Now Reading
C23 is Completed: Right here is What’s on the Menu

C23 is Completed: Right here is What’s on the Menu

2022-08-05 01:42:47

It’s That Weblog Put up. The discharge one, the place we spherical up all of the final of the options accepted because the final time I blogged. Should you’re new right here, you’ll need to go take a look at these earlier articles to study extra about what’s/isn’t going into C23, a lot of them together with examples, explanations, and a few rationale:

The final assembly was fairly jam-packed, and plenty of issues made it by way of on the eleventh hour. We additionally misplaced fairly a couple of good papers and options too, in order that they’ll need to be reintroduced subsequent cycle, which could take us a complete additional 10 years to do. A few of us are agitating for a sooner launch cycle, primarily as a result of we’ve 20+ years of current observe we’ve successfully ignored and there’s plenty of work we ought to be doing to cut back that backlog considerably. It’s additionally simply no enjoyable ready for __attribute__((cleanup(…))) (defer), assertion expressions, higher bitfields, huge pointers (a local pointer + dimension assemble), a language-based generic perform pointer sort (GCC’s void(*)(void)) and like 20 different issues for one more 10 years after they’ve been round for many years.

However, I’ve digressed lengthy sufficient: let’s not speak in regards to the future, however the current. What’s in C23? Nicely, it’s all the things (sans the typo-fixes the Challenge Editors – me ‘n’ one other man – need to do) current in N3047. A few of them fairly huge blockbuster options for C (C++ will principally quietly chortle, however that’s wonderful as a result of C isn’t C++ and we take satisfaction in what we are able to get performed right here, with our group.) The primary large factor that can drastically enhance code is a combination-punch of papers written by Jens Gustedt and Alex Gilding.


I suppose most individuals didn’t see this one coming down the pipe for C. (Un?)Thankfully, C++ was extraordinarily profitable with constexpr and C implementations have been cranking out bigger and bigger fixed expression parsers for critical pace beneficial properties and to do extra sort checking and semantic evaluation at compile-time. Sadly, regardless of many compilers getting highly effective fixed expression processors, normal C simply continued to present everybody who ended up silently counting on these more and more beefy and good-looking compiler’s tips a huge center finger.

For instance, in my last post about C (or simply watching me post on Twitter), I defined how this:

const int n = 5 + 4;
int purrs[n];
// …

is a few extremely unlawful contraband. This creates a Variable-Size Array (VLA), not a constant-sized array with dimension 9. What’s worse is that the Common Compilers™ (GCC, Clang, ICC, MSVC, and most different optimizing compilers truly value compiling with) have been usually highly effective sufficient to principally flip the code era of this object – as long as you didn’t cross it to one thing anticipating an precise Variably-Modified Types (also talked about in another post) – into working like a traditional C array.

This left folks counting on the truth that this was a C array, though it by no means was. And it created sufficient confusion that we needed to settle for N2713 so as to add clarification to the Customary Library to inform those who No, Even If You Can Flip That Into A Fixed Expression, You Can’t Deal with It Like One For The Sake Of the Language. One option to pressure an error up-front if the compiler would probably flip one thing into not-a-VLA behind-your-back is to do:

const int n = 5 + 4;
int purrs[n] = { 0 }; // ????
// …

VLAs are usually not allowed to have initializers, so including one makes a compiler scream at you for daring to jot down one. In fact, should you’re a type of S P E E D junkies, this might waste valuable cycles in probably initializing your knowledge to its 0 bit illustration. So, there actually was no option to win when coping with const right here, regardless of everybody’s psychological mannequin – because of the phrase’s origin in “fixed” – latching onto n being a continuing expression. Compilers “by chance” supporting it by both not treating it as a VLA (and requiring the paper I linked to earlier to be added to C23 as a clarification), or treating it as a VLA however extension-ing and efficient-code-generating the issue away simply resulted in a single too many portability points. So, in true C style, we added a third means that was DEFINITELY unmistakable:

constexpr int n = 5 + 4;
int purrs[n] = { 0 }; // ✅
// …

The brand new constexpr key phrase for C means you don’t need to guess at whether or not that could be a fixed expression, or hope your compiler’s optimizer or frontend is highly effective sufficient to deal with it like one to get the code era you need if VLAs with different extensions are on-by-default. You’re assured that this object is a continuing expression, and if it’s not the compiler will loudly yell at you. Whereas doing this, the wording for fixed expressions was additionally improved dramatically, permitting:

  • compound literals (with the constexpr storage class specifier);
  • buildings and unions with member entry by .;
  • and, the standard arithmetic / integer fixed expressions,

to all be fixed expressions now.

Oh No, These Evil Committee Persons are Ruining™ my Favourite LanguageⓇ with C++ NonsenseⒸ!

Truthfully? I sort of want I may break C generally, however imagine it or not: we are able to’t!

Word that there are not any perform calls included on this, so no person has to flip out or fear that we’re going to go the C++ route of stacking on a thousand completely different “please make this perform constexpr so I can commit compile-time crimes”. It’s only for objects proper now. There may be curiosity in doing this for features, however in contrast to C++ the intent is to supply a stage of constexpr features that’s so weak it’s worse than even the very first C++11 constexpr mannequin, and considerably worse than what GCC, Clang, ICC, and MSVC can present at compile-time proper now of their C implementations.

That is to maintain it straightforward to implement analysis in smaller compilers and stop feature-creep just like the C++ characteristic. C can be protected against further characteristic creep as a result of, in contrast to C++, there’s no template system. What justified half of the enhancements to constexpr features in C++ was “properly, if I simply rewrite this perform in my favourite Purposeful Language – C++ Templates! – and tax the compiler even more durable, I can do precisely what I need with worse compile-time and much more object file bloat”. This was a scary consideration for a lot of on the Committee, however we is not going to truly go that route exactly as a result of we’re within the C language and never C++.

