feat: no-std support #455

huitseeker · 2025-10-21T10:04:38Z

Adds support for no-std.

This is a slight modernization of the miette fork by @bitwalker mentioned in #443, found at bitwalker/miette@main...no-std

All the credit for the contribution goes to @bitwalker, I'm just doing janitorial work around the edges.

This commit finishes full no_std support to miette.

…implementation - Add dedicated no-std CI job that builds for wasm32-unknown-unknown target - Fix Infallible StdError implementation for no-std environments The CI job validates: 1. Core no-std functionality (no default features) 2. fancy-no-syscall feature set compatible with no-std environments

miette-derive/src/code.rs

cgettys-microsoft · 2025-11-14T23:27:58Z

src/error.rs

        }
    }

+    #[cfg(not(feature = "std"))]


Curiosity: why not a different struct, instead of using one in two configs.
If the answer is "won't compile", then this is infectious / not additive, no?

cgettys-microsoft · 2025-11-14T23:28:39Z

src/eyreish/context.rs

    }

    impl Diag for Report {
+        #[cfg_attr(track_caller, track_caller)]


is track_caller needed for the no_std implementation, or is that a separate change?

src/eyreish/error.rs

cgettys-microsoft · 2025-11-14T23:30:23Z

src/eyreish/into_diagnostic.rs

-use std::{error::Error, fmt::Display};
+extern crate alloc;
+
+use crate::StdError as Error;


why re-name StdError to Error here, but not elsewehre?

cgettys-microsoft · 2025-11-14T23:31:18Z

src/eyreish/kind.rs

 //     let error = $msg;
 //     (&error).miette_kind().new(error)

+#[cfg(not(feature = "std"))]


can all these repeated declarations live in a single imported prelude instead, or nah?

cgettys-microsoft · 2025-11-14T23:35:36Z

src/eyreish/mod.rs

-    let hook = HOOK.get_or_init(|| Box::new(get_default_printer)).as_ref();
+pub(crate) fn capture_handler_with_location(
+    error: &(dyn Diagnostic + 'static),
+) -> Box<dyn ReportHandler> {


Why the Once? can't it just be OnceLock?
I'm having a hard time following what this change is achieving, which means more comments might be needed here. I don't see why two different once's are needed.

OnceLock isn't available in no_std, hence spin::Once

cgettys-microsoft · 2025-11-14T23:36:00Z

src/eyreish/wrapper.rs

    fn source(&self) -> Option<&(dyn StdError + 'static)> {
        self.0.source()
    }
-


This is a breaking change, no?

These have been deprecated since rust 1.42/1.33.

cgettys-microsoft · 2025-11-14T23:36:42Z

src/eyreish/wrapper.rs

+        boxed::Box,
+        string::{String, ToString},
+    };
+    #[cfg(feature = "fancy")]


shouldn't many of these still be alloc:: qualified types instead, in case someone wants to make fancy no-std too in future?

Fair point for future-proofing. Right now the fancy features all require std (terminal detection, env vars, etc.) so these std imports are fine. If someone wanted no_std_fancy later they'd need to rework the syscall detection anyway. Left it as-is to keep the diff smaller but happy to change if you feel strongly.

src/handler.rs

src/handlers/theme.rs

cgettys-microsoft · 2025-11-15T00:02:49Z

src/highlighters/mod.rs

-                Self(Arc::new(SyntectHighlighter::default()))
+        #[cfg(all(feature = "syntect-highlighter", not(feature = "fancy-no-syscall")))]
+        {
+            use std::io::IsTerminal;


It looks like there's some deduplication that should be done vs src/handlers/theme.rs
It's already pretty complicated and presumably the two should be kept in sync.
Maybe a helper function? Ideally that memo-izes it to avoid many redundant syscalls if necessary

