I always thought that C was a stepping stone to learn other languages. Like Pascal, it was educational to learn. My Comp Sci courses in 1986-1990 used Turbo Pascal and Turbo C.
None of the alternatives have stability. What was exemplary & idiomatic rust pre-pandemic would be churlish & rejected now and the toolchain use would be different anyway.
Carpenters, plumbers, masons & electricians work on houses 3-300 yrs old, navigate the range of legacy styles & tech they encounter, and predictably get good outcomes.
Only C has, yet, given use that level of serviceability. C99, baby, why pay more?
When there’s an alternative that can compete with that sort of real-world durability, C will get displaced.
It’s weird how whiny this post is. Like there’s zero intellectual curiosity about why C got this way, and why C gets to be the foundation for how systems software is written.
I could write a whole essay about why, but now isn’t the time. I’m just going to enjoy the fact that TFA and the author don’t get it.
This is the same author who found and reported ABI incompatibilities between clang and gcc: https://faultlore.com/abi-cafe/book/trophies.html — Given these incompatibilities have since been addressed, I think it's fair to say there were indeed issues with C.
I really don't understand why people keep misunderstanding this post so badly. It's not a complaint about C as a programming language. It's a complaint that, due to so much infrastructure being implemented in C, anyone who wants to interact with that infrastructure is forced to deal with some of the constraints of C. C has moved beyond merely being a programming language and become the most common interface for in-process interoperability between languages[1], and that means everyone working at that level needs to care about C even if they have no intention of writing C.
It's understandable how we got here, but it's an entirely legitimate question - could things be better if we had an explicitly designed interoperability interface? Given my experiences with cgo, I'd be pretty solidly on the "Fuck yes" side of things.
(Of course, any such interface would probably end up being designed by committee and end up embodying chunks of ALGOL ABI or something instead, so this may not be the worst possible world but that doesn't mean we have to like it)
[1] I absolutely buy the argument that HTTP probably wins out for out of process
Of course things could be better. That doesn’t mean that we can just ignore the constraints imposed by the existing software landscape.
It’s not just C. There are a lot of things that could be substantially better in an OS than Linux, for example, or in client-server software and UI frameworks than the web stack. It nevertheless is quite unrealistic to ditch Linux or the web stack for something else. You have to work with what’s there.
I mean… an ABI is more like an agreement. It isn’t a specification. It’d be nice if everything was neatly specified and sound. But as the author notes near the end… there’s a lot of platforms and they all have their own quirks.
There has to be an ABI that has to be agreed upon by everyone. Otherwise there wouldn’t be any interoperability. And if we didn’t have the SystemV ABI — what would we use instead? Prepare for a long debate as every language author, operating system designer, and platform under the sun argues for their respective demands and proposals. And as sure as the sun rises in the East someone, somewhere, would write an article such as this one decrying that blessed ABI.
SystemV shouldn’t be the be all and end all, IMO. But progress should be incremental. Because a lingua franca loses its primary feature and utility when we all return to our own fiefdoms and stop talking to one another in the common tongue.
It’s a pain in the metaphorical butt. But it’s better, IMO, than the alternatives. It’s kind of neat that SystemV works so well let alone at all.
We didn't do it to annoy you or to foist bad APIs on you. We did it because it was the best language for writing machine code at the time. By miles. Not understanding why this is true will lead you to make all the same mistakes the languages "bested" by C made.
The trouble with C as an API format is that there's no size info. That's asking for buffer overflows.
There's an argument for full type info at an API, but that gets complicated across languages. Things that do that degenerate into CORBA. Size info, though, is meaningful at the machine level, and ought to be there.
Apple originally had Pascal APIs for the Mac, which did carry along size info. But they caved and went with C APIs.
> Anyone who spends much time trying to parse C(++) headers very quickly says “ah, actually, fuck that” and asks a C(++) compiler to do it.
That's exactly my case. For my programming language I have wrote a tool for C headers conversion using libclang. And even with help of this library it wasn't that easy, I have found a lot of caveats by trying converting headers like <windows.h>.
- unspecified default type sizes. Should have had i8, u16, i32, u64, f32, f64 from the beginning.
- aliasing pointers being restricted by default (ie an alias keyword should have been added). Performance matters. All these benchmarks which show something beating C or C++ are mostly due to dealing with aliasing pointers. C++26 still doesnt have standardised restrict keyword.
