Nothing particularly notable here. A lot of it seems to be 'We have something in-house designed for our use cases, use that instead of the standard lib equivalent'.
The rest looks very reasonable, like avoiding locale-hell.
Some of it is likely options that sand rough edges off of the standard lib, which is reasonable.
> We have something in-house designed for our use cases, use that instead of the standard lib equivalent
Yea, you encounter this a lot at companies with very old codebases. Don't use "chrono" because we have our own date/time types that were made before chrono even existed. Don't use standard library containers because we have our own containers that date back to before the STL was even stable.
I wonder how many of these (or the Google style guide rules) would make sense for a project starting today from a blank .cpp file. Probably not many of them.
For the containers in particular this makes a lot of sense because the C++ stdlib containers are just not very good. Some of this is because C++ inherited types conceived as pedagogic tools. If you're teaching generic programming you might want both (single and double) extrusive linked list types for your explanation. But for a C++ programmer asking "Which of these do I want?" the answer is almost always neither.
The specification over-specifies std::unordered_map so that no good modern hash table type could implement this specification, but then under-specifies std::deque so that the MSVC std::deque is basically useless in practice. It requires (really, in the standard) that std::vector<bool> is a bitset, even though that makes no sense. It sometimes feels as though nobody on WG21 has any idea what they're doing, which is wild.
Intrusive linked lists still firmly have a place in modern code, for reasons other than performance. I don’t know many good reasons for extrusive linked lists, even before caches. There might be a few, but a dynamic array is (and has always been?) usually preferable to an extrusive list.
Don't use standard library containers because we have our own containers that date back to before the STL was even stable.
Flashback to last job. Wrote their own containers. Opaque.
You ask for an item from it, you get back a void pointer. It's a pointer to the item. You ask for the previous, or the next, and you give back that void pointer (because it then goes through the data to find that one again, to know from where you want the next or previous) and get a different void pointer. No random access. You had to start with the special function which would give you the first item and go from there.
They screwed up the end, or the beginning, depending on what you were doing, so you wouldn't get back a null pointer if there was no next or previous. You had to separately check for that.
It was called an iterator, but it wasn't an iterator; an iterator is something for iterating over containers, but it didn't have actual iterators either.
When I opened it up, inside there was an actual container. Templated, so you could choose the real inside container. The default was a QList (as in Qt 4.7.4). The million line codebase contained no other uses; it was always just the default. They took a QList, and wrapped it inside a machine that only dealt in void pointers and stripped away almost all functionality, safety and ability to use std::algorithm
I suspect but cannot prove that the person who did this was a heavy C programmer in the 1980s. I do not know but suspect that this person first encountered variable data type containers that did this sort of thing (a search for "generic linked list in C" gives some ideas, for example) and when they had to move on to C++, learned just enough C++ to recreate what they were used to. And then made it the fundamental container class in millions of lines of code.
The complete refactor, bringing it forwards from VS2008 to VS2022, and from a home-built, source-code edited Qt 4.7.4 to Qt 6.something, took about two years from start to finish.
> I wonder how many of these (or the Google style guide rules) would make sense for a project starting today from a blank .cpp file. Probably not many of them.
The STL makes you pay for ABI stability no matter if you want it or not. For some use cases this doesn't matter, and there are some "proven" parts of the STL that need a lot of justification for substitution, yada yada std::vector and std::string.
But it's not uncommon to see unordered_map substituted with, say, sparsehash or robin_map, and in C++ libraries creating interfaces that allow for API-compatible alternatives to use of the STL is considered polite, if not necessarily ubiquitous.
> I wonder how many of these (or the Google style guide rules) would make sense for a project starting today from a blank .cpp file. Probably not many of them.
That also depends on how standalone the project is. Self-contained projects may be better off with depending on standard library and popular third-party libraries, but if a project integrates with other internal components, it's better to stick to internal libraries, as they likely have workarounds and special functionality specific to the company and its development workflow.
I'd argue that the optimum was in long run to migrate to the standard version, that everyone (e.g. new employees) know. Replacing the usually particular (or even weird) way implemented own flavour.
