From 0c51b4fe419f96241e1204daba6ae6414a00aa6d Mon Sep 17 00:00:00 2001
From: link2xt <link2xt@testrun.org>
Date: Mon, 21 Jul 2025 16:27:22 +0000
Subject: [PATCH] docs(STYLE.md): prefer `try_next()` over `next()`

---
 STYLE.md | 21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)

diff --git a/STYLE.md b/STYLE.md
index 537f9c9db..08a01f792 100644
--- a/STYLE.md
+++ b/STYLE.md
@@ -78,6 +78,27 @@ All errors should be handled in one of these ways:
 - With `.log_err().ok()`.
 - Bubbled up with `?`.
 
+When working with [async streams](https://docs.rs/futures/0.3.31/futures/stream/index.html),
+prefer [`try_next`](https://docs.rs/futures/0.3.31/futures/stream/trait.TryStreamExt.html#method.try_next) over `next()`, e.g. do
+```
+while let Some(event) = stream.try_next().await? {
+    todo!();
+}
+```
+instead of
+```
+while let Some(event_res) = stream.next().await {
+    todo!();
+}
+```
+as it allows bubbling up the error early with `?`
+with no way to accidentally skip error processing
+with early `continue` or `break`.
+Some streams reading from a connection
+return infinite number of `Some(Err(_))`
+items when connection breaks and not processing
+errors may result in infinite loop.
+
 `backtrace` feature is enabled for `anyhow` crate
 and `debug = 1` option is set in the test profile.
 This allows to run `RUST_BACKTRACE=1 cargo test`