There are more but I understand the logic/usability/history behind them. The above points should have been addressed in the 80’s.
> - unspecified default type sizes. Should have had i8, u16, i32, u64, f32, f64 from the beginning.
I strongly disagree. The programmer should rather prescribe intent and shouldn't constantly think about what size this should exactly have. Does this variable represent the size of an object, an address, a difference or just a random positive integer? Then use size_t, uintptr_t, ptrdiff_t and unsigned int respectively. Why should I care what exact sizes these are? I hate that modern (system) languages completely throw away that concept. "I want a unsigned integer" "oh, you mean u32" "No! I really mean an unsigned integer."
Also when I do want e.g. 32 bit available, there is no reason, I need to use a suboptimal 32 bit wrapping behaviour when I don't need it. The correct type to use for computation for that would be uint32_fast_t and the implementation chooses what makes sense to use for this.
Which systems in the 70s or even 90s would have had good hardware support for all these lengths?
I think overall perhaps its better that the thing named int works on both the arduino and my amd64 otherwise we might needlessly need to rename all types for each platform to fit the most natural type lengths there. Imagine the 32 bit to 64 bit transition under those conditions.
Do someone pays for anti-C propaganda ?? All that logic breaking accusations...
Eg. here, from memory:
> ...you want to read 32 bits from file but OH NOOES long is 64 bit ! The language ! The imposibility !
But when you read something ot unserialize some format you just need to know based on format schema or domain knowledge. Simple and straightforward like that ! You do not do some "reflections" on what language standard provide and then expect someone send you just that !!
So that anti-C "movement" is mostly based on brainless exampless.
Not saying C is perfect.
But it is very good and I bet IBM and other big corps will keep selling things written and actively developed in C/C++ + adding hefty consulting fees.
In the meantime proles has been adviced to move to cpu-cycle-eating inferior languages and layers over layers of cycle burning infra in cloud-level zero-privacy and guaranteed data leaks.
Oh, btw. that femous Java "bean" is just object with usually language delivered "basic type"... How that poor programmer from article should know what to read from disc when he just have types Java provides ?? How ? Or maybe he should use some domain knowledge or schema for problem he is trying to solve ??
And in "scripting language" with automatic int's - how to even know how many bits runtime/vm actually use ? Maybe some reflection to check type ? But again how that even helps if there is no knowledge in brain how many bits should be read ?? But calling some cycle burning reflection or virtual and as much as posible indirect things is what fat tigers love the moust :)
C isn't really the protocol though, its just a way every language has of exporting simple symbols. Otherwise how else does this work, you'd never every language to understand every other languages symbol name encoding scheme, some of which are complete jibberish (I'm looking hard your way C++).
The real protocol in action here is symbolic linking and hardware call ABIs.
You could always directly call Rust functions, but you'd have to know where to symbolically look for them and how to craft its parameters for example.
If this is well defined then its possible. If its poorly defined or implementation specific (c++) then yeah its a shit show that is not solvable.
I remember reading some crank on slashdot decades ago seriously claiming that Windows was not an operating system. This type of argument hasn’t gotten any more intelligent since.
Following up, I see that's not really what the author was trying to say: it's that for a growing number of purposes, like gluing together two languages that aren't C, that the C ABI is still dominating things where the C language shouldn't even be involved.
Write clickbait, get kneejerk. I guess some other things are at least as old as C.
OP, I would strongly recommend you try to play devil's advocate against your own case before making it, so you have a chance to refine your thoughts and rebut at least the top 2-3 counterarguments. More than that, you would do well to understand the concept of Chesterton's Fence as it applies to engineering:
"In the matter of reforming things, as distinct from deforming them, there is one plain and simple principle; a principle which will probably be called a paradox. There exists in such a case a certain institution or law; let us say, for the sake of simplicity, a fence or gate erected across a road. The more modern type of reformer goes gaily up to it and says, "I don't see the use of this; let us clear it away." To which the more intelligent type of reformer will do well to answer: "If you don't see the use of it, I certainly won't let you clear it away. Go away and think. Then, when you can come back and tell me that you do see the use of it, I may allow you to destroy it.""
-G.K. Chesterton
#1: What you consider a "language" is one with its own official runtime. By your logic, Javascript is not a language either because the interpreter is usually written in C++ (e.g. V8, spidermonkey).
