Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Zig Is Self-Hosted Now, What's Next?

kristoff.it

Post reply on HN
Print view View original

31–40 of 159 posts

Re: Zig Is Self-Hosted Now, What's Next?

#31

Post by kristoff_it »

mattgreenrocks wrote: post #24
j-pb wrote: post #12

Earlier quoted context omitted.

Yeah, theres a surprising amount of stuff that's buggy, think `if` or `recursive structs with pointers` level stuff.

Can you substantiate this further? I don't see how basic stuff like that could be subtly buggy, the Zig devs are very qualified IMO.

We do have a good number of known miscompilations, I can substantiate for the parent poster with a simple link: https://github.com/ziglang/zig/labels/miscompilation

It's also true that the self-hosted compiler has caused new bugs to pop up, which in turn made us write new behavior tests. To me that's to be expected with every project aimed at rewriting an existing system.

It's also true that proper testing infrastructure is critical, and we do have it, although the requirement of more than 8gb of ram has not played in our favor in the past. We plan to revamp our CI systems once we are done with this release.

That said, it's a v0 language. If you choose to use it, you know you're coming with us on a ride. The choice of how much to prioritize correctness vs progress is influenced by many things that are going on in the project and there is no one clear strategy that trumps all others.

Quoting j-pb from another comment:

> A language that (does or doesn't) compile your code in 0.1 seconds to a broken binary is worse than piping your code into dev/null.

Eh, yes, but also no. Obviously a compiler needs to work correctly to be useful, but our priority right now is to move the project forward using our best judgment, not ensure that it has maximal utility to early adopters. I do agree with the sentiment that the compiler needs to be reliable, it's just that Zig is v0 for a reason.

In concrete terms I think that once we're done with an initial adjustment phase, you will find that the compiler will have a steady increase in stability over time. That's how it was with the bootstrap compiler before we started the self-hosted work.

Also, quoting dist1 from up the comment tree:

> I'm not so familiar with Zig, are you saying that the compiler fails to compile correct (i.e. according to spec) code?

Keeping in mind everything that I just wrote, note that Zig doesn't have a spec yet. Sometimes, some issues in the compiler are the result of discovering that a design choice needs to be amended.

Recent example of that: https://github.com/ziglang/zig/issues/12251

Re: Zig Is Self-Hosted Now, What's Next?

#32

Post by kristoff_it »

generalizations wrote: post #28
j-pb wrote: post #15

Earlier quoted context omitted.

Your quoted sentence talks about bugs in the old C++ compiler, I'm talking about bugs in both compilers. Pretty much all of the talks and blog posts on the `in-Zig` compiler are about better performance, lower memory footprint, e.t.c. I would much rather have heard a talk about how they improved correctness through automatic testing, formal methods, simple language design e.t.c. A language that (does or doesn't) comp…

I see the acknowledgements of the bugs in the old compiler, but what bugs in the new compiler are you talking about? Also, a recurring theme in zig has been the advantages of their simple language design.

See my reply in another comment tree https://news.ycombinator.com/item?id=33333827

Re: Zig Is Self-Hosted Now, What's Next?

#33

Post by setr »

thadt wrote: post #30
j-pb wrote: post #3

Correctness please. Zig is a way too simple language to justify all the really easy to hit bugs. I'd give everything for a zig compiler that's 10 times as slow but actually compiles my code. Also testing infrastructure. Zig really needs fuzzing and property testing tooling build-in. Hitting stage 2 with arbitrary autogenerated zig code and IR, would have really helped detect regressions for example.

Yep, I've also run into a myriad of gotchas while building things or poking around in written code. Just last week I was trying to build someone's year old repo that blew up on nightly, yet compiled on 0.9 just fine. But - it's plastered all over the entire project that it is a work in progress. This post itself is talking about how for loops are getting broken. It's an ambitious project with a talented and ambitious…

> Just last week I was trying to build someone's year old repo that blew up on nightly, yet compiled on 0.9 just fine.

Isn’t that the point of nightly

Re: Zig Is Self-Hosted Now, What's Next?

#36

Post by AndyKelley »

j-pb wrote: post #3

Correctness please. Zig is a way too simple language to justify all the really easy to hit bugs. I'd give everything for a zig compiler that's 10 times as slow but actually compiles my code. Also testing infrastructure. Zig really needs fuzzing and property testing tooling build-in. Hitting stage 2 with arbitrary autogenerated zig code and IR, would have really helped detect regressions for example.

If bugs are your concern, then this upcoming release is good news for you. Total number of open bug reports is 1,124. However if we remove bugs labeled "stage1" - that is - bugs that are no longer applicable because this release changes the default compiler to the self-hosted one - then the number drops to 514. It would be a fair characterization to say that this release cuts the number of known bugs in half. We have those stage1 bug reports still open because I mandated that every one must have test coverage before being closed, even though they are already fixed in master branch.

