(I've used Go and Rust daily for the past few years. I love them both.)
Still, the fact that all those languages are in the top 10 of programming languages kind of says that people don't really care.
Go that is?
To be even clearer: I don't think Rust's value proposition depends on whether you absolutely must avoid GC or not.
It had GC as core part of language, about six or more years ago?
PS. Pozdrawiam rodaka.
If you don't know anything about rust, you shouldn't respond to a question about rust.
The leaders of Go foster an attitude of exclusiveness (just like a month ago they wanted to get rid of the Go subreddit in favor of a solely Google owned option of Google groups).
The leaders of Rust are very receptive and helpful to new people. They are on IRC / Reddit and many other channels.
I'd much rather invest my time into a truly open language and to me, that is not Go.
To be fair, this was a proposal by a single person on the Go mailing list, and it was in reaction to Reddit's CEO publicly admitting to editing other users' comments. The person who proposed closing the Go subreddit was also under the mistaken impression that the subreddit was hosted by the Go team, which wasn't the case. In the end, there was a lot of discussion, and nothing was deleted. Tempest in a teapot, as usual.
Go has a serious culture problem, but that's not a good example of it.
In my experience, there's a lot of hostility and arrogance. More than in any other community, I've seen Go developers shut down discussion by closing comment threads on Github; reject pull requests and refuse to debate the merits of the change; castigate people for "not following proper procedure"; and being dismissive or contemptuous instead of humble when they fail to understand a problem that is being discussed (on, say, the official Go Slack channel).
My very personal hypothesis is that this is a case of mirroring. The Go development team may be said to have a laconic, authoritarian style, which is, of course, their prerogative, and which befits their position; unfortunately, a lot of Go developers seem to be under the impression that they, too, can behave like demigods. In particular, Go developers, more than other cultures, seem to get an ego boost out of telling people "no".
But this is a generalization. There are lots of friendly Go devs around, to be sure (I hope I'm one of them). I'm particularly happy with the Kubernetes team. At the same time, I do think it's an issue and one that the community needs to be aware of.
This is very misleading. There is no official Go slack channel. The slack channel you are referring is just a slack channel and is by no means official.
> In particular, Go developers, more than other cultures, seem to get an ego boost out of telling people "no".
This is misleading as well. Go was designed from the start to exclude certain features e.g. inheritance and many others. There are usually very good reasons for those decisions that people who've been following Go from its creation know and understand very well.
Now what happens when someone who does not understand those design choices comes to Go? They want their favorite feature obviously! When Go members attempt to explain them why something like that does not exist in Go (and probably will never be included) it starts feeling like "no". And people do not take "no" well. They get emotional and fail to see reason. If they had bothered to explore the language or at least read the official documentation and FAQ, that state of mind might have been avoided.
Including everyone's favorite feature does not make a language better.
Go has created a community around a certain school of programming. Nobody claims that it is better than other schools but it is a fact that it exists. But who is the one that is close-minded in this case? The new person that doesn't bother with the teachings of the school or the school that dismisses the ideas of the new person? Who is really the one that says "no"?
From my experience, the Go developers always carefully consider every new idea that is brought to the table. But it is also a fact that after the 10th time you've seen the same idea, you are not going to sit down and spend time considering it. You are just gonna link a previous discussion and say "Sorry this has been brought up before, please check these". How does that feel to a new person? I bet it feels like "no" again.
I find it pretty terrible that Kubernetes develops on Slack/Github/Videochats. For being an Open Source project, it rejects open source tooling and makes it difficult for people with poor english skills to participate (text-based meetings are much better for inclusiveness). Even if you do listen in on the video meetings, they give you a feeling as an outsider that many of the decisions have already been made elsewhere (some obscure github issue conversation, a google doc comment on a 1 year old doc, etc). It seems the only way to meaningfully participate in the k8s community is to be a known insider... :(
Wow, really? Do you have a link to that discussion? That's wild.
It should be a good summary of the event.