You can’t look sideways and squint and say “properly, if I simply write this in essentially the most tousled means potential, I can compute a continuing expression on this backdoor Turing full practical language”; it simply doesn’t exist in C. Subsequently, there isn’t a prior artwork or justification for an ever-growing collection of fixed expression library features or marked-up headers. Even when we get constexpr features, will probably be actually and deliberately be underpowered and weak. It is going to be so unhealthy that the very best you are able to do with it’s write a non-garbage max perform to make use of because the behind-the-scenes for a max macro with _Generic. Or, possibly substitute a couple of macros with one thing small and tiny.

Some folks will have a look at this and go: “Nicely. That’s crap. The explanation I exploit constexpr in my C++-like-C is so I can write beefy compile-time features to do a number of heavy computation as soon as at a compile-time, and have it up-to-date with the construct on the identical time. I can actually crunch an ideal hash or create an ideal desk that’s hardware-specific and tailor-made while not having to drop right down to platform-specific tips. If I can’t try this, then what good is that this?” And it’s a great collection of questions, pricey reader. However, my response to this for many C programmers craving for higher is that this:

we get what we shill for.

With C we don’t in the end have the collective will or implementers courageous sufficient to take-to-task making a big fixed expression parser, even when the C language is lots less complicated to jot down one for in comparison with C++. On daily basis we preserve proclaiming C is an easy and delightful language that doesn’t want options, even options which can be compile-time solely with no runtime overhead. Meaning, sooner or later, the one sort of fixed features on the desk are ones with no recursion, just one single assertion allowed in a perform physique, plus further restrictions to get in your means. However that’s a part of the attraction, proper? The compilers could also be weak, the code era could also be terrible, more often than not you need to abandon truly working in C and as an alternative simply use it as a macro assembler and drop right down to bespoke, hand-written platform-specific meeting nested in a god-awful compiler-version-specific #ifdef, however That’s The Shut-To-The-Steel C I’m Talkin’ About, Babyyyyy!!

“C is easy” additionally means “the C normal is underpowered and can’t adequately categorical all the things it’s essential to get the job performed”. However should you ask your vendor properly and promise them cash, cookies, and ice cream, possibly they’ll deign at hand you one thing good. (However will probably be exterior the usual, so I hope you’re able to put an costly ring ???? in your vendor’s finger and marry them.)


Earlier, I form of glazed over the truth that Compound Literals at the moment are a part of issues that may be fixed expressions. Nicely, that is the paper that allows such a factor! It is a characteristic that truly solves an issue C++ was having as properly, whereas additionally fixing plenty of annoyances with C. For these of you at the hours of darkness and who haven’t caught up with C99, C has a characteristic known as Compound Literals. It’s a option to create any sort – normally, buildings – which have an extended lifetime than regular and may act as a brief going right into a perform. They’re used fairly often in code examples and stuff performed by Andre Weissflog of sokol_gfx.h fame, who writes some fairly lovely C code (excerpted from the hyperlink):

#outline SOKOL_IMPL
#outline SOKOL_GLCORE33
#embody <sokol_gfx.h>
#embody <GLFW/glfw3.h>

int principal(int argc, char* argv[]) {

	/* create window and GL context by way of GLFW */
	/* … CODE ELIDED … */

	/* setup sokol_gfx */
	sg_setup(&(sg_desc){0}); // ❗ Compound Literal

	/* a vertex buffer */
	const float vertices[] = {
		// positions            // colours
		0.0f,  0.5f, 0.5f,     1.0f, 0.0f, 0.0f, 1.0f,
		0.5f, -0.5f, 0.5f,     0.0f, 1.0f, 0.0f, 1.0f,
		-0.5f, -0.5f, 0.5f,     0.0f, 0.0f, 1.0f, 1.0f
	sg_buffer vbuf = sg_make_buffer(&(sg_buffer_desc){  // ❗ Compound Literal
		.knowledge = SG_RANGE(vertices)

	/* a shader */
	sg_shader shd = sg_make_shader(&(sg_shader_desc){  // ❗ Compound Literal =
			"#model 330n"
			"format(location=0) in vec4 place;n"
			"format(location=1) in vec4 color0;n"
			"out vec4 coloration;n"
			"void principal() {n"
			"  gl_Position = place;n"
			"  coloration = color0;n"
			"}n", =
			"#model 330n"
			"in vec4 coloration;n"
			"out vec4 frag_color;n"
			"void principal() {n"
			"  frag_color = coloration;n"

	/* … CODE ELIDED … */
	return 0;

C++ doesn’t have them (although GCC, Clang, and some different compilers assist them out of necessity). There may be a paper by Zhihao Yuan to assist Compound Literal syntax in C++, however there was a dangle up. Compound Literals have a particular lifetime in C known as “block scope” lifetime. That’s, compound literals in features behave as-if they’re objects created within the enclosing scope, and subsequently retain that lifetime. In C++, the place we’ve destructors, unnamed/invisible C++ objects being l-values (objects whose deal with you possibly can take) and having “Block Scope” lifetime (lifetime till the place the subsequent } was) resulted within the standard intuitive conduct of C++’s temporaries-passed-to-functions turning right into a nightmare.

