22 November 2024
The ISO C++ Standard Committee held its November 2024 meeting in Wrocław, Poland. It’s an important meeting for C++26 as it’s the last meeting for the Evolution Working Group to approve C++26 features needing library response. This means that proposals having both core language and standard library impact need to get either Evolution Working Group (EWG) or Library Evolution Working Group (LEWG) approved by the end of the Wrocław meeting to be included in C++26. For most of the time, I was sitting in the Evolution Working Group as I am mostly interested in language features that may impact compiler designs. Here are some thoughts on some of the major discussions I have participated in.
P2996 is a paper that progressed very quickly through the previous three meetings. It was first seen in Kona in November 2023. During the St. Louis meeting, the paper received strong consensus to be forwarded to LWG and CWG for review. I am personally very excited about this feature. It’s a transformative feature for C++26 and will empower libraries to significantly improve ergonomics for daily programming. (e.g. think about deriving serialization, logging entire structs, and command line argument parsing.)
P3451: The biggest open
question as a challenge to the status quo of P2996 is whether reflection APIs like
members_of should respect accessibility. This was the main
argument in EWG on Wednesday.
In the end, no consensus was reached whether we should make inconvenient spellings for accessing private members. The general feeling of the room is that we need to retain some form of access to private and protected members.
Pattern Matching is a feature seen in many functional languages that
provides very powerful and ergonomic control flow. Currently, it’s
extremely inconvenient to pattern match standard library constructs like
std::variant<Ts...> . As of C++23, people have to do
library tricks to emulate pattern matching that is nowhere nearly as
powerful and elegant. I believe there’s good reasons to be excited about
P2688: Pattern Matching:
match Expression. Michael Park presented the paper in EWG
on Thursday morning where the committee made progress on the
discussions. There is consensus that we’d like to encourage more work on
Michael Park’s P2688 for C++26 inclusion.
There’s also an alternative proposal P2392 by Herb Sutter to adopt ‘is’ and ‘as’ for pattern matching. My main concern on ‘is’ and ‘as’ is that it’s introducing plenty of cascading rules, potentially making it difficult to distinguish dynamic or static casts, which is particularly problematic under dependent contexts. The working group reached no consensus on whether we want such a feature.
P2900R11: Contracts for C++ is a much requested language feature for safety critical application. The EWG reserved a day and a half to discuss contracts and make a decision on whether Contracts should be included in C++26. The notion of safety in contracts is different from UB safety. In short, the current status of P2900 allows users to define dynamic assertions as preconditions and postconditions on function parameters as well as using contract assertion statements within a function body. A brief example can be found on section 2.2 of the paper. Although the group voted to forward this paper for library and wording review, the consensus for the current design is weak. There are major objections around multiple aspects of the design.
An alternative proposal for what’s called a conveyor function would like to limit what expression you can put into the predicate and hence address UB and the motivation of constification. This, on the other hand, makes contracts much less powerful. P2900 and its competing papers Since the consensus is weak we will likely see objections later on in the process.
Herb Sutter presented P3081: Core safety Profiles: Specification, adoptability, and impact to EWG on Friday Morning. Profiles has been a popular topic among folks who cite the February White House paper on the recommendation of Programming Languages. There was also a discussion referred to a post by Chandler Carruth about how Google enabled bounds checking with minimal overhead. While I am in support for safety and security, I voiced my concerns about how diverging perception of safety can be harmful. If we are on our way to convince the US government about C++ being safe (or safer now,) we might be sending the wrong message to the developers who may start to treat memory safety less seriously. Since we have only seen a published version of this paper a bit over a month ago, time for inclusion to C++26 is limited. The committee reached consensus to make a policy exception for Profiles to be reviewed again in Hagenberg in February for potential inclusion to C++26, despite its simultaneous language and library impact.
On Thursday afternoon, JF Bastien presented P3477: There are exactly 8 bits in a byte. This paper had seen hot internet engagement before the meeting and made us aware of many exotic architectures. In the end, we forwarded this paper to CWG for inclusion in C++26 with consensus.
Overall, it’s an overwhelmingly positive experience to meet with the other 144 committee members who attended in person. It’s a good reminder to all of us that different interests and strong opinions occur based on the vastly different backgrounds of committee members. We may not agree with one another but with our enthusiasm, the collective goal is to make C++ experience better over time for millions. Personally, it’s an opportunity for growth in my comfort zone and practicing technical trade off discussions at high stakes.