-
-
Notifications
You must be signed in to change notification settings - Fork 2.4k
This wiki page lists Zig community gathering places where you can ask questions.
https://github.com/ziglang/zig/wiki/Community
Here it is: Standard Library Documentation
However, please note the remaining issues with generated documentation.
You can also:
- Browse the stdlib code to see what public functions and types are available.
- Check out the community and ask questions if you need help. Newcomers are always welcome!
Because no human and no contemporary code editor is capable of handling tabs correctly. Humans tend to mix tabs and spaces on accident, and editors don't have a way to "indent with tabs, align with spaces" without pressing the space bar many times, leading programmers to use tabs for alignment as well as indentation.
Tabs would be better than spaces for indentation because they take up fewer bytes. But in practice, what ends up happening is incorrectly mixed tabs and spaces. In order to simplify everything, tabs are not allowed. Spaces are necessary; we can't ban spaces. But tabs are not strictly needed, so the null hypothesis is to not have them.
Maybe someday, we'll switch to tabs for indentation, spaces for alignment and make it a compile error if they are incorrectly mixed. But if we did that today, writing Zig code would be too hard. For now your options are to configure your editor to insert spaces when you press the tab key, or configure your editor run zig fmt
on save (recommended).
What will make it into the final language specification? It isn't decided yet and it doesn't really matter. Just run zig fmt
on save.
The biggest reason to enforce an indentation and line endings is that it eliminates energy spent on debating what the standard should be, since the standard is enforced by the compiler.
The issue of other whitespace characters has been discussed too, and similar decisions were made. Zig aims to offer only one way to do things whenever possible. This makes the cognitive load lower for programmers and keeps the compiler code base simpler and easier to understand.
In the words of the Go community,
gofmt
is nobody's favorite, yetgofmt
is everybody's favorite.
They were copied verbatim from other projects rather than trying to be self-consistent. The Command Line Interface is not finalized. We can make sure it is consistent and intuitive in an organization pass before releasing 1.0.
zig fmt
will parse comments for special directives.
In this example all code between // zig fmt: off
and // zig fmt: on
will be excluded from formatting:
// zig fmt: off
const matrix = Matrix{1.0, 0.0, 0.0, 0.0,
0.0, 1.0, 0.0, 0.0,
0.0, 0.0, 1.0, 0.0,
0.0, 0.0, 0.0, 1.0};
// zig fmt: on
When building or compiling with Zig a build-cache is used. This particular error indicates filesystem security has prevented access to a directory or file used in the global build-cache. Fixing permissions should solve the issue.
note: It is safe to manually remove cache directories when no zig compiler process is active.
The build cache can be found in the following locations unless overridden with command-line options:
TYPE | OS | DIRECTORY | NOTE |
---|---|---|---|
global | all | $XDG_CACHE_HOME/zig | if env is set |
$HOME/.cache/zig | |||
Windows | %LOCALAPPDATA%\zig | ||
local | all | $PWD/zig-cache |
In summary, Jimmi made a good attempt at implementing a StringSwitch
in comptime
and concluded that good old chained if
statements were fastest.
For details see match.zig . Note that switching on variable identifiers as const strings is easy with the @""
syntax.
- The first step is to reduce your code to isolate the issue as much as possible.
- The second step is to setup your code to not require an executable with
main()
or similar. Instead we define anexport
function to force the compiler into thinking the function must be compiled. Note this means you are probably going to have to usezig build-obj
instead ofzig build-exe
orzig run
. - The third step is to define a skeleton panic handler to override the more functional default.
Here is a reduction boiler plate. Note little tricks like using easily seen variable names or literal values can help. For example, a
is a difficult variable to grep for. And the value 99
is much easier to find than 0
:
export fn entry() void {
var hello: usize = 99;
}
pub fn panic(msg: []const u8, error_return_trace: ?*@import("std").builtin.StackTrace) noreturn {
@breakpoint();
unreachable;
}
and search for fn entry
(you'll see it twice because zig first produces "IR0" from source then analyzes IR0 and produces IR). Either or both may be of interest depending on the issue at hand:
zig build-obj reduction.zig --verbose-air |& less
see https://github.com/ziglang/zig/issues/208#issuecomment-393777148
When compiling without -O2
or -O3
, Zig infers Debug Mode. Zig passes -fsanitize=undefined -fsanitize-trap=undefined
to Clang in this mode. This causes Undefined Behavior to cause an Illegal Instruction. You can then run the code in a debugger and figure out why UB is being invoked.
From the UBSAN docs:
UndefinedBehaviorSanitizer is not expected to produce false positives. If you see one, look again; most likely it is a true positive!
However you can suppress UBSAN. Here is how to affect the build mode that Zig selects for C code:
-
-O2
or-O3
: ReleaseFast -
-O2
or-O3
and-fsanitize=undefined
: ReleaseSafe -
-Os
: ReleaseSmall -
-Og
or no optimization flags: Debug
You can also pass -fno-sanitize=undefined
.
zig build run -- one two three
const run_cmd = exe.run();
if (b.args) |args| run_cmd.addArgs(args);
zig build run -Dhello=false
zig build run -Dhello=true
exe.addBuildOption(bool, "hello", hello);
const std = @import("std");
pub fn main() anyerror!void {
const options = @import("build_options");
if (options.hello) {
std.debug.warn("Hello.\n", .{});
} else {
std.debug.warn("All your base are belong to us.\n", .{});
}
}
First, some basics:
- Zig packages are just Zig source trees, requiring a root file.