For C, this didn’t matter and – in lots of instances – the conduct was even relied on to have longer-lived “temporaries” that survived past the period of a perform name to, say, chain with different perform calls in a macro expression. For C++, this meant that some kinds of RAII useful resource holders – like mutexen/locks, or simply knowledge holders like dynamic arrays – would maintain onto the reminiscence for means too lengthy.

The conclusion from the most recent dialog was “we are able to’t have compound literals, as they’re, in C++, since C++ gained’t take the semantics of how they work from the C normal of their implementation-defined extensions and not one of the implementations need to change conduct”. Which is fairly crappy: taking an extension from C’s syntax after which sort of simply… smearing over its semantics is a little bit of a rotten factor to do, even when the brand new semantics are higher for C++.

However, Jens Gustedt’s paper saves us plenty of the difficulty. Whereas default, plain compound literals have “block scope” (C) or “non permanent r-value scope” (C++), with the brand new storage-class specification characteristic, you possibly can management that. Borrowing the sg_setup perform above that takes the sg_desc construction sort:

#embody <sokol_gfx.h>

SOKOL_GFX_API_DECL void sg_setup(const sg_desc *desc);

we’re going to add the static modifier, which signifies that the compound literal we create has static storage period:

int principal (int argc, const char* argv[]) {
	/* … CODE ELIDED … */
	/* setup sokol_gfx */
	sg_setup(&(static sg_desc){0}); // ❗ Compound Literal	
	/* … CODE ELIDED … */

Equally, auto, thread_local, and even constexpr can go there. constexpr is probably essentially the most pertinent to folks right this moment, as a result of proper now utilizing compound literals in initializers for const knowledge is technically SUPER unlawful:

typedef struct crime {
    int criming;
} crime;

const crime crimes = (crime){ 11 }; // ❗ ILLEGAL!!

int principal (int argc, char* argv[]) {
    return crimes.criming;

It is going to work on plenty of compilers (unless warnings/errors are cranked up), however it’s much like the VLA state of affairs. The minute a compiler decides to get snooty and choosy, they’ve all of the justification on the earth as a result of the usual is on their facet. With the brand new constexpr specifier, each buildings and unions are thought of fixed expressions, and it can be utilized to compound literals as properly:

typedef struct crime {
    int criming;
} crime;

const crime crimes = (constexpr crime){ 11 }; // ✅ LEGAL BABYYYYY!

int principal (int argc, char* argv[]) {
    return crimes.criming;



Go learn this to search out out all in regards to the characteristic and the way a lot of a bloody pyrrhic victory it was.


This paper was a very long time coming. C++ received it first, making it barely hilarious that C harps on standardizing current observe a lot however C++ tends to beat it to the punch for options which clear up long-standing Preprocessor shenanigans. Should you’ve ever had to make use of __VA_ARGS__ in C, and also you wanted to cross 0 arguments to that , or attempt to use a comma earlier than the __VA_ARGS__, you recognize that issues received genuinely tousled when that code needed to be ported to different platforms. It received a special entry in GCC’s documentation due to how blech the state of affairs ended up being:

… GNU CPP allows you to fully omit the variable arguments on this means. Within the above examples, the compiler would complain, although because the enlargement of the macro nonetheless has the additional comma after the format string.

To assist clear up this downside, CPP behaves specifically for variable arguments used with the token paste operator, ‘##’. If as an alternative you write

#outline debug(format, …) fprintf (stderr, format, ## __VA_ARGS__)

and if the variable arguments are omitted or empty, the ‘##’ operator causes the preprocessor to take away the comma earlier than it. Should you do present some variable arguments in your macro invocation, GNU CPP doesn’t complain in regards to the paste operation and as an alternative locations the variable arguments after the comma. …

That is solved by way of the C++-developed __VA_OPT__, which expands out to a authorized token sequence if and provided that the arguments handed to the variadic are usually not empty. So, the above might be rewritten as:

#outline debug(format, …) fprintf (stderr, format __VA_OPT__(,) __VA_ARGS__)

That is secure and accommodates no extensions now. It additionally avoids any preprocessor undefined conduct. Moreover, C23 permits you to cross nothing for the argument, giving customers a means out of the earlier constraint violation and murky implementation behaviors. It really works in each the case the place you write debug("meow") and debug("meow", ) (with the empty argument handed explicitly). It’s a very elegant design and we’ve Thomas Köppe to thank for bringing it to each C and C++ for us. This may permit a very nice normal conduct for macros, and is very good for formatting macros that now not have to do bizarre tips to special-case for having no arguments.

Which, talking of 0-argument features…


This paper is fairly easy. It acknowledges that there’s actually no motive to not permit

to exist in C. C++ has it, and all of the arguments get handed efficiently, and no person’s misplaced any sleep over it. It was additionally a vital filler since, as talked about in old blog posts, we’ve lastly taken the older perform name model and put it down after 30+ years of being in existence as a characteristic that by no means received to see a single correct C normal launch non-deprecated. This was nice! Besides, as that earlier weblog put up mentions, we had no means of getting a general-purpose Utility Binary Interface (ABI)-defying perform name anymore. That turned out to be unhealthy sufficient that after the deprecation and removing we wanted to push for a repair, and fortunate for us void f(…); had not made it into normal C but.

So, we put it in. Not needing the primary parameter, and now not requiring it for va_start, meant we may present a clear transition path for everybody counting on Okay&R features to maneuver to the -based perform calls. Because of this mechanical upgrades of outdated codebases – with instruments – is now on-the-table for migrating outdated code to C23-compatibility, whereas lastly placing Okay&R perform calls – and all their lack of security – within the dust. 30+ years, however we may lastly capitalize on Dennis M. Ritchie’s dream right here, and put these perform calls to mattress.