I know, I know, long run does not exists in today's investor dominated scenarios. Code modernization is a fairytale. So far I seen no exception in my limited set of experiences (but with various codebases going back to the early 90's with patchy upgrades here and there, looking like and old coat fixed many many times with diverse size of patches of various materials and colour).
When I led C++ style/modernization for Chromium, I made this argument frequently: we should prefer the stdlib version of something unless we have reason not to, because incoming engineers will know it, you can find advice on the internet about it, clang-tidy passes will be written for it, and it will receive optimizations and maintenance your team doesn't have to pay for.
There are cases, however, when the migration costs are significant enough that even those benefits aren't really enough. Migrating our date/time stuff to <chrono> seemed like one of those.
Are there really any good reasons to start a brand new project in c++ though? No one who can write modern c++ has any trouble with rust in my experience, and all the other common options are even quicker to pick up. Creating bindings isn't hard anymore if your niche library doesn't have any yet. Syntactic preference I guess, but neither c++ or rust are generally considered elegant or aesthetic choices.
Because "brand new" doesn't mean devoid of context. Within your domain, there will still be common libraries, interfaces, and tools.
C++ is very flexible, with a lot of very mature tooling and incredibly broad platform support. If you're writing some web server to run on the hardware of your choosing, then sure, that doesn't matter. But if you're writing something deeply integrated with platform/OS interfaces, or graphics, or needs to support some less common platforms, then C++ is often your only practical option for combining expressiveness and performance.
This is the sort of info I was trolling for, but what are those platforms and os? Targets llvm doesn't handle yeah c++ makes sense, or c. A sibling mentions xcode, which makes sense. Graphics seems questionable, vulkan support is fine. Windows support has seemed finetoo, the same gui has worked as what we wrote for Linux.
Dependencies. There are billions of lines of C++ out there that have been optimized and production hardened over decades that you might want to reuse. Rust lang interoperability with anything but C sucks in practice.
Maybe, maybe not. But either way it's just plain rude to charge into a C++ thread to drop a comment saying how the language sucks and you should use (insert other language) instead.
C++ was the GUI king during the 1990's, and none of the Rust toolkits is half as good as the surviving frameworks, like C++ Builder, Qt, wxWidgets, heck even MFC has better tooling.
Yes. If you're targeting Apple platforms and want to allow clients to use your product in Xcode (the common case) or even need Swift/ObjC interop yourself, using rust or anything not explicitly supported by Apple and Xcode is just too fiddly.
(Shrug) If I want Rust, I'll feed my C++ to an LLM and tell it to port it to Rust. Since we've been assured that Rust magically fixes everything that's wrong, bad, or unsafe about C++, this seems like a sound approach.
We probably aren't that far off actually. Even taking asm with no symbols back into rust works well. You have truth, just have the agent repeat until the asm matches. Doesn't work on giant codebases, but on a few functions it absolutely does. And while the llm may get the algorithm wrong, the type system does seem to help it generate useful code in rust for a starting place v
Yeah, but then just let the agent generate proper C++ code, contrary to an human it doesn't forget about best practices, or how ownership is supposed to be handled.
You missed the other take, with AI assisted coding, you can stay in C++, as it will take care everything is coded with enough care.
Or why bother with Rust, when the LLM gets to generate C++ code with best practices.
While I like Rust, I think AI as the next abstraction step in programming has kind of taken its relevance away, when computer assisted programming is part of the workflow.
Strange. I wouldn’t trust the output of a coding agent and I would want stronger review of its output. If it passes a strict compiler that gives me more confidence than if it passed a lax one.
But sure, if you trust it to have written C++ to a higher standard than the experts, then go for it.
It's weird to me, as the former lead maintainer of this page for ten years or so, that this got submitted to both r/c++ and HN on the same day. Like... what's so exciting about it? Was there something on the page that caught someone's eye?
Somewhat notable is that `char8_t` is banned with very reasonable motivation that applies to most codebases:
> Use char and unprefixed character literals. Non-UTF-8 encodings are rare enough in Chromium that the value of distinguishing them at the type level is low, and char8_t* is not interconvertible with char* (what ~all Chromium, STL, and platform-specific APIs use), so using u8 prefixes would obligate us to insert casts everywhere. If you want to declare at a type level that a block of data is string-like and not an arbitrary binary blob, prefer std::string[_view] over char*.
