this paper through. So I will be presenting and advocating for it. Originally targeted at LEWG by just changing the
const-ness of the
initializer_list parameter, this paper now needs a trip through Evolutionary groups to approve the change to the core language wording discussed in my previous write-up of it in paper reviews.
As the paper’s new champion, there is one fundamental question that I need answers to, particularly from a few people who seemed to champion this change when
initializer_list was first created:
Why is the backing storage
const and what is the key motivation for such a restriction on the compiler?
If I cannot receive a good answer to this question, then the change is essentially justified from the lack of good reasoning. We can also swat down the
movable_initializer_list debate, because it’s been crushed before precisely because it’d be a whole set of complicated rules on when to un-
const the backing array and provide this new type on top. If there wasn’t a good reason to
const the array (or the reason no longer Applies in This Day and Age™), then we can just fix
initializer_list without really, really poor patch jobs.
A thing to note is that in prior discussions everyone was talking about how to extend
initializer_list or add a new type, not to just fix the current definition. There are, as always, concerns about doing this but I think there are far too many people who are extremely annoyed about
initializer_list’s behaviors and do not have dependencies on its
const-ness to begin with. Fixing this makes
initializer_lists far more suitable with their desired usage.
I like this paper. I was not sure I was going to write it before quite a few people encouraged me to. I think this paper helps bring a lot of useful syntax into the language that people have wanted for quite some time. I also think it is incredibly important that we do not need to
std::invoke for everything, and that the language can reasonably handle some of this without excessive amounts of template spew.
This was not my paper; I borrowed it from someone to lighten their paper load. It’s a dirt-simple paper: let library authors provide a reason why they do not want you to discard this return value. You can even put something like
"this is a legacy work around and okay to (void)". Not really too bent out of shape about this going through or not, but it would be nice.
Headed for LEWG(I)
If you were around, I posted in my Rapperswil Trip Report about the
Humans Library Evolution Working Group’s acronym being LEWG and sounding like loog? Well, I said it’s one or two letters off from sounding like loogie.
And then this group appears.
The absolute madlads did it. They created a group whose acronym is actually said loogie! Feelings of prophetic joy aside, LEWGI is the
Half-Humans Library Evolution Working Group Incubator. Again, it’s meant to handle all sorts of proposals and try to filter out the good ones to either forward to LEWG or – with some approval from LEWG’s chair – send it right to the Elves LWG.
This paper paves the way for getting rid of people overloading unary address-of operator
& and violating a fundamental C++-ism: that every object has a (unique) address that you can get to. Of course, this does not help when you have a smart pointer that you want to
&my_ptr into a C API that takes
T**. So, this proposal steps in.
The one known case of someone overloading unary operator
CComPtr. The shiny new windows library WRL doesn’t make this mistake again and, as discussed in the paper, goes a route similar to what is designed in p1132. VMWare and a bunch of other Fortune 100 companies also have this in their codebase. In short: it’s a long-standing existing practice, gives people who overloaded a bad operator a way out, and solves a serious problem.
It has wording now, and I fully intend to get it to LWG by the end of the week.
I wrote about this paper in one of my other posts so there’s no reason to go over it a lot. But, I really believe this one needs to reach LEWG and LWG this cycle. It is critical we do not make our lives worse.
Headed for NWHR
These are papers headed for the Nowhere Study Group, SG0. They’re for the dark void of my imagination, and will probably take C++23 or C++26 to see the light of day.
- p1039: I got you, FAM - Flexible Array Members for C++
- p1193: Explicitly Specified Returns for (Implicit) Conversions
- p1129: The Holy War
- pXXXX: string_literal and basic_c_string_view (just an idea I’m toying with for now)
That’s all for now!
Catch you laaater, on the flip-side of San Diego. ♥