In fact, compilers that assist each C and C++, and compilers that already had void f(…); features as an extension, could have deployed an ABI that’s incompatible with the outdated Okay&R declarations of void f();. Because of this a mechanical improve might want to examine with their distributors, and:

  • guarantee that this occupies the identical calling conference;
  • or, the one who is looking the perform can’t replace the opposite facet that is perhaps pulling meeting/utilizing a special language,

then the improve that goes by way of to exchange each void f(); could have to additionally add a vendor attribute to ensure the perform calling conference is suitable with the outdated Okay&R one. Personally, I counsel:

[[vendor::kandr]] void f();

, or one thing related. However, ABI exists exterior the usual: you’ll want to speak to your vendor about that one while you’re able to port to an exlusively-post-C23 world. (I doubt anybody will compile for an solely C23-and-above world, however it’s good to know there’s a well-defined migration path for customers nonetheless hook up a 30+ yr deprecated characteristic). Astute readers could discover that in the event that they don’t have a parameter to go off of, how do they commit stack-walking sins to get to the arguments? And, properly, the reply is you continue to can: ztd.vargs has a proof-of-concept of that (on Home windows). You continue to want some option to get the stack pointer in some instances, however that’s been one thing compilers have supplied as an intrinsic for some time now (or one thing you possibly can do by committing register crimes). In ztd.vargs, I needed to drop down into meeting to start out fishing for stuff extra immediately once I couldn’t commit extra direct built-in compiler crimes. So, that is everybody’s probability to get actually in contact with that bare-metal they preserve bragging about for C. Polish off these dusty manuals and compiler docs, it’s time to get intimately conversant in all of the sins the platform is doing on the down-low!


What can I say about this paper, besides…

What The Hell, Man?

It’s completely bananas to me that in C – the techniques programming language, the language the place const int n = 5 isn’t a continuing expression so folks inform you to make use of enum { n = 5 } as an alternative – simply had this example happening, since its inceptions. “16 bits is sufficient for everybody” is what Unicode stated, and we paid for it by having UTF-16, a most restrict of 21 bits for our Unicode code factors (“Unicode scalar values” should you’re a nerd), and all the C and C++ normal libraries with respect to textual content encoding simply being fully not possible to make use of. (On high of the library not working for Big5-HKSCS as a multibyte/slender encoding). So in fact, once I lastly sat down with the C normal and skim by way of the factor, noticing that enumeration constants “should be representable by an int” was the precise wording in there was infuriating. 32 bits could also be good, however there have been loads of platforms the place int was nonetheless 16 bits. Worse, should you put code right into a compiler the place the worth was too huge, not solely would you not get errors on most compilers, you’d generally simply get straight up miscompiles. This isn’t as a result of the compiler vendor is a jerk or unhealthy at their job; the usual actually simply telephones it in, and each compiler from ICC to MSVC allow you to go previous the low 16-bit restrict and sometimes even exceed the 32-bit INT_MAX with out a lot as a warning. It was a nugatory clause in the usual,

and it took lots out of me to combat to right this one.

The paper subsequent on this weblog put up was seen because the repair, and we determined that the outdated code – the code the place folks used 0x10000 as a bit flag – was simply going to be non-portable rubbish. Did you go to a compiler the place int is 16 bits and INT_MAX is smaller than 0x10000? Congratulations: your code was non-standard, you’re now in implementation-defined territory, pound sand! It took plenty of convincing, practically received voted down the primary time we took a critical ballot on it (simply barely scraped by with consensus), however the paper rolled in to C23 on the final assembly. An enormous shout out to Aaron Ballman who described this paper as “value-preserving”, which went a extremely great distance in connecting everybody’s understanding of how this was meant to work. It added a really express algorithm on do the computation of the enumeration fixed’s worth, in order that it was giant sufficient to deal with constants like 0x10000 or ULLONG_MAX. It retains it to be int wherever potential to protect the semantics of outdated code, but when somebody exceeds the scale of int then it’s truly authorized to improve the backing sort now:

enum my_values {
	a = 0, // 'int'
	b = 1, // 'int'
	c = 3, // 'int'
	d = 0x1000, // 'int'
	f = 0xFFFFF, // 'int' nonetheless
	g, // implicit +1, on 16-bit platform upgrades sort of the fixed right here
	e = g + 24, // makes use of "present" sort of g - 'lengthy' or 'lengthy lengthy' - to do math and set worth to 'e'
	i = ULLONG_MAX // 'unsigned lengthy' or 'unsigned lengthy lengthy' now

When the enumeration is accomplished (the closing brace), the implementation will get to pick a single sort that my_values is suitable with, and that’s the kind used for all of the enumerations right here if int isn’t large enough to carry ULLONG_MAX. Meaning this subsequent snippet:

int principal (int argc, char* argv[]) {
	// when enum is full,
	// it will probably choose any sort
	// that it needs, as long as its
	// large enough to signify the kind
	return _Generic(a,
		unsigned lengthy: 1,
		unsigned lengthy lengthy: 0,
		default: 3);

can nonetheless return any of 1, 0, or 3. However, on the very least, you recognize a, or g or i won’t ever truncate or lose the worth you set in as a continuing expression, which was the objective. The sort was at all times implementation-defined (see: -fshort-enum shenanigans of outdated). All of that outdated code that was flawed is now now not flawed. All of these individuals who tried to jot down wrappers/shims for OpenGL who used enumerations for his or her integer-constants-with-nice-identifier-names are additionally now right, as long as they’re utilizing C23. (That is additionally one motive why the OpenGL constants in among the unique OpenGL code are written as preprocessor defines (#outline GL_ARB_WHATEVER …) and never enumerations. Enumerations would break with any of the OpenGL values above 0xFFFF on embedded platforms; they needed to make the transfer to macros, in any other case it was busted.)