In Go I can only write libraries for Go programs. In Rust I can write libraries for any program.
i.e. Rust can easily produce static and dynamic libraries that are linkable with C programs and any language with a C FFI. I can write Rust code that works for programmers using C, C++, C#, D, Go, Swift, Python, PHP, Java, etc.
http://www.darkcoding.net/software/building-shared-libraries...
That's not true. It's quite simple to create a loadable shared object in go and call it using anything with a c ffi.
[1] https://www.cockroachlabs.com/blog/the-cost-and-complexity-o...
As proof, notice that it barely happens, and only as an oddity in Go, yet in rust there are actual uses (e.g. ruby and python library optimization)
I was going to say that rust has performance advantages over Go (due to GC), but look at benchmarks:
http://benchmarksgame.alioth.debian.org/u64q/compare.php?lan...
Go wins some and looses some, but it's all in the ballpark (except Binary trees [1] which it loses even to Java(!)).
It's true that rust is a new language, but so is Go.
[1]: I assume it's because it's a test of GC, but Go loses to Java (which, like Go, is a GC language)
For example, many of the Java/C/C++ benchmarks are using custom optimizations that should be illegal for the benchmarking. Case in point, some are featuring custom hash maps that feature hashing algorithms that, while fast, would never be useful as they provide no protection against collisions. You'll see a hashing algorithm in a C preprocessor, for example, that just fakes having an actual algorithm whereas Rust examples are sticking to the tried and tested production-grade algorithms shipping in the standard library.
They are useful in this case, aren't they? Custom hash map implementation can be also useful in other scenarios. This is feature of the language that allow you to do that so it's not "illegal" to use it. If Rust doesn't allow you to do something while other language does, it doesn't mean it should be illegal.
I don't see any tricks in Go and it's faster than Rust and more memory efficient in almost half of the benchmarks. So what's your excuse for those cases ?
I don't speak Rust, so I can't quite tell what's going on here:
http://benchmarksgame.alioth.debian.org/u64q/program.php?tes...
But the C hash function is very simple, and probably not at all collision-resistant:
#define CUSTOM_HASH_FUNCTION(key) (khint32_t)((key) ^ (key)>>7)
I am sure that the Rust could be made to do the same thing, but I am not sure that, as-is, this is an apples to apples comparison.
Another factor could also be that the GC in the Go examples just never runs since it doesn't allocate enough memory. It's hard to say exactly what's happening without tracing the runtime, which would also affect the benchmarks a bit.
It's important to note that while both languages are natively compiled (and I suspect Rust would inch out in that category due to LLVM's backing) most of the overhead would probably be from memory, whether allocations, or cache usage, which makes comparing them in microbenchmarks a little inaccurate.
This is an allocator feature independent of garbage collection; jemalloc does this for example. GCd languages have the potential to do this better since they can do additional analysis whilst tracing for little additional cost, but non-GCd languages can still benefit from this.
That is, I think I'm thinking of this:
> k-nucleotide will explicitly require built-in / library HashMap.
https://alioth.debian.org/tracker/?func=detail&group_id=1008...
Since C doesn't have a standard library hashmap, you can write an entirely custom one just for the benchmark. But since Rust has a built-in hashmap in the standard library, you cannot. Even though you could write Rust code that's the same as the custom C hashmap.
http://benchmarksgame.alioth.debian.org/u64q/knucleotide-des...
But they say 'don't write a custom hash table', not 'don't write a custom hash function'. Maybe the problem is that the data in this benchmark is just not a good way to exercise the hash tables in a given language the way the benchmark intended. That probably means the benchmark should be modified; the complaint that some implementations use 'laughably bad' hash functions that seem to be measurably decent hash functions for the data at hand seems really strange.
> Since C doesn't have a standard library hashmap, you can write an entirely custom one just for the benchmark.
NOT TRUE!
#315195 states the opposite!
"k-nucleotide will explicitly require built-in / library HashMap"
Am I allowed to use a custom HashMap for the game, or am I required to use the one in std::collections? My understanding is that it's the latter, and so that penalizes Rust or any other language that includes a map in their standard library.
In fact, the current Rust k-nucleotide DOES use a custom Hash function with a library Hash map.
No one is allowed to -- in your words -- "write an entirely custom [hashmap] just for the benchmark".
