C++26: Standard Library Hardening Experiments
cppstories.com
cppstories.com
> The C++26 draft has not yet had its final ballot
This is co-authored by Bjarne, and so I'm sure it's not trivially false, but maybe I am just missing some detail. I know Bjarne was threatening to cast a veto of C++26 over this, but I thought he did not?
Anyone who follows the process a bit more than me have some context here?
People (from the committee of experts appointed as national body representatives or invited experts) submit proposals which is essentially merge requests to make changes to the C++ draft. A proposal, by the way of it's changes, can either add a feature to C++ or remove features from C++ (kind of like deleted code).
In each C++ meeting, they have a vote for each proposal in the meeting and decide whether that proposal is allowed to act on the C++, thereby determining if the features gets in the draft (if the proposal was proposing a feature) or if the features is kicked out of the draft (if the proposal was advocating for removal of a feature).
Once a proposal has been voted in the draft, only another proposal can remove that feature from the draft.
Every 3 years, they seal the current state of the draft into a standard and send it for voting to the official ISO C++ committee of national bodies (different from committee of experts we mentioned above). And I think each national body has a veto. (that is the vote has to be unanimous, but not sure here).
At this point, first the National bodies can make comments where they can threaten to veto the standard if their comment is not looked at. Then, the committee of experts can either respond to the comment or follow the comment or do whatever.
The official committee (essentially national body members) then votes and can only approve or reject the standard in the whole. So, if they reject of example C++ next now, there will be no C++26.
The whole process from sealing to actual voting takes around a year or so. In the mean time the committee of experts starts working on the next c++ standard from the sealed state forward without waiting for the committee to ratify the current one.
What Bjarne is doing is that he was threatening to get a national body to veto the proposal so that there is no C++26 at all. So, his point was, either accept his paper as it is (which calls for removal of contracts from the C++ draft before it goes to committee of national bodies), or he probably gets US or Denmark or some NB (not sure which) to vote no on the C++ standard (when it eventually goes to the committee of national bodies with the contracts in it, kind of like throwing baby with the bathwater).
The true goal of C++ is the self-preservation and funding of the ISO Committee specification “engine”, while compilers are merely an accidental by-product.
Once you see this then suddenly a lot of odd behavior makes sense. For example the committee can’t break ABI because then the old companies wouldn’t fund the new c++ standard, but would fund a committee that maintains/forks the old thing. The complexity / bureaucracy also provides jobs for those involved in many ways including writing new books every year and or consulting.
While end consumers might really care about compile speed, modularization, security, and all of the things modern languages give, that is secondary to keeping the committee alive.
Younger me would be like, oh no this is horrible someone will fix this! Older me is more sober and understands that once a system is up and running it will continue as is (this being an example of Pournelle’s Iron Law of Bureaucracy). Security by the way of migrations to Rust is causing what the committee sees as the first real threat to its existence as its patrons start adopting more and more Rust. As a result the committee is proposing to break compatibility as a means to keep them around just a little longer. Note: Actually fixing the security issues is not important only giving enough for sponsors to not leave so we will see what actually happens.
For context, under Leonardo Chiariglione's stewardship, MPEG attempted to ship a royalty-free codec called Internet Video Coding (IVC) and it was immediately torpedoed by Samsung claiming ownership. In any sane environment you would amend the standard to remove the claimed technology, but ISO specifically does not allow you to do this. There is no obligation for an ISO participant to identify the patents being claimed unless they refuse to offer a license on FRAND terms. Ergo, Samsung just said "we own this and will sell licenses for money", and IVC was dead in the water as a standard[0].
Leonardo attempted to bring this to the attention of ISO, but the actual formal process for this was blocked by... you guessed it, a national body vetoing it. So the patent policy that prohibits ISO from distinguishing between royalty-free and royalty-bearing codecs stayed. Oh, and then they cut MPEG up into six pieces so Leonardo would have nothing to run anymore.
C++ was an unusual language in that, for about 10-15 years, the only other competing systems programming language was C - another ISO standard, and the one C++ was originally based off of. The standard answer to "but I want memory safety" was "use a garbage collector"[1]. I distinctly remember hearing interviews with compiler developers talking about how adding even these basic features to C++ would be interminably painful, because you'd inevitably have to debug some deeply nested macro in a library somewhere, so shut up and just enjoy that we gave you Vector<T>.at().
The day Rust becomes an ISO standard is the day it becomes dead to me.
[0] IVC was specifically designed to be inferior to the royalty-bearing H.265 standard, as Leonardo himself wanted to keep the royalty-free codecs second-tier.
[1] To be clear, there was research into non-GC languages with memory safety - that's where Rust got its ideas from - but it was a very small niche until Mozilla put their feet down and gave legitimacy to something.
