Xr0 verifier, guarantee the safety of C programs at compile time

by Alifatisk • 7 months ago

114 67 7 months ago

pkhuong• 7 months ago

I don't see any explanation of what niche this targets relative to pre-existing tools like Checked C, CMBC, or Frama C...

aw1621107• 7 months ago

Previously:

Show HN: Xr0 – Vanilla C Made Safe with Annotations (https://news.ycombinator.com/item?id=37536186, 2023-09-16, 13 comments)

Xr0 Makes C Safer than Rust (https://news.ycombinator.com/item?id=39858240, 2024-03-28, 41 comments)

Xr0: C but Safe (https://news.ycombinator.com/item?id=39936291, 2024-04-04, 144 comments)

Show HN: Xr0 is a Static Debugger for C (https://news.ycombinator.com/item?id=40472051, 2024-05-05, 4 comments)

tialaramex• 7 months ago

Thanks! I thought this felt familiar. 2023 is a long time to still be in this "Initial signs are promising" mode. There are no code commits since March 2025.

I've had projects which stalled for a few months or even a year, but generally if I said I'll get to it "soon" and two years later I haven't, that's not getting done. There are two boxes of books I planned to unbox "soon" after discovering that the bookshelves for my new flat were full & so I had nowhere to put them. That was when Obama was still President of the United States of America. Don't expect me to ever get around to unboxing those books, I clearly haven't even missed them.

muldvarp2• 7 months ago

The initial sign for Xr0 never seemed promising for anyone with experience in formal verification. Neither the code nor the ideas they presented were new. I asked them multiple times to clarify how their project differed from the dozens of already existing options for formal verfication of C programs and never got a concrete answer.

rurban• 7 months ago

Well, apparently someone rewrote it in rust, which was better than their version. Which they needed some time to process how to continue.

And they got stuck with the bounds checker and loops.

But other such checkers are far more advanced, with a better contract syntax.

thealistra4• 7 months ago

What happens if a function allocates not deterministically, like

if (turing_machine_halts(tm)) return malloc(1); else return NULL;

How is this handled?

jonhohle• 7 months ago

NULL is a valid return for malloc. Wouldn’t that case already be handled?

dnautics5• 7 months ago

as someone building an analyzer for zig, if you sent something like this through the system im building, it would get handled by Zig's optional type tagging; but lets say you did "free" (so your result is half freed, half not freed): it would get rejected because both branches produce inconsistent type refinements, you don't have to solve the halting problem, just analyze all possible branches and seek convergence.

dwattttt• 7 months ago

Rust is the poster child for these complaints, but this is a great example of "the language rejects a valid program". Not all things that can be expressed in C are good ideas!

This is "valid" C, but I wholly support checking tools that reject it.

SkiFire13• 7 months ago

just analyze all possible branches and seek convergence.

This sounds like a very simple form of abstract interpretation, how do you handle the issues it generally brings?

For example if after one branch you don't converge, but after two you do, do you accept that? What if this requires remembering the relationship between variables? How do you represent these relationships?

Historically this has been a tradeoff between representing the state space with high or perfect precision, which however can require an exponential amount of memory/calculations, or approximate them with an abstract domain, which however tend to lose precision after performing certain operations.

Add dynamically sized values (e.g. arrays) and loops/recursion and now you also need to simulate a possibly unbounded number of iterations.

RossBencina2• 7 months ago

In general, symbolic execution will consider all code paths. If it can't (or doesn't want to) prove that the condition is always true or false it will check that the code is correct in two cases: (1) true branch taken, (2) false branch taken.

thealistra4• 7 months ago

I understand how this works in general. I had static analyzers at Uni, I know lattice theory and all this - I am just wondering how Xr0 handles it.

tgv3• 7 months ago

I think all paths have to return the same allocation. You would have to solve this in another way.

thealistra4• 7 months ago

Isn’t this a very restrictive way to write? Hard for me to imagine as never wrote with such annotations so no idea how viable it is for a large codebase to have this constraint.

nextaccountic• 7 months ago

There will always be valid programs that are nonetheless rejected by some verifier (Rice's theorem). That is, programs that have really nothing wrong but nonetheless are rejected as invalid

In those cases you generally try to rewrite it in another way

jokoon• 7 months ago

Does it require annotations or can it validate any c code?

It's odd that so many people promote rust, yet we don't even use static analysis and validators for c or C++.

How about enforcing coding standards automatically first, before switching to a new language?

RossBencina2• 7 months ago

we don't even use static analysis and validators for c or C++

There is some use, how much I don't know. I guess it should be established best practice by now. Also run test suites with valgrind.