Suffice to say I’m extraordinarily completely happy this paper received it and that we retroactively fastened plenty of code that was not imagined to be compiling on plenty of platforms, in any respect. The underlying sort of an enumeration can nonetheless be some implementation-defined integer sort, however that’s what this subsequent paper is for…


This was the paper everybody was actually after. It additionally received in, and moderately than being about “value-preservation”, it was about sort preservation. I may write lots, however whitequark – as standard – describes it greatest:

i spotted right this moment that C is so unhealthy at its job that it wants the assistance of C++ to make some options of its ABI usable (since you possibly can specify the width of an enum in C++ however not C)

Catherine (@whitequark), May 25th, 2020

C getting dumpstered by C++ is a typical prevalence, however truthfully? For a characteristic like this? It’s past unacceptable that C couldn’t give a selected sort for its enumerations, and subsequently made the usage of enumerations in e.g. bit fields or related toxic, unhealthy, and non-portable. There’s already a lot to take care of in C to jot down good close-to-the-hardware code: now we are able to’t even use enumerations portably with out 5000 static checks particular flags to ensure we received the fitting sort for our enumerations? Utter hogwash and a blight on the entire C group that it took this lengthy to repair the issue. However, as whitequark additionally said:

on this case the answer to “C is unhealthy at its job” is certainly to “repair C” as a result of, even should you hate C a lot you need to eradicate it fully from the face of the earth, we’ll nonetheless be caught with the C ABI lengthy after it’s gone

Catherine (@whitequark), May 25th, 2020

It was time to roll up my sleeves and do what I at all times did: take these abominable programming languages to activity for his or her inexcusably poor conduct. The worst half is, I nearly let this paper slip by as a result of another person – Clive Pygott – was dealing with it. In actual fact, Clive was dealing with this even earlier than Catherine made the tweet; N2008, from…

oh my god, it’s from 2016.

I had not realized Clive had been engaged on it this lengthy till, throughout one assembly, Clive – when requested in regards to the standing of an up to date model of this paper – stated (paraphrasing) “yeah, I’m not carrying this paper ahead anymore, I’m drained, thanks”.

That’s not, uh, good. I rapidly snapped up in my chair, slammed the Mute-Off button, and practically fumbled the mechanical mute on my microphone as I sputtered a bit of so I may communicate up: “hey, uh, Clive, may you ahead me all of the suggestions for that paper? There’s lots of people that need this characteristic, and it’s actually vital to them, so ship me all of the suggestions and I’ll see if I can do one thing”. True to Clive’s phrase, minutes after the ultimate day on the mid-2021 assembly, he despatched me all of the notes. And it was…

… lots.

I didn’t notice Clive had this a lot push again. It was late 2021. 2022 was across the nook, we have been principally out of time to workshop stuff. I often went to twitter and ranted about enumerations, from October 2021 and onward. The worst half is, most individuals didn’t know, so they simply assumed I used to be cracked up about one thing till I pointed them to the phrases in the usual after which revealed all of the non-standard conduct. Really, the C specification for enumerations was one thing terrible.

In fact, regardless of how a lot I fumed, anger is ineffective with out route.

I honed that virulent ranting right into a weapon: two papers, that finally grew to become what you’re studying about now. N3029 and N3030 was the crystallization of how a lot I hated this a part of C, hated it’s specification, loathed the best way the Committee labored, and despised a course of that led us for over 30 years to finish up on this precise spot. This man – Clive – had been at this since 2016. It’s 2022. 5 years in, he gave up attempting to placate all of the suggestions, and that left me just one yr to scrub these items up.

Truthfully, if I didn’t have a bizarre righteous anger, the paper would’ve by no means made it.

By no means underestimate the facility of anger. Numerous people and plenty of cultures spend time attempting to get you to “handle your feelings” and “discover serenity”, typically to the whole exclusion of getting mad at issues. You wanna know what I believe?

???? ““Serenity””

Serenity, peace, all of that may be taken and shoved the place the solar don’t shine. We have been delivered a sizzling rubbish language, made Clive Pygott – one of many smartest folks engaged on the C Reminiscence Mannequin – gargle Committee suggestions for five years, get caught in a rocky specification, and in the end abandon the hassle. Then, we needed to do some heroic enhancing and WAY an excessive amount of time of three folks – Robert Seacord, Jens Gustedt, and Joseph Myers – simply to hammer it into form whereas I needed to drag that factor kicking and screaming throughout the end line. Even I can’t preserve that up for a very long time, particularly with all of the work I additionally needed to do with #embed and Trendy Bit Utilities and 10+ different proposals I used to be combating to repair. “Indignant” is kind of frankly not a robust sufficient phrase to explain a course of that may make one thing so obligatory spin its wheels for five years. It’s completely bananas that is how ISO-based, Committee-based work needs to be performed. To all the opposite languages eyeing the mantle of C and C++, considering that an precise under-ISO working group will present something to them.

Do. Not.