(This was all discussed to death, in early December on https://www.reddit.com/r/rust/ ).
Meanwhile, from the task description: Please don't implement your own custom "hash table" - it will not be accepted.
If you want an additional Rust hashmap implementation then you need to persuade the Rust community to provide one in std::collections.
No one is allowed to -- in your words -- "write an entirely custom [hashmap] just for the benchmark".
Again, this means that for Rust, it _has_ to be in the standard library. But C doesn't have one in the standard library. So C gets to write their own, and Rust doesn't.
I still can't get all the moral over this.
The best thing here for Rust would be to achieve the best performance on this game and nothing else.
To me this game is important and winning on it is more yet. So, if Rust don't have better result by "morals" this is so wrong that hurts. But each time I read these responses I ask myself if there is not a real problem there.
I like Rust, but performance is decisive. It is really sad to see that Rust keeps going down on this game and that makes me question myself when trying to invest time on Rust.
If, on the other hand, you're just treating the benchmarks as a fun competition akin to code golf, then by all means, hacks ahoy!
That's absurd. The benchmarks game doesn't measure real-world performance in any sense whatsoever. And many of the implementations are written in ways that you'd never ever put into production code (for example, the laughably bad hashing functions that are mentioned elsewhere in this thread). From what I understand, the Rust implementations tend to not get up to those shenanigans, which makes them appear worse than the implementations from other languages that do.
It's not the goal of a programming language to be the top on a silly benchmarks game. The goal is to be an excellent language for real-world use.
I don't know the background of these people, but they couldn't be more wrong. So that looks like "morals" or just they don't ever saw real C code out there.
For example https://github.com/haipome/fnv/blob/master/fnv.c
http://benchmarksgame.alioth.debian.org/u64q/program.php?tes...
Imagine you got a million strings of variable length such that the last 4 bytes are guaranteed to be unique value between 0 and 999999. Lopping off the last four bytes is a perfectly good hash function for that kind of data.
So yeah, if you can come up with a tailored hash algorithm that works correctly and is DoS-immune for all possible input for the program, go ahead and use that tailored hash algorithm. But if your hash algorithm is only correct for the specific input data the benchmark game uses, and would not be collision-resistant under real world conditions, then you really shouldn't use that hash as it does not demonstrate anything useful.
Your argument seems to be that that's not the intent of the benchmark which may be true but it's not clear from the rules provided at all. To me, it looks like the opposite is true - they talk about using a standard hash table and most of those allow user-specified hash functions.
Btw. this site is extremly bad for benchmarks since it also measure's the startup time of the runtime in java/go/rust.
[1] https://github.com/duneroadrunner/SaferCPlusPlus-BenchmarksG...
Quote?
I've tried learning C/C++ at several times but I just don't have the inclination to have to bother about null-terminating strings, etc in 2016. I don't mind spending a little more time getting something to compile if it prevents silly mistakes.
Having said all that, it's obviously too early for me to say whether I like Rust. I'm picking it up pretty quickly since I know FP thanks to Scala, but I'll see how much time I spend fighting the borrow checker.
Feature-wise, Rust features generics and functional programming via higher-order functions and iterators, which is something that Go especially lacks in. Go doesn't have nice concepts like `Option`, `Result`, or `Iterator`. That's not something I'd personally want to live without today. The Go method is effectively writing boiler plate code everywhere, which leads to much room for error prone implementations that require more testing.
I haven't felt that Rust was more complex than Go, at least when you're actually writing software in Rust. Rust libraries feature semantic versioning and are automatically downloaded and verified at build time based on the contents of your `Cargo.toml` and `Cargo.lock` files. No importing of Git repositories directly required. Go does not provide an equal on that front.
There are a lot of great libraries out there to bring you extreme performance, simply, such as the bytecount crate, which is just a library that features a single function, a function that counts the occurrence of a specific byte, 32 bytes at a time with AVX, with additional SSE/SIMD implementations depending on what the processor supports.
All there is to truly know about Rust is the borrowing and ownership mechanism and how to implement a custom `Iterator`. If you have a solid understanding of both then you've pretty much mastered all you need to know about Rust.