Historically many of the C/C++ static analyzers were proprietary. I haven't checked lately but I think Coverity was (is?) free for open source projects.

haileys• 7 months ago

Not sure what this post has to do with Rust, but people do use static analysis on C and C++. The problem is that C and C++ are so flexible that retrofitting static verification after the fact becomes quite difficult.

Rust restricts the shape of program you are able to write so that it's possible to statically guarantee memory safety.

Does it require annotations or can it validate any c code?

If you had clicked through you would see that it requires annotations.

01HNNWZ0MV43FF• 7 months ago

Today's experts are the novices of 40 years ago, and today's novices will be the experts in 40 years.

C and C++ don't require static analysis, and it's difficult to set up, and so most of us slide down the incentive gradient of using C / C++ without any helpers except CMake and gdb.

Rust requires it, so the noobies use it, so in 40 years the experts will accept it.

ranger_danger• 7 months ago

it's difficult to set up

Is it though? I've only ever had to run "scan-build make" for my projects and it spits out a full folder of HTML pages that details any static analysis issues, and I didn't have to change my build system at all.

WalterBright13• 7 months ago

From the tutorial:

    #include <stdio.h>
    int main()
    {
        int i;
        return i;
    }
    the behaviour is undefined because i’s value is indeterminate.

D solves that problem by initializing variables with the default initializer (in this case 0) if there is not an explicit initializer.

That was in the first version of D because I have spent days tracking down an erratic bug that turned out to be an uninitialized variable.

WalterBright13• 7 months ago

BTW, when D compiles C code, it will default initialize the C variables, too.

dadoum• 7 months ago

I think that the right thing to do is to error out though. When the behaviour of some code cannot be guaranteed, that code should just be ruled out imo. Manually initializing a variable generally doesn't clutter the code, arguably it's making it clearer.

Alifatisk• 7 months ago

Dart does this, you can mark a variable as "late" which tells the compiler that you know for certain the variable will be written too before read. If something reads the variable before it's initialized, then the runtime will error. Maybe even on compile time if it can be caught, I am not certain.

pseudohadamard3• 7 months ago

Pretty much any existing C compiler will also solve that problem by telling you that it's uninitialized.

WalterBright13• 7 months ago

I tried:

    int test()
    {
        int i;
        return i;
    }

using clang on my Mac mini, and:

    clang -c test.c

and it compiled without complaint.

rwmj2• 7 months ago

You can do that in GCC now (-ftrivial-auto-var-init=zero). There are a few projects that decided to enable this by default.

WalterBright13• 7 months ago

Should add it to the C standard.

Animats2• 7 months ago

Cute. They've introduced ownership to C. The annotation system is expressing ownership, in a rather clunky way. You have to explicitly declare that functions create or release allocated space. That's ownership tracking.

But "We currently model all buffers as infinitely-sized. We will be adding out-of-bounds checking soon." That's the hard problem.

LiamPowell• 7 months ago

But "We currently model all buffers as infinitely-sized. We will be adding out-of-bounds checking soon." That's the hard problem.

It's not particularly difficult for the prover. You essentially just need to do a translation from C in to your ATP language of choice, with a bunch of constraints that check undefined behaviour never occurs. Tools such as Frama-C and CBMC have supported this for a long time.

The difficult part is for the user as they need to add assertions all over the place to essentially keep the prover on the right track and to break down the program in to manageable sections. You also want a bunch of tooling to make things easier for the user, which is a problem that can be as difficult as you want it to be since there's always going to be one more pattern you can detect and add a custom handler/error message for.

Animats2• 7 months ago

It's not particularly difficult for the prover.

I know that. Did that decades ago. I wanted to see how they did the annotations and proofs. What can you express? Partially filled arrays are moderately hard. Sparsely initialized arrays need a lot of scaffolding. This is an area where most "safe C" variants have foundered.

rwmj2• 7 months ago

Oddly they didn't even reuse the annotation for this that GCC already has: https://stackoverflow.com/questions/18485447/gcc-attribute-m...

171862744010• 7 months ago

Cute. They've introduced ownership to C.

Ownership semantics are described in every serious C interface. Linters for checking it have also existed for decades. I find the notion that Rust invented it to be incredible stupid. Rust just has different ownership semantics and makes it an enforced part of the language (arguable a good idea). And yes they of course also do bounds-checking.

pjmlp• 7 months ago

Rust did not invent affine types of course, those that affirm that lack background in type systems.

Xr0 verifier, guarantee the safety of C programs at compile time

Related Stories