Okay, strictly speaking, Rust types are affine, not linear. Leaking Rust values is a safe operation. In a linear Rust the compiler would have to prove all values do not leak, but this would be incredibly difficult and probably make reference counting illegal.
Rust did not invent this (and I don’t believe Rust’s creators claim it did), see e.g. “Linear regions are all you need”[1] (ESOP 2006) and in general the work around Cyclone[2] in the 2000s. There was also Mezzo[3], which did very similar things but described them in a different style (effects instead of affine types), and a fair bit of work that went beyond what Rust can currently express (e.g. fractional capabilities). What Rust did is succeed at bringing a conservative subset of this academic work to production—which is not a small deal, mind you; others (including the Cyclone team) tried and failed.
[1] https://link.springer.com/chapter/10.1007/11693024_2, https://www.cs.cornell.edu/people/fluet/research/substruct-r...
[2] https://gallium.inria.fr/~fpottier/slides/fpottier-2007-05-l...
The actually innovative thing about Rust is how they made object lifetimes into a H-M-esque type system and used this 50 year old algorithm to detect dangling pointers at compile time.
They want to fund code they will use and that requires updating versions to be doable for like a million not twenty million.
Everyone loses site of the massive sizes of the codebases involved. When your codebase is measured in millions LoC minor breaks can be devastating.
Minor steps are fine and appreciated but "rip off the bandaid" is better left to alternative projects like Rust.
I think that my misunderstanding here is because I thought that the vote in Croydon back in March was the last chance to veto things, but re-reading https://herbsutter.com/2026/03/29/c26-is-done-trip-report-ma...
> are now producing the final document to be sent out for its international approval ballot
So yeah, I guess in most cases this final ballot is a rubber stamp (not a bad thing, actually, I'd argue it's the way it should be) but maybe this time it is not.
By the way, didn't realise when I answered this that I was answering to you. Thanks for all the work on Rust! Your views on Rust and C++ are both much appreciated.
Just to be extra clear about it, while there's clearly some dysfunction going on in WG21, I don't think Rust's process is free of it either. I do think there's a lot of pros to the way Rust does things, but this stuff is just very hard, in general.
I wish I had actually written "The elephant in the room" back when the safe c++/profiles stuff was going on, but I never actually did. Oh well. Probably wouldn't have changed much anyway.
No veto. Every P-member has a vote (abstain/approve/disapprove). Whether C++26 gets published is decided by numerical consensus.
I personally agree that contracts probably shouldn't be in the standard, but this whole situation looks like a mess.
Contracts has the problem that everybody has a slightly different idea of what they want from the feature. What exists today in the draft standard is a messy compromise that has a very broad consensus. However, there are few people who are vehemently against the current compromise and are taking every opportunity to try to upset that compromise.
Now on top of that, there is a separate issue in that the C++ standards committee has started to become too aggressive in adopting new features, and implementers are starting to complain about that. Contracts is one of the new features that is in the camp of the-standard-is-moving-too-quickly, but it's far from the only one and probably not the worst offender, given that there is fairly high demand for contracts. (Constexpr-all-the-things is probably the worst offender from a cost/benefit ratio?)
Currently, the C++26 draft is out to the national bodies for balloting, and the vote in the US national body passed by the thinnest of margins. I haven't heard of the results from other national bodies yet.
C++26 contracts are written by people with experience in ADA/SPARK. While we don't have experience in C++ contracts, there is plenty of experience elsewhere. I find it odd that the paper didn't mention Spark at all!
The criticism that it is experimental in all compilers is a fact that can never be anything else. Chicken and egg - nobody will make this non-experimental until it is in the standard. There have many small scale experienements with contracts - enough to agree we want to use this on a larger scale but we need it in the standard first.
This is a recurring pattern across large changes to c++. Modules and coroutines both are great examples of this.
> but it is good enough that they can start figuring out the details of making all the different factions happy.
I don’t think there’s really a plan to make the factions happy, there’s a plan to get X in, say it’s in and then ignore any actual criticism in favor of “it was the best we could do” while pointing at the other groups for not accepting the massive flaws.
Coroutines are ready since C++20, there is some stuff missing in the language (`for co_await` being the most obvious thing), but the feature is useful as is...
You might argue that modules aren’t the solution to compile times, but for a very long time they were. Now they’re a better way to structure code, except they’re not really backwards compatible. That dorsnt matter because the committee doesn’t consider that bit at the same time refuses other changes that may have similar impacts.
I think c++ is caught between a rock and a hard place. There’s a group that wants only existing behaviours standardised, and another that wants the language to be a force that brings the ecosystem with it. What we get is the latter group proposing changes that look like progress, but no (or limited) real change happening due to the first group, while the first group are forced to have the language change almost for the sake of it to placate the latter group.
What's wrong with static polymorphism that have been around for at least 31 years? (1995 per [1]). Also know as CRTP [2] Apart, of course, from ugly verbose compilation errors...