- Packages are imported with
@import("package-name")
without the.zig
extension. - Packages have the same visibility rules as other source files, they just don't need to be in a relative directory to your own source tree.
- Each package has their own dependencies, so packages might use a same-named package that is backed by different source files.
This example shows how to add two packages "first" and "second", where "second" depends on a package "first", which is different:
.
├── first /path/to/one/first.zig
└── second /path/to/second.zig
└── first /path/to/a/different/first.zig
zig build-exe \
--pkg-begin \
first \
/path/to/one/first.zig \
--pkg-end \
--pkg-begin \
second \
/path/to/second.zig \
--pkg-begin \
first \
/path/to/a/different/first.zig \
--pkg-end \
--pkg-end
const exe = b.addExecutable(...);
// simple package
exe.addPackage(std.build.Pkg {
.name = "first",
.path = "/path/to/one/first.zig",
});
// with deps
exe.addPackage(std.build.Pkg {
.name = "second",
.path = "/path/to/second.zig",
.dependencies = &[_]std.build.Pkg {
std.build.Pkg {
.name = "first",
.path = "/path/to/a/different/first.zig",
},
}
});
Why am I seeing hard errors that a framework could not be found when cross-compiling to macOS from a different host OS?
Zig, and zig ld
in particular, has the ability to link against frameworks when cross-compiling, however, a copy of the Darwin's sysroot (also called the SDK) is not bundled with Zig's installation. Therefore, it is necessary to provide the sysroot yourself, and provide Zig with valid paths to the sysroot and any search directories that may be required.
A typical invocation when linking against Foundation
framework may look as follows,
$ zig cc -target aarch64-macos --sysroot=/home/kubkon/macos-SDK -I/usr/include -L/usr/lib -F/System/Library/Frameworks -framework Foundation -o hello hello.c
error: file exists in multiple modules
If you see this, it means that Zig was asked to use a file in multiple modules at the same time. This is undesirable because it would have the following consequences:
- There would be multiple instances of global variables and thread-local variables. That is very likely a bug, or at least suffer from subtle performance problems.
- All functions would be duplicated in each package the file is used in, which bloats the executable, can cause performance issues, and may cause bugs if functions are compared for equality.
- Any exported functions or variables would collide with themselves, causing linker errors.
It's almost always a mistake for the same file to exist in multiple modules simultaneously.
note: posted by mlugg0 on reddit
Here is an objective summary of the status of async/await as of Zig 0.12.0.
The async/await functionality of Zig was originally implemented in the "bootstrap compiler" (written in C++). Zig 0.10.0 began the migration to the "self-hosted compiler" (written in Zig), and Zig 0.11.0 finished this transition by removing the legacy codebase entirely (and hence the option to use the bootstrap compiler). Throughout this process, async/await was not implemented in the self-hosted compiler. Originally, this was just to push the project forward: async/await was not deemed important enough to block releases which otherwise made huge progress in many areas. However, after Andrew (and, in fact, myself) looked at implementing async/await in the self-hosted compiler, we began to consider some issues with it.
The implementation of async/await in the bootstrap compiler contained many bugs (although admittedly most parts of that compiler did), and was of experimental quality at best. Over the years, several issues with the feature became apparent:
-
LLVM struggles to optimize async functions. We can lower Zig async to LLVM IR in two ways: using LLVM coroutines, or a manual lowering where an async function becomes a switch statement which dispatches to the code following a suspension point. Both of these lowerings were tried in the bootstrap compiler. LLVM coroutines are just bad -- they optimize poorly, explode compile times, and add implicit calls to
malloc
. Manual lowering is better, but LLVM seriously struggles to optimize it (particularly in threaded builds, due to an atomic flag used to prevent races between a frame and its awaiter). This would make async/await unusable in high-performance projects. -
LLVM does not allow us to store the function frame off of the call stack. This means we need to spill values into a function's
@Frame
manually, storing the real function frame on the normal call stack. Not only does this hurt optimizability (LLVM assumes that the spills we specify have to happen!), it also makes safe recursion -- a language goal we would like to achieve -- completely impossible. -
Debuggers do not work on async code. There isn't much to say here: debuggers just do not have the functionality required to allow proper debugging of async functions.
-
The cancellation problem. Because Zig doesn't have RAII, we need a concise way to cancel an in-flight frame and clean it up if the caller errors. This is currently an unsolved problem.
With everything listed out, it becomes obvious that this language feature has many problems. Ultimately, if we can't solve all of these issues, async/await will not be a part of Zig; the complexity cost is just too high for a flawed feature. Some Zig core team members believe that async/await should be removed regardless of the resolution of the above issues, due to the complexity it brings. However, if it is to return, the following are all necessary preconditions:
-
One or more self-hosted code generation backends becomes competitive with LLVM and optimizes async functions effectively. This would allow async/await to be used in serious software. It would also solve the
@Frame
issue, since controlling code generation lets us do things that LLVM simply does not support. -
A third-party debugger -- existing or new -- gains the ability to debug async functions. If this happens, it'll almost certainly be a new debugger written in Zig, for Zig. Of course, this would require a certain degree of cooperation from the compiler itself.
-
We solve cancellation. The issue I linked before tracks this.
Personally, I would definitely like to see async/await return to the language. If it were to be well-supported and optimize properly, there would be some key uses for it within the compiler itself! However, we aren't just going to throw it in carelessly to tick a box. If Zig is to support async functions, it will do so properly.