Nothing about ISO or IEC or its numerous subcommittees incentivizes progress. It incentivizes limitless suggestions loops, heavy weighted processes, particular person burn out, and low return-on-investment. Do something – actually something – else along with your time. Should you want the ISO sticker since you need ASIL A/B/C/D certification in your language, than by all means: determine a option to make it work. However preserve your core course of, your core suggestions, your core identification out of ISO. You possibly can standardize current practices means higher than this, and with out practically this a lot gnashing of tooth and pullback. Regardless of how politely its structured, the insurance policies of ISO and the best way it expects Committees to be structured is a deeply-embedded type of bureaucratic violence towards the least of those, its contributors, and also you deserve higher than this. A lot of this CIA sabotage field manual’s list:

A picture of the CIA handbook covering lessons (1) through (8), which you can read at the link which covers more detail. From the pictured page: (1) Insist on doing everything through “channels.” Never permit short-cuts to be taken in order to expedite decisions. (2) Make “speeches.” Talk as frequently as possible and at great length. Illustrate your “points” by long anecdotes and accounts of personal experiences. (3) When possible, refer all matters to committees, for “further study and consideration.” Attempt to make the committee as large as possible — never less than five. (4) Bring up irrelevant issues as frequently as possible. (5) Haggle over precise wordings of communications, minutes, resolutions. (6) Refer back to matters decided upon at the last meeting and attempt to re-open the question of the advisability of that decision. (7) Advocate “caution.” Be “reasonable” and urge your fellow-conferees to be “reasonable” and avoid haste which might result in embarrassments or difficulties later on. (8) Be worried about the propriety of any decision - raise the question of whether such action as is contemplated lies within the jurisdiction of the group or whether it might conflict with the policy of some higher echelon.

shouldn’t have a directly-applicable analogue that describes how an Worldwide Requirements Group conducts enterprise. However whether it is what it’s, then it’s time to roll up the sleeves. Don’t be sad. Get mad. Get even.

In any case, enumerations. You possibly can add varieties to them:

enum e : unsigned quick {

int principal (int argc, char* argv[]) {
    return _Generic(x, unsigned quick: 0, default: 1);

Not like earlier than, this can at all times return 0 on each platform, no exceptions. You possibly can stick it in buildings and unions and use it with bitfields and so long as your implementation isn’t fully off its rocker, you’re going to get fully reliable alignment, padding, and sizing conduct. Take pleasure in! ????


It is a comparatively easy paper, however closes up a gap that’s existed for some time. Nominally, it’s undefined-behavior to change an originally-const array – particularly a string literal – by way of a non-const pointer. So,

why precisely was strchr, bsearch, strpbrk, strrchr, strstr, memchr, and their huge counterparts principally taking const in and stripping it out within the return worth?

The reason being as a result of these needed to be singular features that outlined a single externally-visible perform name. There’s no overloading in C, so again within the outdated days when these features have been cooked up, you possibly can solely have one. We couldn’t exclude individuals who wished to jot down into the returned pointers of those features, so we made the optimum (on the time) selection of merely eradicating the const from the return values. This was not preferrred, however it received us by way of the door.

Now, with type-generic macros within the desk, we shouldn’t have this limitation. It was only a matter of somebody getting creative sufficient and writing the specification up for it, and that’s precisely what Alex Gilding did! It appears a bit of humorous within the standardese, however:

See Also

#embody <string.h>
QChar *strchr(QChar *s, int c);

Describes that should you cross in a const-qualified char, you get again a const-qualified char. Equally if there isn’t a const. It’s a pleasant little addition that may assist enhance read-only reminiscence security. It’d imply that individuals utilizing any one of many aforementioned features as a free-and-clear “UB-cast” to trick the compiler must fess up and use an actual forged as an alternative.


To me, this one was a bit clearly in want, although not everybody thinks so. For a very long time, folks preferred utilizing NULL, (void*)0, and literal 0 because the null pointer fixed. And they’re actually not flawed to take action: the primary one in that record is a preprocessor macro resolving to both of the opposite 2. Whereas nominally it could be good if it resolved to the primary, compatibility for older C library implementations and the code constructed on high of it calls for that we not change NULL. In fact, this made for some fascinating issues in portability:

#embody <stdio.h>

int principal (int argc, char* argv[]) {
	printf("ptr: %p", NULL); // oops
	return 0;

Now, no person’s passing NULL on to printf(…), however in a roundabout means we had NULL – the macro itself – filtering down into perform calls with variadic arguments. Or, extra critically, we had folks simply passing straight up literal 0. “It’s the null pointer fixed, that’s completely wonderful to cross to one thing anticipating a pointer, proper?” This was, in fact, flawed. It could be good if this was true, however it wasn’t, and on sure ABIs that had penalties. The identical registers and stack areas for passing a pointer weren’t at all times the identical as have been used for literal 0 or – worse – they have been the identical, however the literal 0 didn’t fill in all of the anticipated area of the register (32-bit vs. 64-bit, for instance). That meant folks doing prtinf("%p", 0); in some ways have been relying purely on the luck of their implementation that it wasn’t producing precise undefined conduct! Whoops.

nullptr and the related nullptr_t sort in <stddef.h> fixes that downside. You possibly can specify nullptr, and it’s required to have the identical underlying illustration because the null pointer fixed in char* or void* kind. This implies it’s going to at all times be handed accurately, for all ABIs, and also you gained’t learn rubbish bits. It additionally aids within the case of _Generic: with NULL being implementation-defined, you possibly can find yourself with void* or 0. With nullptr, you get precisely nullptr_t: this implies you don’t have to lose the _Generic slot for each int or void*, particularly should you’re anticipating precise void* pointers that time to stuff. Small addition, eliminates some Undefined Habits instances, good change.