Isn't the real reason to use char8_t over char that it that char8_t* are subject to the same strict aliasing rules as all other non-char primitive types? (i.e., the compiler doesn't have to worry that a char8_t* could point to any random piece of memory like it would for char*?).
> there is no guarantee `char` is 8 bits, nor that it represents text, or even a particular encoding.
True, but sizeof(char) is defined to be 1. In section 7.6.2.5:
"The result of sizeof applied to any of the narrow character types is 1"
In fact, char and associated types are the only types in the standard where the size is not implementation-defined.
So the only way that a C++ implementation can conform to the standard and have a char type that is not 8 bits is if the size of a byte is not 8 bits. There are historical systems that meet that constraint but no modern systems that I am aware of.
char8_t also isn't guaranteed to be 8-bits, because sizeof(char) == 1 and sizeof(char8_t) >= 1. On a platform where char is 16 bits, char8_t will be 16 bits as well
The cpp standard explicitly says that it has the same size, typed, signedness and alignment as unsigned char, but its a distinct type. So its pretty useless, and badly named
Wouldn't it be rather the case that char8_t just wouldn't exist on that platform? At least that's the case with the uintN_t types, they are just not available everywhere. If you want something that is always available you need to use uintN_least_t or uintN_fast_t.
It is pretty consistent. It is part of the C Standard and a feature meant to make string handling better, it would be crazy if it wasn't a complete clusterfuck.
> There's no guarantee char8_t is 8 bits either, it's only guaranteed to be at least 8 bits.
Have you read the standard? It says: "The result of sizeof applied to any of the narrow character types is 1." Here, "narrow character types" means char and char8_t. So technically they aren't guaranteed to be 8 bits, but they are guaranteed to be one byte.
Well platforms with CHAR_BIT != 8. In c and c++ char and there for byte is atleast 8 bytes not 8 bytes. POSIX does force CHAR_BIT == 8. I think only place is in embeded and that to some DSPs or ASICs like device. So in practice most code will break on those platforms and they are very rare. But they are still technically supported by c and c++ std. Similarly how c still suported non 2's complement arch till 2023.
That's where the standard should come in and say something like "starting with C++26 char is always 1 byte and signed. std::string is always UTF-8" Done, fixed unicode in C++.
But instead we get this mess. I guess it's because there's too much Microsoft in the standard and they are the only ones not having UTF-8 everywhere in Windows yet.
Of course it can be made UTF-8. Just add a codepoints_size() method and other helpers.
But it isn't really needed anyway: I'm using it for UTF-8 (with helper functions for the 1% cases where I need codepoints) and it works fine. But starting with C++20 it's starting to get annoying because I have to reinterpret_cast to the useless u8 versions.
Is there a single esoteric DSP in active use that supports C++20? This is the umpteenth time I've seen DSP's brought up in casual conversations about C/C++ standards, so I did a little digging:
Aside from that, from what I can tell, those esoteric architectures are being phased out in lieu of running DSP workloads on Cortex-M, which is just ARM.
I'd love it if someone who was more familiar with DSP workloads would chime in, but it really does seem that trying to be the language for all possible and potential architectures might not be the right play for C++ in 202x.
Besides, it's not like those old standards or compilers are going anywhere.
Cadence DSPs have C++17 compatible compiler and will be c++20 soon, new CEVA cores also (both are are clang based).
TI C7x is still C++14 (C6000 is ancient core, yet still got c++14 support as you mentioned).
AFIR Cadence ASIP generator will give you C++17 toolchain and c++20 is on roadmap, but not 100% sure.
But for those devices you use limited subset of language features and you would be better of not linking c++ stdlib and even c stdlib at all (so junior developers don't have space for doing stupid things ;))
How common is it to use Green Hills compilers for those DSP targets? I was under the impression that their bread was buttered by more-familiar-looking embedded targets, and more recently ARM Cortex.
Judging by the lack of modern C++ in these crufty embedded compilers, maybe modern C++ is throwing too much good effort after bad. C++03 isn't going away, and it's not like these compilers always stuck to the standard anyway in terms of runtime type information, exceptions, and full template support.