Note: the double or triple negation in the old version was making my head hurt, so I made it exhaustive instead

enum StyleDetection { NoColor, Color } fn detect_color_support() -> { #[cfg(any(feature = "fancy-no-syscall", not(feature = "std")))] { StyleDetection::NoColor } #[cfg(all(not(feature = "fancy-no-syscall"), feature = "std"))] { match std::env::var("NO_COLOR") { if !std::io::stdout().is_terminal() || !std::io::stderr().is_terminal() => { // TODO: address todo by making this StyleDetection::NoColor? //TODO: should use ANSI styling instead of 24-bit truecolor here StyleDetection::Color } Ok(string) => { if string == "0" { // NO_COLOR = "0" means enable color StyleDetection::Color } else { // any other value means disable color. StyleDetection::NoColor } } // This could be NotPresent (color since disablement env var not set) or NotUnicode (hopefully never happens, but we'llt reat it the same) Err(_) => StyleDetection::Color } } }

Thought about this. The two defaults have different return types (GraphicalTheme vs MietteHighlighter) so a shared helper would need callbacks or an enum. The duplicated logic is small (just the NO_COLOR check) and both places already have cfg guards. A helper could work but might add more complexity than it saves. Open to ideas if you see a clean way to do it.

cgettys-microsoft · 2025-11-15T00:06:13Z

src/lib.rs

@@ -1,6 +1,8 @@
+#![no_std]
 #![deny(missing_docs, missing_debug_implementations, nonstandard_style)]


If this is going to be merged, I think at least some of these should probably be enabled?

![deny(clippy::alloc_instead_of_core,clippy::std_instead_of_core,clippy::std_instead_of_alloc)]

src/lib.rs

src/protocol.rs

cgettys-microsoft · 2025-11-15T00:09:52Z

tests/test_boxed.rs


    let error = MyError {
-        source: io::Error::new(io::ErrorKind::Other, "oh no!!!!"),
+        source: io::Error::other("oh no!!!!"),


Might be worth pulling all the io::Error changes out into another PR to keep the diff down, but idk.

Cargo.toml

cgettys-microsoft · 2025-11-15T00:12:01Z

Cargo.toml

-unicode-width = "0.2.0"
+unicode-width = { version = "0.2.0", default-features = false }
 cfg-if = "1.0.0"
+spin = { version = "0.9", default-features = false, features = ["mutex", "spin_mutex", "lazy"] }


Doesn't seem like it should be a dependency for std environments. I see the simplicity benefit, but adding a dependency for consumers for whom OnceLock is an option, doesn't seem like it makes sense.
Also, do you need all those features?

spin is optional = true and only activated when not(feature = "std") (see the cfg guards in src/eyreish/mod.rs). Std users don't pull in spin at all. The minimal features (once only) are intentional to keep the dependency small for no_std users who need it.

Cargo.toml

build.rs

cgettys-microsoft · 2025-11-15T00:20:12Z

To be clear, my comments are just my opinion - I'm neither the author nor am I a maintainer of this library, I'm just a person who has submitted a few contributions previously and was curious. Hope the feedback is helpful.

- Fixed compilation errors in no-std mode (TestError Debug derive, ToString imports) - Addressed consistency issues: Option qualification, StdError naming, Borrow impl - Fixed feature flag logic for supports_color function with proper priority ordering - Added conditional compilation for std-dependent features in panic.rs, handler.rs, theme.rs - Made syntect-highlighter feature explicitly require std - Consolidated duplicate Borrow implementations using core::borrow::Borrow

- Add #[allow(unused_assignments)] to test structs/enums with unused fields - Fix Miri test failures caused by derive macro feature interactions - Miri tests now pass: 52 passed, 0 failed, 6 ignored - Issue occurs when fancy feature changes how derive macros process fields - Suppression is appropriate as fields are only unused under specific feature combinations

Cargo.toml

src/handler.rs

cgettys-microsoft · 2025-11-17T02:01:54Z

src/panic.rs

+#[cfg(feature = "std")]
 use std::boxed::Box;
+#[cfg(feature = "std")]
 use std::{


maybe make the whole module conditional? I don't think a fake no-op set_panic_hook should be provided when std is not set, it's correct to not have it there and will not break any program that depends on miette with std support feature

- Remove rustc_version build dependency that was no longer needed - Remove nightly detection since nightly cfg flag wasn't used anywhere - Simplify build.rs to only set track_caller cfg flag