The borrowing and ownership mechanism can be simplified down to: - Passing a variable by value will move ownership, dropping the original variable from memory - Passing a variable by mutable reference will keep the original variable, but allow you to modify the variable. - You may only borrow a variable mutably once at a time, and you may not immutably borrow while mutably borrowing. - You may have as many immutable borrows as you want, so long as you aren't modifying that value. - You may mutably borrow a field in a struct, and then mutably borrow a different field in the same struct simultaneously, so long as you aren't also mutably borrowing the overall struct. - You can use `Cell` and `RefCell` to allow for mutably modifying an immutable field in a struct. - You may mutably borrow multiple slices from the same array simultaneously so long as there is no overlap. - Safe memory practices means that instead of mutably borrowing the same variable in multiple places, you queue the changes to make in a separate location and apply them serially one after another.
Then for the `Iterator` trait, you would know that all traits have required methods, whereby as long as you implement the required methods for your type, you will automatically gain all of the other methods associated with the trait. For the `Iterator` type, you only need to implement the `next` method, and that looks something like so:
``` struct DataIterator<'a> { data: &'a [u8], index: usize, }
enum Token<'a> { One(&'a [u8]), Two(&'a [u8]) } impl<'a> Iterator for DataIterator<'a> { type Item = Token<'a>; fn next(&mut self) -> Option<Token<'a>> { let start = self.read; for element in self.data.iter().skip(self.read) { self.read += 1; // if next value is found then return Some(&value[start..self.read]) } None } } ```
There are only seven instances of unsafe: https://github.com/BurntSushi/ripgrep/search?q=unsafe&type=C...
Four of them are related to calling libc/kernel32 functions, which need unsafe to be called. One is due to using a memory map, which needs unsafe to be called. Only two are actual unsafe rust functions.
That's a "real-world" example though, you're asking for a more specific implementation. I can't compare because my Go would be very poor; if you posted a Go implementation, I'd be willing to give this a shot and write a Rust one.
(Mostly out of personal interest, I don't think it really proves anything larger about the two languages.)
Your benchmark proposal is interesting on the surface, but if done correctly, its speed will come down to highly optimized SIMD routines. For example, in Go, the code to search for a single byte on amd64 is written in Assembly (as it is in glibc too). This means it's probably not a good indicator of language performance overall.
Then compare it with similar tools written in Go, like sift and the platinum searcher.
The problem is, searching a file quickly is not as simple as you want to believe. If you show me a naive line by line approah, I'll show you an approach that is much faster but more complex. Top speed requires work, sometimes algorithmic work, regardless of the language you choose.
Of course, I did it myself many times. But this is NOT the point, I've already wrote it. The point is that I wrote naive approach in both languages and it's a lot faster in Go. Which is a reply to what OP wrote (don't forget where this discussion started). In this case this is the fact and I don't see any reason to fight with facts if we get bias out of the equation. In other cases? I don't know. What I would expect is that naive searching for one string in haystack to be faster than Go in a language that is performance/zero-cost abstractions oriented like Rust but its false in this case. But it doesn't mean it's false in every case. And to be honest writing that "but they have faster implementation in assembler" is not an excuse at all, you can also write your own, especially if it works so well for those languages that have custom asm for specific platforms. In the end average Joe will not care if it's hand written assembler, he will care that his naive solution using standard library without any magic is just faster.
I tried your challenge, and the first data point I uncovered contradicts this. Here is the source code of both programs: https://gist.github.com/anonymous/f01fc324ba8cccd690551caa43... --- The Rust program doesn't use unsafe, doesn't explicitly use C code, is shorter than the Go program, faster in terms of CPU time and uses less memory. I ran the following:
$ /usr/bin/time -v ./lossolo-go /tmp/OpenSubtitles2016.raw.sample.en the
$ /usr/bin/time -v ./target/release/lossolo-rust /tmp/OpenSubtitles2016.raw.sample.en the
The Go program takes ~6.5 seconds and has a maximum heap usage of 7.7 MB. The Rust program takes ~4.2 seconds and has a maximum heap usage of 6 MB. (As measured by GNU time using `time -v`.)