Besides, I would argue that the selling point of C++ wasn't portability per se, but the fact that it was largely compatible with existing C codebases. It was embrace, extend, extinguish in language form.
> Judging by the lack of modern C++ in these crufty embedded compilers,
Being conservative with features and deliberately not implementing them are two different thing. Some embedded compilers go through certification, to be allowed to be used producing mission critical software. Chasing features is prohibitively expensive, for no obvious benefit. I'd bet in 2030s most embedded compiler would support C++ 14 or even 17. Good enough for me.
Related: in C at least (C++ standards are tl;dr), type names like `int32_t` are not required to exist. Most uses, in portable code, should be `int_least32_t`, which is required.
In a lot of places, they point out the std implementation is strictly inferior to theirs in some way, so its not always organizational inertia, it's that the C++ standard types could have been designed strictly better with no tradeoff.
Not an Googler, but my, probably way too much romanticized, understanding of Google was that they never ask you about specific tech because for everything there's an in-house version.
The problem is that too many people drank too much koolaid and trying to parrot everything to a letter without understanding the bigger picture.
The best example would be Kubernetes. Employed by many orgs that have 20 devs and 50 services.
Reasonable summary. There's some massive NIH syndrome going on.
Another piece is that a lot of stuff that makes sense in the open source world does not make sense in the context of the giant google3 monorepo with however many billions of lines of code all in one pile.
> Nothing particularly notable here. A lot of it seems to be 'We have something in-house designed for our use cases, use that instead of the standard lib equivalent'.
The bulk of the restrictions are justified as "Banned in the Google Style Guide."
In turn the Google Style Guide bans most of the features because they can't/won't refactor most of their legacy code to catch up with post C++0x.
So even then these guidelines are just a reflection of making sure things stay safe for upstream and downstream consumers of Google's largely unmaintained codebase.
I don't think that's an accurate representation. There are a few features like that, but the majority of things banned in the Google style guide are banned for safety, clarity, or performance concerns. Usually in such cases Google and/or Chromium have in-house replacements that choose different tradeoffs.
A good decision. I tried to use it once and realized that it can't even work with UTF-8 properly. It's a mystery for me how such flawed design was standardized at all.
People that keep bringing this up always miss the rationable that Google code was written initially in a C style that isn't exception safe.
Key takeaway => "Things would probably be different if we had to do it all over again from scratch."
"On their face, the benefits of using exceptions outweigh the costs, especially in new projects. However, for existing code, the introduction of exceptions has implications on all dependent code. If exceptions can be propagated beyond a new project, it also becomes problematic to integrate the new project into existing exception-free code. Because most existing C++ code at Google is not prepared to deal with exceptions, it is comparatively difficult to adopt new code that generates exceptions.
Given that Google's existing code is not exception-tolerant, the costs of using exceptions are somewhat greater than the costs in a new project. The conversion process would be slow and error-prone. We don't believe that the available alternatives to exceptions, such as error codes and assertions, introduce a significant burden.
Our advice against using exceptions is not predicated on philosophical or moral grounds, but practical ones. Because we'd like to use our open-source projects at Google and it's difficult to do so if those projects use exceptions, we need to advise against exceptions in Google open-source projects as well. Things would probably be different if we had to do it all over again from scratch."
I'm quite happy to NOT have exceptions. I think they're a mistake as a language feature. What we need is first -class support for returning errors and propagating them, like what zig does. The next best thing are those RETURN macros that Google uses.
"first-class support for returning errors and propagating them" certainly sounds like exceptions! In fact, the compiler can even emit special tables that let the runtime completely skip over stack frames that don't need to do any cleanup during that propagation step!
Some languages have even innovated new kinds of exceptions that you can throw but that you are admonished should almost certainly never be caught.
Unfortunately even this is also better for tool support, a problem that using a bunch of macros solves. It's cool and good when a variable gets declared inside the guts of some macro expansion (and--critically--escapes those guts).
Both parts of your sentence refer to the Google style guide. This doc isn't the Google style guide. It's the Chromium modern c++ features doc. We don't talk about exceptions or platform-specific stuff (save a note on [[no_unique_address]]) in this doc.
Gonna venture a guess and say probably https://www.chromium.org/developers, as that's where all the information for folks who actually need to know that kind of thing lives.