And now c++23 "deducing `this`" [3,4] modern replacement which is supposed to improve compilation time and remove the verbose errors.
[1] https://www.artima.com/articles/the-most-important-c-non-boo...
[2] https://en.cppreference.com/cpp/language/crtp
[3] https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2021/p08...
[4] https://devblogs.microsoft.com/cppblog/cpp23-deducing-this/
Take the example from the article. The contract trips at runtime, not at compile time.
There is a lot of interest in static analysis enforcing contracts, but this will likely require some additions to the contract specification.
> this will likely require some additions to the contract specification.
You need a lot more machinery to move this stuff to compile time, and not everything can be checked at compile time. The first example in the post would require you to validate that a <= v's len before indexing in order to be evaluated at compile time, for example. (Which of course is a runtime check anyway...)
By whole program analysis I mean you need the entire call tree of a function - stopping only when you can validate that the range of values is constrained to legal values. For some functions this is really simple, while for others the whole flow can be thousands of functions.
Also take a look at David Crocker's Escher Technologies "Escher C Verifier and Escher C++ Verifier" (https://www.eschertech.com/products/ecv.php) and articles on "Verified Design-By-Contract" etc. linked to in my comment chain here - https://news.ycombinator.com/item?id=48544022
Gcc can already do link-time devirtualization in some cases.
Traditionally you write code which checks the Widget to see if it's Gonzo and if not you throw an exception. Callers can pick, for this function in particular, whether to handle the Exception, in which case they get that Exception to look at, or they can "bubble it up" to be handled in their caller, and so on all the way to the top of the program where if it bubbles up it's reported and then exits the program.
With Contracts you write a contract for the function with a pre-condition that the Widget is Gonzo. Your users (programmers who might call doodle_widget) can pick: If they fail a contract the program exits immediately reporting a violation ("quick enforce"), it reports the violation via a global contract handler and then exits ("enforce") or it just reports to the handler but doesn't exit ("observe") or finally, they ignore it entirely ("ignore")
These just aren't that different. The contract is maybe slightly better because of the enhanced semantic discovery - you could imagine tooling which gives you a yellow squiggly line because your code violates a contract requirement for example, it's definitely not practical to check exception raising that way.
However contracts cover a lot of other cases (and as the other replies point out contracts are probably the wrong answer here - a concept is your right answer allow someone else to write a new/different Gonzo complaint widget in the fiture). A contract can check cases where have the right type, but something is wrong anyway. If you need a sorted list a contract can check that....
Likewise exception is useful for a lot of things that should not be a contract. Running out of disk space is still a common problem - you do not want a contract that there is enough disk space, this is an error you need to ask how to handle (often the user would free up disk space external to your program and retry a save).
Can't you just use concepts?
Obviously you could re-design the software to follow a Rust-style type-state paradigm, have a NonGonzoWidget and GonzoWidget type which both inherit from Widget and now you can have your function take a GonzoWidget - but that's not what my comparison was about and this technique while possible is less common in C++
Edited to add: Actually, I should clarify that concepts have an idea called "modelling" and you can model anything you want, but the problem is that the machine doesn't care. So you can have a concept which models strings being about how awesome Donald Trump is, and the compiler won't and can't check that, but now a C++ program with a string matching that concept and the value "Trump is a moron" is an invalid C++ program, still compiles, still probably works, but your concept wasn't "modelled" so the program was not valid for whatever that's worth....
But keep in mind that there are multiple kinds of exceptions.
Some exceptions, indicate scenarios the caller could reasonably have avoided. This include things like argument bound or null checks. A caller absolutely could avoid these exceptions by doing its own checks. This can be split between checks that the caller really should have done, like not passing null to a function that cannot take null, vs those where baking the relevant knowledge into the caller would be undesirable (perhaps because future versions of the library expect to accept more values).
There are exceptions that in theory a caller could avoid, but in practice it is impractical, like an exception thrown by a parser of some complex format if the provided input isn't legal. The only good way to avoid such an exception is to have a non-throwing parser that you can check with, but you probably don't want to parse twice. An alternative interface for the parser might be able to totally avoid the exception, returning either succeeded (with result tree) or fail with error message, but that would be a design choice of the implementor, not the caller.
There can be exceptions that indicate a logic bug in the impleentation (things like throwing if some invariant the implementation is in charge of is violated), but more often this is assertions instead.
Lastly, you have exceptions that there is no possible way the caller could always avoid. IO exceptions are among these. While you can sometimes do existence, space, or permission checks or similar to reduce the probability of getting certain exceptions, something else could race your app between the check and performing IO, and make them happen anyway.
The primary target for contracts are exceptions that callers both could and should have avoided, and the exception/assertion case that is trying to verify the implementation is working as expected.
The other categories of exceptions are more or less totally out of scope.