---
IMO, both programs reflect "naive" solutions. The point of me doing this exercise is to show just how silly this is, because now we're going to optimize these programs, but we'll limit ourselves to smallish perturbations in order to put a reasonable bound on the task.
If I run the Go program through `perf record`, the top hotspot is runtime.mallocgc. Now, I happen to know from experience that Scanner.Text is going to allocate a new string while Scanner.Bytes will not. I also happen to know that the Go standard library `bytes` package recently got a nice optimization that makes bytes.Contains as fast as strings.Contains: https://github.com/golang/go/commit/44f1854c9dc82d8dba415ef1... --- Since reading into a Go `string` doesn't actually do any UTF-8 validation, we don't lose anything by switching to using raw bytes.
Knowing this, we can tweak the Go program to great effect: https://gist.github.com/anonymous/c98dc8f6be6d414ae3e7aa6931... --- Running the same command as above, we now get a time of ~2.3 seconds and a maximum heap usage of 1.6 MB. That's impressive.
Now let's see if we can tweak Rust, which is now twice as slow as the Go program. Running perf, it looks like there's an even split between allocation, searching and UTF-8 validation, with a bit more towards searching. Like the Go program, let's attack allocation. In this case, I happen to know that the `lines` method returns an iterator that yields `String` values, which implies that it's allocating a fresh `String` for every line, just like our Go program was. Can we get rid of that? The BufReader API provides a `read_line` method, which permits the caller to control the `String` allocation. If we use that, our Rust program is tweaked to this: https://gist.github.com/anonymous/a6cf1aa51bf8e26e9dda4c50b0... --- It's not quite as symmetrical as a change as we made to the Go program, but it's pretty straight-forward IMO. Running the same command as above, we now get a time of ~3.3 seconds and a maximum heap usage of 6 MB.
OK, so we're still slower than the Go program. Looking at the profile again, the time now seems split completely between searching and UTF-8 validation. The allocation doesn't show up at all any more.
Is this where you got stuck? The next step from here isn't straight-forward because getting rid of the UTF-8 validation isn't possible to do safely while still using the String/&str search APIs. Notably, Rust's standard library doesn't provide a way to search an `&[u8]` directly using optimized substring search routines. Even if you knew your input was valid UTF-8 before hand, there's no obvious place to insert an unsafe `from_utf8_unchecked` because the BufReader itself is in control of producing the string contents. (You could do this by switching to using `BufReader.read_until` and then transmuting the result into an &str, but that would require unsafe.)
Let's take a leap. Rust's regex library has a little known feature that it can actually search the contents of an &[u8]. Rust's regex library isn't part of the standard library, but it is maintained as an official crate by the Rust project. If you know all of this, then it's possible to tweak the Rust program just a bit more to regain the speed lost by UTF-8 checking: https://gist.github.com/anonymous/bfa42d4f86e03695f3c880aace... --- Running the same command as above once again, we now get a time of ~2.1 seconds and a maximum heap usage of 6.5 MB.
In sum, we've beaten Go in CPU time, but lost the Battle for Memory and the battle for obviousness. Beating Go required noticing the `read_until` API of BufReader and knowing that 1) Rust's regexes are fast and 2) they can search &[u8] directly. It's not entirely unreasonable, but to be fair, I've done this without explicitly using any unsafe or any C code.
None of this process was rocket science. Both the Go and Rust programs were initially significantly sub-optimal because of allocation, but after some light profiling, it was possible to speed up both programs quite a bit.
---
Compared to the naive solution, some of our search tools can be a lot faster. Performing the same query on the same corpus:
ripgrep 1.13 seconds, 7.7 MB
ripgrep 1.35 seconds, mmap
GNU grep 1.73 seconds, 2.3 MB
ag 1.80 seconds, mmap
pt 6.41 seconds, mmap
sift 50.21 seconds, 16.6 MB
From here, the next step would be stop parsing lines and start searching the entire buffer directly. (I hope to make even this task very easy by moving some of the searching code inside of ripgrep to an easy to use library.)