The banned list proves that context matters more than having the newest tools. These features work well for small apps but they cause problems in a project this size.
IIRC a big part of Google’s coding guidelines is also about making it easy for people not heavily invested in a specific language to contribute safely. So not necessarily a project size but rather an organizational concern.
They’d rather see it done the same way it would’ve been in any other similar language than with a language specific feature.
There are also portability concerns in mind given that projects like Chromium have to be easily portable across a vast amount of platforms (this shows with things like long long which is also on the list).
Some of it is historical reasons or portability more than anything else. Chrome is an old C++ project and evolved many of its own versions of functionality before standardization; and there's benefit to staying on its own implementations rather than switching.
Agreed. I also prefer conformity over sporadic use of new features going against an already set of standards in a codebase. it's overall less cognitive load on whoever is reading it imho.
I'm not a c++ user but i'm pretty sure you should be able to pull-off a macro to do that ; in c you could alias the lib for something that breaks + alert ; I don't know how I would integrate such additional compiler checks in rust for other kinds of rules however - it's interesting to think about
Rust has been better than C++ about marking where there's a foot gun and where practical just nerfing it. core::mem::uninitialized() is an example. In 2016 if you'd said "Initializing my Doodad is too expensive, what do I do?" you might be pointed at the (unsafe) core::mem::uninitialized() which "initializes" any variable, such as your Doodad but er, doesn't actually initialize it.
But that's a foot gun, there are niche cases where this crazy stunt actually is correct (ZSTs are one), but in general the compiler was in effect licensed to cause absolute mayhem because this isn't a Doodad at all, which isn't what you wanted. So, they did three things:
1. Made a new type wrapper for this purpose, MaybeUninit<Doodad> might not be a Doodad yet, so we can initialize it later and until then the compiler won't set everything on fire but it's the same shape as the Doodad.
2. Marked core::mem::uninitialized deprecated. If you use it now the compiler warns you that you shouldn't do that.
3. De-fanged it by, despite its name, scrawling the bit pattern 0x01 over all the memory. The compiler can see this has some particular value and for common types it's even valid, a bool will be true, a NonZeroU32 will be 1, and so on. This is slow and probably not what you intended, but we did warn you that it was a bad idea to call this function still so...
Rust's compiler has six what it calls "lint levels" two of which can't be overridden by compiler flags, these handle all of its diagnostics and are also commonly used by linters like clippy.
Allow and Expect are levels where it's OK that a diagnostic happened, but with Expect if the diagnostic expected was not found then we get another diagnostic telling us that our expectation wasn't fulfilled.
Warn and Force-Warn are levels where we get a warning but compilation results in an executable anyway. Force-warn is a level where you can't tell the compiler not to emit this warning.
Deny and Forbid are levels where the diagnostic is reported and compilation fails, so we do not get an executable. Forbid, like Force-warn cannot be overriden with compiler flags.
You almost never see a list of banned Java features (or even banned C# features). On the other hand any serious C++ development team is going to have a list of banned features. Java eliminated the features that you would want to ban.
This seems factually incorrect and ignorant of history. Java has tons of things which shouldn't be used. Serialization (use Jackson now, not the built-in stuff), date/time (there's an entirely different namespace so you don't accidentally use garbage classes), etc.
C# similarly has old warts that are discouraged now. .NET Framework is a great example (completely different from modern c#, which used to be called "dotnet core"). WPF and MAUI are also examples. Or when "dynamic" was used as a type escape hatch before the type system advanced to not need it. ASP being incompatible with ASP.NET, the list goes on.
They're just languages, there's no reason to pretend they're perfect.
> C# similarly has old warts that are discouraged now. .NET Framework is a great example (completely different from modern c#, which used to be called "dotnet core"). WPF and MAUI are also examples. Or when "dynamic" was used as a type escape hatch before the type system advanced to not need it. ASP being incompatible with ASP.NET, the list goes on.
Almost all of this is incorrect or comparing apples to oranges.
.net framework and .net core are runtime and standard library impl, not languages. C# is a language that can target either runtime or both. Framework is still supported today, and you can still use most modern C# language features in a project targeting it. WPF and Maui are both still supported and widely used. ASP predates .net - c# was never a supported language in it. ASP.net core has largely replaced ASP.net, but it's again a library and framework, not a language feature.