#2: Dogfooding a compiler and a runtime is common practice because most people believe that it will help the makers of a language identify bugs in their implementation or identify undefined behavior in the very definition of their language. However, every language must bootstrap from another language: most languages bootstrapped from a compiler written in C. If you trace the lineage from compiler to compiler to ... compiler, you'll most likely find that the second C compiler ever written was written in C and the first compiler was written in some architecture's assembly language. The same still applies for the new, new language of C99 which is less than half the age of C. C is so mature that rooting out new undefined behaviors is no longer a significant concern.
#3: libc is not just a "language runtime", it's the interface into the OS kernel for most operating systems. The runtime for every "real language" as you describe it ultimately uses libc. It doesn't make sense to separate the functions in the C99 standard from the OS's own syscalls and special-purpose user land functions call C standard library functions, and C's standard library relies on the OS's special functions. If they were separate libraries, there would be a circular dependency.
Think of C not as "not a language" but as a language that serves the unusual purpose of being the foundation for nearly everything else.
You have a point on integer types being a mess, but we have stdint.h for that.
30 comments
[ 2.9 ms ] story [ 61.0 ms ] threadCarpenters, plumbers, masons & electricians work on houses 3-300 yrs old, navigate the range of legacy styles & tech they encounter, and predictably get good outcomes.
Only C has, yet, given use that level of serviceability. C99, baby, why pay more?
When there’s an alternative that can compete with that sort of real-world durability, C will get displaced.
I could write a whole essay about why, but now isn’t the time. I’m just going to enjoy the fact that TFA and the author don’t get it.
It's understandable how we got here, but it's an entirely legitimate question - could things be better if we had an explicitly designed interoperability interface? Given my experiences with cgo, I'd be pretty solidly on the "Fuck yes" side of things.
(Of course, any such interface would probably end up being designed by committee and end up embodying chunks of ALGOL ABI or something instead, so this may not be the worst possible world but that doesn't mean we have to like it)
[1] I absolutely buy the argument that HTTP probably wins out for out of process
It’s not just C. There are a lot of things that could be substantially better in an OS than Linux, for example, or in client-server software and UI frameworks than the web stack. It nevertheless is quite unrealistic to ditch Linux or the web stack for something else. You have to work with what’s there.
There has to be an ABI that has to be agreed upon by everyone. Otherwise there wouldn’t be any interoperability. And if we didn’t have the SystemV ABI — what would we use instead? Prepare for a long debate as every language author, operating system designer, and platform under the sun argues for their respective demands and proposals. And as sure as the sun rises in the East someone, somewhere, would write an article such as this one decrying that blessed ABI.
SystemV shouldn’t be the be all and end all, IMO. But progress should be incremental. Because a lingua franca loses its primary feature and utility when we all return to our own fiefdoms and stop talking to one another in the common tongue.
It’s a pain in the metaphorical butt. But it’s better, IMO, than the alternatives. It’s kind of neat that SystemV works so well let alone at all.
We didn't do it to annoy you or to foist bad APIs on you. We did it because it was the best language for writing machine code at the time. By miles. Not understanding why this is true will lead you to make all the same mistakes the languages "bested" by C made.
The whole world shouldn't "need to be fixed" because you won't spend the time to learn something.
Rust doesn't even have a stable Internal ABI that's why you have to re-compile everything all the time.
There's an argument for full type info at an API, but that gets complicated across languages. Things that do that degenerate into CORBA. Size info, though, is meaningful at the machine level, and ought to be there.
Apple originally had Pascal APIs for the Mac, which did carry along size info. But they caved and went with C APIs.
That's exactly my case. For my programming language I have wrote a tool for C headers conversion using libclang. And even with help of this library it wasn't that easy, I have found a lot of caveats by trying converting headers like <windows.h>.
- unspecified default type sizes. Should have had i8, u16, i32, u64, f32, f64 from the beginning.
- aliasing pointers being restricted by default (ie an alias keyword should have been added). Performance matters. All these benchmarks which show something beating C or C++ are mostly due to dealing with aliasing pointers. C++26 still doesnt have standardised restrict keyword.
There are more but I understand the logic/usability/history behind them. The above points should have been addressed in the 80’s.