Somebody just lately challenged me, nonetheless: they stated this variation isn’t obligatory and bollocks, and we should always merely pressure everybody to outline NULL to be void*. I stated that in the event that they’d like that, then they need to go to these distributors themselves and ask them to alter and see the way it goes. They stated they might, and so they’d like an inventory of distributors defining NULL to be 0. Drawback: fairly a couple of of them are proprietary, so right here’s my Open Problem:

should you (sure, you!!) have gotten a C normal library (or shim/substitute) the place you outline NULL to be 0 and never the void-pointer model, ship me a mail and I’ll get this particular person in contact with you so you possibly can duke it out with one another. In the event that they handle to persuade sufficient distributors/maintainers, I’ll persuade the Nationwide Physique I’m with to jot down a Nationwide Physique Remark asking for nullptr to be rescinded. In fact, they’ll have to not solely attain out to those folks, however persuade them to alter their NULL from 0 to ((void*)0), which. Nicely.

Good luck to the one who signed up for this.


Keep in mind how there have been all these directions obtainable since like 1978 – you recognize, within the Earlier than Instances™, earlier than I used to be even born and my dad and mom have been nonetheless younger? – and the way we had quick access to them by way of all our C compilers as a result of we rapidly standardized current observe from final century?

… Yeah, I don’t keep in mind us doing that both.

Trendy Bit Utilities isn’t a lot “trendy” as “catching as much as 40-50 years in the past”. There have been some specification issues and I spent means an excessive amount of time combating on so many fronts that, finally, one thing needed to undergo: though the paper offers wording for Rotate Left/Proper, 8-bit Endian-Conscious Hundreds/Shops, and 8-bit Reminiscence Reversal (fancy means of claiming, “byteswap”), the specification had too many tiny points in it that opposition mounted to stop it from being included-and-then-fixed-up-during-the-C23-commenting-period, or simply included in any respect. I used to be additionally too drained by the final assembly day, Friday, to really attempt to combat exhausting for it, so though a couple of different members of WG14 sacrificed half-hour of their block to get Rotate Left/Proper in, others insisted that they wished to do the Rotate Left/Proper features in a special model. I used to be too drained to combat too exhausting over it, so I made a decision to simply defer it to post-C23 and are available again later.


Nonetheless, with the brand new <stdbit.h>, this paper offers:

  • stdc_popcount
  • stdc_bit_width
  • stdc_leading_zeroes/stdc_leading_ones/stdc_trailing_zeros/stdc_trailing_ones
  • stdc_first_leading_zero/stdc_first_leading_one/stdc_first_trailing_zero/stdc_first_trailing_one
  • stdc_has_single_bit
  • stdc_bit_width
  • stdc_bit_ceil
  • stdc_bit_floor

“The place’s the endian macros for Honeywell architectures or PDP endianness?” You will get that if __STDC_ENDIAN_NATIVE__ isn’t equal to both the little OR the massive macro:

#embody <stdbit.h>
#embody <stdio.h>

int principal (int argc, char* argv[]) {
		printf("little endian! uwun");
	else if (__STDC_ENDIAN_NATIVE__ == __STDC_ENDIAN_BIG__) {
		printf("huge endian OwO!n");
	else {
		printf("what is that this?!n");
	return 0;

Should you fall into the final department, you have got some bizarre endianness. We don’t present a macro for that title as a result of there may be an excessive amount of confusion round what the precise correct byte order for “PDP Endian” or “Honeywell Endian” or “Bi Endian” would find yourself being.

“What’s that ugly stdc_ prefix?”

For the bit features, a prefix was added to them within the type of stdc_…. Why?

popcount is a actually well-liked perform title. If the usual have been to take it, we’d successfully be loading up a gun to shoot a ton of current codebases proper within the face. The one correct decision I may get to the issue was including stdc_ in entrance. It’s not preferrred, however truthfully it’s the very best I may do on quick discover. We shouldn’t have namespaces in C, which suggests any time we add performance we principally need to sq. off with customers. It’s most actually not a enjoyable a part of proposal improvement, for positive: thus, we get a stdc_ prefix. Maybe will probably be the primary of many features to make use of such prefixes so we shouldn’t have to step on person’s toes, however I think about for enhancements and fixes to current performance, we are going to preserve writing perform names by the outdated guidelines. This will probably be determined later by a coverage paper, however that coverage paper solely applies to papers after C23 (and after we get to have that dialogue).


It is a fairly easy paper, all issues thought of. Should you ever used __auto_type from GCC: that is that, with the title auto. I describe it like this as a result of it’s explicitly not like C++’s auto characteristic: it’s considerably weaker and much more restricted. Whereas C++’s auto permits you to declare a number of variables on the identical line and even deduce partial qualifiers / varieties with it (reminiscent of auto* ptr1 = factor, *ptr2 = other_thing; to demand that factor and other_thing are some sort of pointer or convertible to 1), the C model of auto is modeled fairly immediately after the weaker model of __auto_type. You possibly can solely declare one variable at a time. There’s no pointer-capturing. And so forth, and so forth:

int principal (int argc, char* argv[]) {
	auto a = 1;
	return a; // returns int, no mismatches

It’s most helpful in macro expressions, the place you possibly can keep away from having to duplicate expressions with:

#outline F(_NAME, ARG, ARG2, ARG3) 
	typeof(ARG + (ARG2 || ARG3)) _NAME = ARG + (ARG2 | ARG3);

int principal (int argc, char* argv[]) {
	F(a, 1, 2, 3);
	return a;

as an alternative being written as:

#outline F(_NAME, ARG, ARG2, ARG3) 
	auto _NAME = ARG + (ARG2 | ARG3);

int principal (int argc, char* argv[]) {
	F(a, 1, 2, 3);
	return a;

Being much less liable to make refined or small errors that will not be caught by the compiler you’re utilizing is sweet, in relation to working with particular expressions. (You’ll discover the left hand facet of the _NAME definition within the first model had a refined typo. Should you did: congratulations! Should you didn’t: properly, auto is for you.) Expressions in macros can get exceedingly difficult, and worse if there are unnamed structs or related getting used it may be hard-to-impossible to call them. auto makes it potential to understand these varieties and use them correctly, leading to a smoother expertise.