Dynamic in c# and the dlr are definitely not widely used because it's both difficult to use safely and doesn't fit well with the dominant paradigm of the language. If you're looking for STD lib warts binaryserializer would have been an excellent example.
C# can be used in both .NET Framework and modern .NET (ex-core). In fact, it is possible for a C# project to target both .NET Framework and .NET with the exact same code, or to target to .NET Standard, where the same .DLL file can be loaded by both. Since the old Framework is in maintenance mode, some modern language features will not work there, but you can still be productive with the old framework.
Dynamic is largely unnecessary, and it was unnecessary even when it was introduced.
ASP and ASP.NET are completely unrelated. ASP was designed to allow dynamic webpages to be written in VBScript (like CGI). This is not something you want to do in modern languages.
There are clearly demarcated language features sections and library sections. That definitely makes sense for C++, it is a poorly designed language and you definitely have to know what features to avoid.
In C#, you would normally implement rules like this with a custom Roslyn Analyzer or with https://github.com/dotnet/roslyn/blob/main/src/RoslynAnalyze.... It’s fair to say C# projects tend to have smaller denylists than mature C++ projects, but banned APIs definitely exist in mature C# projects.
I think Java has plenty of features that end up implicitly banned at least. e.g. you will really never see a `java.util.Vector` in any modern code base. No one `implements Serializable` anymore in practice. `Objects.equals()` and `Objects.hashCode()` get you through 99% of equals/hash code implementations. The list goes on.
I guess the difference is it's rarely "dangerous" or "hard to reason about" using the old features unlike what I see in the C++ list. Java replaces things with better things and momentum shifts behind them kind of naturally because the better things are objectively better.
Java has so few features that there's not much to ban. Nevertheless Java definitely has language features (in addition to deprecated library APIs, which are many) that aren't recommended today. Off the top of my mind:
* `synchronized` blocks
* raw types
* anonymous inner classes
* checked exceptions
as well as specific uses of otherwise fine language features:
* "AbstractX" style abstract classes when static and default interface methods suffice
* old-style POJO classes when records are a better choice
As a Java programmer I can only think of one thing:
Reflection - unless you really need to do something fancy almost certainly a very bad idea in normal application code.
Other than that it’s either just limiting yourself to a specific JVM version or telling people not to use old syntax/patterns that were replaced with better options long ago.
Unless you develop games in Unity, and have banned C# features like classes or LINQ (because they allocate on heap and Unity garbage collector is bad and make your game to micro freeze / stutter). Sure there are cases where classes are fine (singletons, pooling), but still...
Since Chromium stopped allowing manifest-v2 extensions, i.e. significantly crippled what extensions can do and made it impossible to use some key extensions like uBlock Origin, I've decided to avoid it.
Anyway, about these C++ conventions - to each software house its own I guess. I don't think banning exceptions altogether is appropriate; and I don't see the great benefit of using abseil (but feel free to convince me it's really that good.)
I spent a while on an open source project debugging some bizarre crashes. Cant remember exact detail but something like Someone was throwing an exception inside a destructor which had been triggered inside another exception. It was layers deep inside a massively templated 3rd party dependency library. so I like wound up parsing the string in the exception message and doing weird logic to make the program keep going since the exception wasnt actually a dire situation as far as the main program was concerned. So Exceptions can be fine in theory but I understand the idea that a ban can simplify a lot of things.
If nothing else Abseil gives you state-of-the-art hashmaps that run circles around the STL ones, which are slow pretty much by definition. The spec precludes the modern ways of implementing them.
Sure; you can roll your own, or you can use one of the many other libraries that provide alternatives. There are obvious downsides to the former, and the latter is basically equivalent to using the Abseil ones in terms of practical tradeoffs.
Chromium uses hundreds of other open source libraries. Is there something about Abseil specifically that sticks in your craw?
> Our advice against using exceptions is not predicated on philosophical or moral grounds, but practical ones. ... Things would probably be different if we had to do it all over again from scratch.
They are clearly not against them per se. It simply wasn't practical for them to include it into their codebase.