I strongly disagree. The programmer should rather prescribe intent and shouldn't constantly think about what size this should exactly have. Does this variable represent the size of an object, an address, a difference or just a random positive integer? Then use size_t, uintptr_t, ptrdiff_t and unsigned int respectively. Why should I care what exact sizes these are? I hate that modern (system) languages completely throw away that concept. "I want a unsigned integer" "oh, you mean u32" "No! I really mean an unsigned integer."
Also when I do want e.g. 32 bit available, there is no reason, I need to use a suboptimal 32 bit wrapping behaviour when I don't need it. The correct type to use for computation for that would be uint32_fast_t and the implementation chooses what makes sense to use for this.
Eg. here, from memory:
> ...you want to read 32 bits from file but OH NOOES long is 64 bit ! The language ! The imposibility !
But when you read something ot unserialize some format you just need to know based on format schema or domain knowledge. Simple and straightforward like that ! You do not do some "reflections" on what language standard provide and then expect someone send you just that !!
So that anti-C "movement" is mostly based on brainless exampless.
Not saying C is perfect.
But it is very good and I bet IBM and other big corps will keep selling things written and actively developed in C/C++ + adding hefty consulting fees.
In the meantime proles has been adviced to move to cpu-cycle-eating inferior languages and layers over layers of cycle burning infra in cloud-level zero-privacy and guaranteed data leaks.
Oh, btw. that femous Java "bean" is just object with usually language delivered "basic type"... How that poor programmer from article should know what to read from disc when he just have types Java provides ?? How ? Or maybe he should use some domain knowledge or schema for problem he is trying to solve ??
And in "scripting language" with automatic int's - how to even know how many bits runtime/vm actually use ? Maybe some reflection to check type ? But again how that even helps if there is no knowledge in brain how many bits should be read ?? But calling some cycle burning reflection or virtual and as much as posible indirect things is what fat tigers love the moust :)
With C++ it's the same. Within the Haiku code it's half understandable, the whole spec it's to get driven mad in days.
The real protocol in action here is symbolic linking and hardware call ABIs.
You could always directly call Rust functions, but you'd have to know where to symbolically look for them and how to craft its parameters for example.
If this is well defined then its possible. If its poorly defined or implementation specific (c++) then yeah its a shit show that is not solvable.
Write clickbait, get kneejerk. I guess some other things are at least as old as C.
Good read though. Thinking about C as not just a language but also a protocol is a different perspective that is useful for the mental model.
"In the matter of reforming things, as distinct from deforming them, there is one plain and simple principle; a principle which will probably be called a paradox. There exists in such a case a certain institution or law; let us say, for the sake of simplicity, a fence or gate erected across a road. The more modern type of reformer goes gaily up to it and says, "I don't see the use of this; let us clear it away." To which the more intelligent type of reformer will do well to answer: "If you don't see the use of it, I certainly won't let you clear it away. Go away and think. Then, when you can come back and tell me that you do see the use of it, I may allow you to destroy it.""
-G.K. Chesterton
#1: What you consider a "language" is one with its own official runtime. By your logic, Javascript is not a language either because the interpreter is usually written in C++ (e.g. V8, spidermonkey).
#2: Dogfooding a compiler and a runtime is common practice because most people believe that it will help the makers of a language identify bugs in their implementation or identify undefined behavior in the very definition of their language. However, every language must bootstrap from another language: most languages bootstrapped from a compiler written in C. If you trace the lineage from compiler to compiler to ... compiler, you'll most likely find that the second C compiler ever written was written in C and the first compiler was written in some architecture's assembly language. The same still applies for the new, new language of C99 which is less than half the age of C. C is so mature that rooting out new undefined behaviors is no longer a significant concern.
#3: libc is not just a "language runtime", it's the interface into the OS kernel for most operating systems. The runtime for every "real language" as you describe it ultimately uses libc. It doesn't make sense to separate the functions in the C99 standard from the OS's own syscalls and special-purpose user land functions call C standard library functions, and C's standard library relies on the OS's special functions. If they were separate libraries, there would be a circular dependency.
Think of C not as "not a language" but as a language that serves the unusual purpose of being the foundation for nearly everything else.
You have a point on integer types being a mess, but we have stdint.h for that.
But Rust evangelists cannot stop people from using any particular language
Assembly is the only language that matters. It's the only language the computer "understands"
Everything other language is just an abstraction over it
When it comes to abstractions not everyone has the same preferences