- Use core::error::Error directly (MSRV 1.82 supports it) - Remove build.rs (track_caller is always available with MSRV 1.82) - Simplify feature detection in handler.rs syscall module - Add clippy lints to catch std/core/alloc misuse - Remove custom StdError trait, re-export core::error::Error - Add __alloc module for derive macro compatibility - Clean up verbose comments - Remove extern crate alloc from tests (not needed with std) - Fix derive macros to use miette::__alloc paths

- spin is now optional, only needed for no_std (provides Once) - std builds use OnceLock, avoiding unnecessary dependency - Restore unicode-width default features (keeps CJK support)

miette-derive/src/code.rs

huitseeker · 2026-01-14T20:53:05Z

src/highlighters/mod.rs

-                Self(Arc::new(SyntectHighlighter::default()))
+        #[cfg(all(feature = "syntect-highlighter", not(feature = "fancy-no-syscall")))]
+        {
+            use std::io::IsTerminal;


Thought about this. The two defaults have different return types (GraphicalTheme vs MietteHighlighter) so a shared helper would need callbacks or an enum. The duplicated logic is small (just the NO_COLOR check) and both places already have cfg guards. A helper could work but might add more complexity than it saves. Open to ideas if you see a clean way to do it.

huitseeker · 2026-01-14T20:55:07Z

src/eyreish/wrapper.rs

+        boxed::Box,
+        string::{String, ToString},
+    };
+    #[cfg(feature = "fancy")]


Fair point for future-proofing. Right now the fancy features all require std (terminal detection, env vars, etc.) so these std imports are fine. If someone wanted no_std_fancy later they'd need to rework the syscall detection anyway. Left it as-is to keep the diff smaller but happy to change if you feel strongly.

Cargo.toml

src/handler.rs

huitseeker · 2026-01-14T21:14:09Z

src/eyreish/mod.rs

-    let hook = HOOK.get_or_init(|| Box::new(get_default_printer)).as_ref();
+pub(crate) fn capture_handler_with_location(
+    error: &(dyn Diagnostic + 'static),
+) -> Box<dyn ReportHandler> {


OnceLock isn't available in no_std, hence spin::Once

huitseeker · 2026-01-14T21:20:56Z

src/highlighters/mod.rs

+        #[cfg(feature = "syntect-highlighter")]
+        {
+            use std::io::IsTerminal;
+            match std::env::var("NO_COLOR") {


The two defaults have different return types (GraphicalTheme vs MietteHighlighter) so a shared helper would need callbacks or an enum. The duplicated logic is small (just the NO_COLOR check) and both places already have cfg guards. A helper could work but might add more complexity than it saves. Open to ideas if you see a clean way to do it.

huitseeker · 2026-01-14T21:23:05Z

Cargo.toml

-unicode-width = "0.2.0"
+unicode-width = { version = "0.2.0", default-features = false }
 cfg-if = "1.0.0"
+spin = { version = "0.9", default-features = false, features = ["mutex", "spin_mutex", "lazy"] }


spin is optional = true and only activated when not(feature = "std") (see the cfg guards in src/eyreish/mod.rs). Std users don't pull in spin at all. The minimal features (once only) are intentional to keep the dependency small for no_std users who need it.

- Make spin a required dependency for no_std hook support (OnceLock unavailable) - Simplify cfg guards: not(std) instead of not(std) && spin - Change clippy alloc/std lints from warn to deny - Remove no_std stub for from_current_location (compile-time vs runtime error) - Fix is_graphical cfg logic for fancy-base without fancy-no-backtrace - Add cargo-hack feature powerset CI job to verify all 56 combinations

bitwalker and others added 3 commits October 18, 2025 07:25

fix: support no-std environments

243ea53

feat(no_std): Implement comprehensive no_std support

79fbd0a

This commit finishes full no_std support to miette.

huitseeker force-pushed the no-std branch from d60ac69 to 18c217f Compare October 21, 2025 11:19

huitseeker marked this pull request as ready for review October 21, 2025 11:19