And I think a lot of the cons of exceptions are handled in languages like F#, etc. If f calls g which calls h, and h throws an exception, the compiler will require you to deal with it somehow in g (either handle or explicitly propagate).
My issue with exceptions is also practical. If they didn't introduce significant stability issues, I'd have no problem. As it stands, it's impossible to write robust software that makes use of C++ exceptions.
> the compiler will require you to deal with it somehow in g
In low-level systems software, which is a primary use case for C++, exceptions can introduce nasty edge cases that are difficult to detect and reason about. The benefits are too small to justify the costs to reliability, robustness, and maintainability.
Exceptions in high-level languages avoid many of these issues by virtue of being much further away from the metal. It is a mis-feature for a systems language. C++ was originally used for a lot of high-level application code where exceptions might make sense that you would never use C++ for today.
> In low-level systems software, which is a primary use case for C++
I don't this this is true. There is A LOT of C++ for GUI applications, video games, all kind of utilities, scientific computing and others. In fact, I find that the transition to "modern" alternatives from native GUI toolkits in C/C++ has led to a regression in UI performance in general. Desktop programs performed better 20 years ago when everything was written in Win32, Qt, GTK and others and people did not rely on bloated Web toolkits for desktop development. Even today you can really feel how much more snappy and robust "old school" programs are relative to Electron and whatnot.
To clarify, you think that low-level systems software is only a secondary use case for C++? The part you quoted does not make claims about whether there are other primary use cases, just that low-level systems software is one of them, so it's not clear why it being useful elsewhere is a rebuttal of that.
The model of communicating errors with exceptions is really nice. The implementation in C++ ABI's is not done as well as it could be and that results in large sad path perf loss.
> That's true, except for languages that ensure you can't simply forget that something deep down the stack can throw an exception.
Sometimes it is not safe to unwind the stack. The language is not relevant. Not everything that touches your address space is your code or your process.
Exception handlers must have logic and infrastructure to detect these unsafe conditions and then rewrite the control flow to avoid the unsafety. This both adds overhead to the non-exceptional happy path and makes the code flow significantly uglier.
The underlying cause still exists when you don't use exceptions but the code for reasoning about it is highly localized and usually has no overhead because you already have the necessary context to deal with it cleanly.
If you forget to handle a C++ exception you get a clean crash. If you forget to handle a C error return you get undefined behavior and probably an exploit.
Rust is better here (by a lot), but you can still ignore the return value. It's just a warning to do so, and warnings are easily ignored / disabled. It also litters your code with branches, so not ideal for either I-cache or performance.
The ultimate ideal for rare errors is almost certainly some form of exception system, but I don't think any language has quite perfected it.
Only when you don't need the Ok value from the Result (in other words, only when you have Result<(), E>). You can't get any other Ok(T) out of thin air in the Err case. You must handle (exclude) the Err case in order to unwrap the T and proceed with it.
> It also litters your code with branches, so not ideal for either I-cache or performance.
Anyhow erases the type of the error, but still indicates the possibility of some error and forces you to handle it. Functionality-wise, it's very similar to `throws Exception` in Java. Read my post
Poor man's checked exceptions. That's important. From the `?` you always see which functions can fail and cause an early return. You can confidently refactor and use local reasoning based on the function signature. The compiler catches your mistakes when you call a fallible function from a supposedly infallible function, and so on. Unchecked exceptions don't give you any of that. Java's checked exceptions get close and you can use `throws Exception` very similarly to `anyhow::Result`. But Java doesn't allow you to be generic over checked exceptions (as discussed in the post). This is a big hurdle that makes Result superior.
C++ exceptions are fast for happy path and ABI locked for sad path. They could be much faster than they are currently. Khalil Estell did a few talks/bunch of work on the topic and saw great improvements. https://youtu.be/LorcxyJ9zr4
> "In low-level systems software, which is a primary use case for C++, exceptions can introduce nasty edge cases that are difficult to detect and reason about. The benefits are too small to justify the costs to reliability, robustness, and maintainability."
Interestingly, Microsoft C / C++ compiler does support structured exception handling (SEH). It's used even in NT kernel and drivers. I'm not saying it's the same thing as C++ exceptions, since it's designed primarily for handling hardware faults and is simplified, but still shares some core principles (guarded region, stack unwinding, etc). So a limited version of exception handling can work fine even in a thing like an OS kernel.