---
In sum, your litmus test essentially comes down to these trade offs:
- Rust provides a rich API for its String/&str types, which are guaranteed to be valid UTF-8.
- Rust lacks a rich substring search API in the standard library for Vec<u8>/&[u8] types. Because of this, efficient substring search using only the standard library has an unavoidable UTF-8 validation cost in safe code.
- Go doesn't do any kind of UTF-8 checking and provides mirrored substring search APIs between its `bytes` and `strings` packages.
- The actual performance of searching in both programs probably boils down to optimized SIMD algorithms. Therefore, once you get past the ability to search each line of a file with minimal allocation, you've basically hit a wall that's probably the same in most mainstream languages.
In my opinion, these trade offs strike me as something terribly specific, and it's probably not something that is usefully generalizable. More than that, in the naive case, Rust is doing you a good service by checking that your input is valid UTF-8, which is something that Go doesn't do. I think this could go either way, but I think it's uncontroversial that guaranteeing valid UTF-8 up front like this probably eliminates a few possibly subtle bugs. (I will say that my experience with text encoding in Go has been stellar though.)
Most importantly, both languages at least have a path to writing a very fast program, which is often what most folks end up caring about at the end of the day.
Do you think you could refactor out bytestring-based string manipulation into its own library? Even better would be something that worked for all encodings (using https://github.com/servo/tendril or something)
IIRC, someone was working on making the Pattern trait work on &[u8], but I'm guessing that work is stalled.
To factor it out into a separate crate means copying the &str substring routines, since there's no way to safely use them on an &[u8] from the standard library. (bluss did that in the `twoway` crate, so we could just use that.)
It does seem like a plausible thing to do, at least until std gets better &[u8] support.
> Even better would be something that worked for all encodings
I suspect the standard practice here is something like "transcode to UTF-8 and then search the UTF-8." (This is what I hope to do with ripgrep.)
> (using https://github.com/servo/tendril or something)
I don't think I know what problems tendril is solving, so it's not clear to me what it's role is.
EDIT: and how can you have time to write this? I just usually close the browser tab when this situation occurs...
My opinion in this case is empirically checked. I am not saying this to start flame war, I am just showing my observations in particular example. I can also say that Rust regex implementation eats Go regex implementation by magnitudes (performance wise) and it will also be true and I am not looking for any flame war in this case also. I am only sharing my experience that I've backed up with proofs (code + perf results) for that particular use case. This is a factual discussion, I don't agree it's pointless.
It's 2.6 seconds for Go and 3.5 seconds for Rust, both perf results and code here: http://pastebin.com/WwhvHH6S
Your Go program is not what I would expect. A bufio.Scanner is the idiomatic (and naive) way to read lines in Go. But this is immaterial. Your results are consistent with mine. They aren't different. I just included more analysis and more programs to provide proper context to this challenge of yours.
Rust 4.6s
Go 3.1s
EDIT:
Using grep 0.6s
Then I have installed rigrep and...
Using ripgrep 0.4s
Really nice burntsushi. I am surprised by those (ripgrep) results compared to grep.
Using a `lines()` approach is naive because that allocates owned heap-allocated strings. An optimal, non-naive solution would not use heap allocation for buffering lines but use a stack-allocated array. That alone would bring significant speedup versus the line approach.
As for ripgrep, it's a pretty comprehensive application that makes use of SIMD/AVX so it's only natural that it's fast.
It took 4.6s which only confirms what I wrote on beginning when we started this discussion. Perf counters for your solution in here:
https://gist.github.com/mmstick/a8316ba0514f9d9ab33b18fa9b91...
As for timing, I'm doubtful that Go is any faster than Rust. I don't have this www.js file so I can't test it on my laptop, but I'm pretty sure you didn't even attempt to do things like enabling LTO, setting O3, disabling jemalloc, and using the musl target. All these things can make significant differences to the performance of the binary that's produced.
For me it's not naive solution. Do you have any proof it is? Can you mathematically prove that it's naive solution? I know I can't. For everyone naive solution is something different, what I saw in burntsushi reply and what I wrote myself is the closest to what I think are naive solutions.
> like enabling LTO, setting O3, disabling jemalloc, and using the musl target.