Regardless of being a easy characteristic, I count on this will probably be some of the divisive for C programmers. Individuals already took to the streets in a couple of locations to declare C a useless language, completely ruined by this variation. And, as a Committee member, if that truly finally ends up being the case? If this truly finally ends up fully destroying C for any of the explanations folks have towards auto and sort inference for a language that fairly actually simply allow you to fully elide varieties in perform calls and gave you “implicit int” conduct that compilers right this moment nonetheless need to assist in order that issues like OpenSSL can nonetheless compile?

A picture of an anthropomorphic sheep, their eyes squinted small as they say "Heh.". They look EXTREMELY smug, a not-too-large grin with the corners of their mouth turned up quite high.

Don’t threaten me with a good time, now.


memset_explicit is memset_s from Annex Okay, with out the Annex Okay historical past/baggage. It serves functionally the identical objective, too. It took lots (maybe an excessive amount of) dialogue, however Miguel Ojeda pursued all of it the best way to the top. So, now we’ve a typical, mandated, always-present memset_explicit that can be utilized in security-sensitive contexts, supplied your compiler and normal library implementers work collectively to not Be Evil™.

Hoorah! ????


The writing has been on the wall for properly over a decade now; intmax_t and uintmax_t have been insufficient for all the business over and has been consistently limiting the evolution of C’s integer types year over year, and affecting downstream languages. Whereas we can’t exempt each single integer sort from the trimmings of intmax_t and uintmax_t, we are able to a minimum of bless the intN_t varieties and uintN_t varieties to allow them to transcend what the 2 max varieties deal with. There may be lively work on this space to permit us to transition to a greater ABI and let these two varieties stay as much as their guarantees, however for now the least we may do is let the vector extensions and prolonged compiler modes for uint128_t, uint256_t, uint612_t, and so on. and so on. all get a while within the solar and out of the (u)intmax_t shadow.

This doesn’t assist for the preprocessor, although, since you’re nonetheless caught with the utmost worth that intmax_t and uintmax_t can deal with. Integer literals and expressions will nonetheless be caught coping with this downside, however on the very least there ought to be some small quantity of portability between the Beefy Machines™ and the presence of the newer UINT128_WIDTH and such macros.

Not the very best we are able to do, however progress in the fitting route! ????

Word that I didn’t say “that’s it”: there’s fairly a couple of extra options that made it in, simply my fingers are drained and there’s plenty of papers that have been accepted. I additionally don’t really feel like there are some I can do nice justice with, and fairly frankly the papers themselves make higher explanations than I do. Notably, N2956 – unsequenced functions is a extremely fascinating paper that may allow some intense optimizations with person attribute markup. Its efficiency improves can be utilized domestically:

#embody <math.h>
#embody <fenv.h>

inline double distance (double const x[static 2]) [[reproducible]] {
	// We assert that sqrt is not going to be known as with invalid arguments
	// and the end result solely will depend on the argument worth.
	extern typeof(sqrt) [[unsequenced]] sqrt;
	return sqrt(x[0]*x[0] + x[1]*x[1]);

I’ll depart the paper to elucidate how precisely that’s imagined to work, although! On high of that, we additionally removed Trigraphs??! (N2940) from C, and we made it so the _BitInt feature can be utilized with bit fields (N2969, nice). (Should you don’t know what Trigraphs are, contemplate your self blessed.)

One other actually consequential paper is the Tag Compatibility paper by Martin Uecker, N3037. It makes for outlining generic knowledge buildings by way of macros lots simpler, and doesn’t require a pre-declaration in an effort to use it properly. Lots of people have been thrilled about this one and picked up on the improvement immediately: it helps us get one step nearer to possibly having room to start out transport some cool container libraries sooner or later. You need to be looking out for when compilers implement this, and rush off to the races to start out growing nicer generic container libraries for C along side all the brand new options we put in!

There may be additionally plenty of performance that didn’t make it, reminiscent of Unicode Features, defer, Lambdas/Blocks/Nested Functions, huge perform pointers, constexpr features, the byteswap and different low-level bit performance I spoke of earlier than, assertion expressions, further macro performance, break break (or one thing prefer it), size_t literals, __supports_literal, Transparent Aliases, and extra.

However For Now?

My work is completed. I’ve received to go take a break and chill out. You’ll find the most recent draft copy of the Committee Draft Customary N3047 here. It’s in all probability stuffed with typos and different errors; I’m not an awesome undertaking editor, truthfully, however I do attempt, and I assume that’s all I can do for all of us. That’s it for me and C for the entire yr. So now,

A transformation sequence. An anthropomorphic schoolgirl sheep with lavender-purple hair runs with toast a piece of toast in their mouth, looking rushed and hurried, wearing a big pink bow over their chest with the typical white blouse over a knee-length blue skirt. Their image fades and drifts off to the right, slowly turning turn before they end up fully transformed into what looks like a very realistic sheep, whose floof is the same color as the lavender-purple hair from before.

it’s sleepy time. Nighty night time, and thanks for approaching this wild trip with me ????.

Source Link

What's Your Reaction?
In Love
Not Sure
View Comments (0)

Leave a Reply

Your email address will not be published.

2022 Blinking Robots.
WordPress by Doejo

Scroll To Top