FWIW, I think it is possible to make exception-like error handling work. A lot of systems code has infrastructure that looks like an exception handling framework if you squint.
There are two main limitations. Currently, the compiler has no idea what can be safely unwound. You could likely annotate objects to provide this information. Second, there is currently no way to tell the compiler what to do with an object in the call stack may not be unwound safely.
A lot of error handling code in C++ systems code essentially provides this but C++ exceptions can't use any of this information so it is applied manually.
Exceptions are actually a form of code compression. Past some break even point they are a net benefit, even in embedded codebases. They're "bad" because the C++ implementation is garbage but it turns out it's possible to hack it into a much better shape:
My memory of F# is very rusty, but IIRC, there are two types of error handling mechanisms. One of them is to be compatible with C#, and the other is fully checked.
It really depends on how reliable you want the code to be. Many business application developers prioritize development speed and don't want to think about errors, for them checked exceptions may seem like a hassle. For developers who prioritize reliability unchecked exceptions are a huge problem because they are not part of the contract and can change without notice.
Because java is garbage-collected and doesn't have any of the problems of C++ exceptions, so checked exceptions just become a nuisance of having to try/catch everything.
Most codebases that ban exceptions do it because they parrot Google.
Google’s reasons for banning exceptions are historical, not technical. Sadly, this decision got enshrined in Google C++ Style Guide. The guide is otherwise pretty decent and is used by a lot of projects, but this particular part is IMO a disservice to the larger C++ ecosystem.
I think reasonable people can disagree about whether C++ exceptions are "good" or not.
There are things you can't do easily in C++ without using exceptions, like handling errors that happen in a constructor and handling when `new` cannot alloc memory. Plus, a lot of the standard library relies on exceptions. And of course there's the stylistic argument of clearly separating error-handling from the happy-path logic.
I won't argue that it's popular to ban them, though. And often for good reasons.
For exception-less C++, you'd declare an operator new() that doesn't throw exceptions and just returns NULL on allocation failure along with a simple constructor and a followup explicitly-called init() method that does the real work which might fail and returns an error value on failure.
If you're planning on shutting down, what's the fundamental difference between throwing an exception, vs simply complaining loudly and calling exit() ..?
Sometimes it’s useful to handle the exception somewhere near its origin so you can close related resources, lockfiles, etc. without needing a VB6 style “On Error GoTo X” global error handler that has to account for all different contexts under which the exceptional situation might have occurred.
> a VB6 style “On Error GoTo X” global error handler that has to account for all different contexts under which the exceptional situation might have occurred
... That seems like a pretty accurate description of how exception handling mechanisms are implemented under the hood. :)
The code that's throwing an exception typically does not know that the exception catcher will shut anything down.
And - very often, you would _not_ shut down. Examples:
* Failure/error in an individual operation or action does not invalidate all others in the set of stuff to be done.
* Failure/error regarding the interaction with one user does not mean the interaction with other users also has to fail.
* Some things can be retried after failing, and may succeed later: I/O; things involving resource use, etc.
* Some actions have more than one way to perform them, with the calling code not being able to know apriori whether all of them are appropriate. So, it tries one of them, if it fails tries another etc.
Yet, if you can only explain an exception using the word ‘exception’ you’re not making any head way.
I like the idea of an exception as a way to blow out of the current context in order for something else to catch it and handle in a generic manner. I don’t like the idea of an exception to hide errors or for conditional logic because you have to know what is handling it all. Much easier to handle it there and then, or use a type safe equivalent (like a maybe or either monad) or just blow that shit up as soon as you can’t recover from the unexpected.
> They're good for exceptional situations where foundamental, core assumptions are broken for some reason.
No, that's what assertions or contracts are for.
Most exceptions are supposed to be handled. The alternative to exceptions in C++ are error codes and `std::expected::. They are used for errors that are expected to happen (even if they may be exceptional). You just shouldn't use exceptions for control flow. (I'm looking at you, Python :)
The rest looks very reasonable, like avoiding locale-hell.
Some of it is likely options that sand rough edges off of the standard lib, which is reasonable.
reply