And this is for sure not part of naive solution either.
> You may not want to accept it because that contradicts your claim of the Go solution being faster
This is not true. You can find perf numbers of his solution in my second reply to him. Or you can compare those solutions yourself.
You can't prove a negative.
You are not proving anything until you post your Go code, which you don't seem to have done (please correct me if I'm wrong, I am legitimately curious to see for myself how Go and Rust stack up against each other for this problem). Until then, all you're doing is making vague claims backed up by precisely zero evidence. Why should anyone take you seriously?
When did you ask on IRC? I'll take a look.
I'm not sure what such an example would prove, other than maybe "Go is better than a subset of Rust excluding some of the core features of the language and its greater ecosystem".
This is a tricky requirement. The standard libraries of both languages use a crap ton of unsafe code. You might end up just asking for a comparison of which standard library is bigger.
But at the end of the day, both languages have perfectly standard buffered readers (https://doc.rust-lang.org/std/io/struct.BufReader.html), shouldn't a simple search like this compile to the exact same code?
Go is good, even great, at many things. But it's a language largely defined by its limitations, usually intentionally. It's an engineering language, not made for big abstractions. For me, the largest frustration is that the language gets in the way, and the pain increases with the scale of the problem. Which is to say: I think Go scales to large projects just fine, but there are problems where you'd like to build big building blocks on top of smaller blocks on top of smaller blocks, and Go doesn't lend itself to certain kinds of big, composable, data-oriented abstractions. It's small building blocks all the way.
I've bumped into several very real problems recently where Go's coarse, not-very-data-oriented imperative approach has revealed itself as a liability, and where I found myself fantasizing how I could have done it in just a few elegant lines in Haskell. Sometimes they're about expressing things simply in a composable manner, and sometimes these problems simply manifest themselves in immense blobs of boilerplate/repetition (for example, because you have to implement the same method a few dozen times on different data structures, which in a different language could be solved with a generic implementation), where Go's solution is to either eschew type safety, use slow reflection APIs, or programmatically generate the Go code as part of the build process.
Go's is also frustrating in its selective pragmatism. Where Go has chosen to automate and sugarcoat some complicated things (memory management, goroutines), it's stubbornly unpragmatic about other things (error handling, working with polymorphic data, memory safety). Go has been ridiculed for its simplistic error system, but I'm not an extremist here; I'm all for errors being values, and not a fan of exceptions. But if you look at actual Go code, a huge amount of code has to interact with errors. When nearly every function is riddled with "if err != nil", you should know that your language is crying out for just a little syntactic sugar. Or a solid type-system solution for that matter. Enums (Rust-style) and pattern matching wouldn't go against Go's grain at all, but since Go is "done", we're stuck with how it is.
I think Go's focus on simplicity is very important (I'm a big fan of the Wirth school of languages), and my worry about Rust and Swift is that they never learned this lesson. To me, both Rust and Swift looked more promising early in the design process than in their current state; Swift looks increasingly like Scala every time I visit it, whereas Rust often feels lost in a sea of punctuation. That said, my annoyance with Go is acute enough that I'm willing to deal with a few downsides if I can get a language that better matches the kinds of projects that I build.
FWIW the Rust team does consider simplicity to be important. However, it's not the only goal, which means that sometimes it has to be sacrificed to be able to make something possible. But most new language features do get discussed in the context of a "complexity budget".
So Rust tries pretty hard to ensure it doesn't get more complex than it has to; but it doesn't put simplicity as the overriding goal and hang all else to get it.
> Enums and pattern matching wouldn't go against Go's grain at all
I've said this many times before -- I don't miss generics from Go; I understand why they're not in the language. I miss ADTs and pattern matching.
If I look at everything I used to write in C, I'd say 80% is well suited for Go and the rest I would fallthrough to Rust for. For the stuff where having a GC and slightly less control is OK, I don't see why I would want to use Rust. Rust is just much more complex and I prefer to keep it simple stupid (KISS).
Basically, Go is good for 90% of what I used to use Java for and 80% of what I used to use C for... trying to understand where it makes sense for Rust to fit in.