That said, I would humbly ask for you to trust my leadership on finding the balance on how much effort to spend on bug fixing versus how much to spend on progress. I think you underestimate how much is going to change before Zig tags 1.0. There are still a few experimental and exploratory changes planned, and when software specifications change, bugs are inevitably introduced. Spending the time to iron out 100% of the bugs found from fuzz testing, on a feature that ends up replaced with a different one that also needs to get debugged, is a waste of time, and a disservice to the wider user base of the project. We're better off finding the "sweet spot" for a swift journey, arriving at our destination early, and then busting out the hardcore QA.

I see https://github.com/ziglang/zig/issues/13211 mentioned elsewhere in this thread. I want to point out that getting the C ABI correct when using LLVM is a notoriously perilous process, as any other compiler dev reading HN today can attest, and this regression which lasted for a total of 6 days in master branch without ever gracing a tagged release, is a symptom of the fact that we actually addressed this C ABI problem at its core, added more test coverage, added an abstraction in the LLVM code generation, and fixed the C ABI for several targets.

And finally, I agree that fuzz testing should be integrated into Zig's unit testing system. That would be a nice enhancement.

Re: Zig Is Self-Hosted Now, What's Next?

#37

Post by thadt »

setr wrote: post #33
thadt wrote: post #30

Earlier quoted context omitted.

Yep, I've also run into a myriad of gotchas while building things or poking around in written code. Just last week I was trying to build someone's year old repo that blew up on nightly, yet compiled on 0.9 just fine. But - it's plastered all over the entire project that it is a work in progress. This post itself is talking about how for loops are getting broken. It's an ambitious project with a talented and ambitious…

> Just last week I was trying to build someone's year old repo that blew up on nightly, yet compiled on 0.9 just fine. Isn’t that the point of nightly

Sure - but I believe it was due to a change in a language detail. I was just commenting that its not intended to be a particularly stable target quite yet.

Re: Zig Is Self-Hosted Now, What's Next?

#38

Post by ghoward »

david2ndaccount wrote: post #34

I wonder if Zig is getting too much hype at the moment considering its immaturity (and so many easy to hit bugs as noted by other commenters). Is it getting hype before it's ready? Will it be able to recover from that?

Disclaimer: Not a Zig fan.

I think Zig is getting too much hype. It's unfortunate because Zig could have potential, but hyping something before it's ready is a recipe for alienating early adopters and killing progress towards popularity.

The worst part of it is that I've heard people say that they've got Zig in production; with all of the bugs in the compiler, they are apt to get burned, and too many burns will lead to putting out the fire and rewriting.

It doesn't help that (in my opinion) there is confusion regarding basic things about Zig. [1]

[1]: https://gavinhoward.com/2022/04/i-believe-zig-has-function-c...

Re: Zig Is Self-Hosted Now, What's Next?

#39

Post by kristoff_it »

david2ndaccount wrote: post #34

I wonder if Zig is getting too much hype at the moment considering its immaturity (and so many easy to hit bugs as noted by other commenters). Is it getting hype before it's ready? Will it be able to recover from that?

> Is it getting hype before it's ready? Will it be able to recover from that?

Isn't that the very definition of hype? :^)

We can't hide in a cave to build everything in secret until it's ready because otherwise we wouldn't have the money to pay contributors, or we would be forced to get funding through other means (I mention that point in the blog post). And while we do want to be in the open, you can see that in my writing I never talk about a future that's distant and completely unproven. Even the most forward-looking stuff I mention, like incremental compilation, has at least had a proof of concept to show that it's not pure fiction.

I'm obviously biased, but I think we're in a pretty good spot when it comes to gathering public interest vs getting things done.

Re: Zig Is Self-Hosted Now, What's Next?

#40

Post by thefaux »

Zig seems to have captured a lot of attention in the new language design space, and I personally find this unfortunate. Frankly, I find it irresponsible that the team has marketed a pre 1.0 language so heavily. By constantly evolving the language and making frequent breaking changes, they are quite literally wasting people's time. They also have been very hostile to people who have raised fair points about the misleading claims that they have made.

From my perspective, the entire premise of the language is flawed. The main project goal seems to be to become a replacement for c. This means that in some sense the project is implicitly defined by c and doesn't really stand on its own conceptually. This could be ok if it actually fixed some of the worst problems with c, but most of what it offers appears to be relatively superficial syntactic improvements (which could easily be implemented by a simple transpiler to c). While I am sure that the language designers see things differently, it appears that Zig just lets c programmers keep writing c with some additional compiler checks. This is probably an improvement on c (at least in some respects), but it seems largely incremental. Also because they prematurely evangelized the language, they are likely to be weighed down by the additional complexity that seeped in from trying to support so many different use cases before they had really figured out what the language is.

That it took two years to write the self hosting compiler speaks to the fact that this language is already quite complex. My guess is that this complexity is always going to pose problems. Downstream users are always going to be finding obscure edge cases that don't work because the language supports so many different things that actually testing everything is going to be extremely difficult.

Having said all that, if you love Zig, that's great. I don't doubt that it offers some real quality of life improvements for you. I just personally believe that this project is an evolutionary dead end and that it is very unlikely to have significant mindshare/deployment in say, 10-20 years.

Post reply on HN