mailrs-clean

Email content cleanup primitives for Rust — multi-stage HTML sanitization, tracking-pixel detection, bulk/automated sender heuristics, and quoted-reply splitting. Zero I/O, no async runtime.

Extracted from mailrs so any mail client / inbound pipeline can run an email body through the same battle-tested pipeline that fronts a production server.

What you get

Entry point	What it does
[`clean_email_html`]	strip unsafe elements + tracking pixels + hidden blocks + marketing chrome, then convert to paragraph-aware plain text. Returns a [`CleanResult`] with the text plus flags for tracking / template-heavy / unsubscribe-link presence.
[`detect_bulk_sender`]	RFC 2369 `List-*` header heuristic — true for mailing-list / newsletter traffic.
[`is_automated_sender`]	local-part pattern check (`no-reply@`, `notification@`, …).
[`split_quoted_content`]	separate a fresh reply from quoted ancestry so UIs can collapse old context.

Quick start

use mailrs_clean::{clean_email_html, detect_bulk_sender, split_quoted_content};

let result = clean_email_html(
    r#"<html><body>
        <p>Hi! Big news today.</p>
        <img src="https://tracker.mailchimp.com/pixel.gif" width="1" height="1"/>
        <p style="display:none">spam block</p>
       </body></html>"#,
);
assert!(result.has_tracking_pixel);
// result.text is the cleaned, paragraph-aware plain text

let headers = "From: news@example.com\r\nList-Id: <news.example.com>\r\n";
assert!(detect_bulk_sender(headers));

let (fresh, quoted) = split_quoted_content(
    "Sounds good.\n\nOn Wed wrote:\n> Could we move it to 10am?",
);
// fresh = "Sounds good.", quoted = vec!["Could we move it to 10am?"]

What the HTML cleaner removes

<script>, <style>, <iframe>, <object>, <embed> — unsafe.
<img width=1 height=1> from known tracking domains (Mailchimp, SendGrid, HubSpot, Mailgun, Amazon SES, …) — privacy.
Blocks with display:none, visibility:hidden, opacity:0 — usually hidden spam-keyword harvesting.
Marketing-template chrome: outer <table> wrappers around a single content block, repeated unsubscribe footers.
After cleanup, the residue goes through html2text for paragraph-aware plain-text conversion.

Tracking-domain list is const and exhaustive for the top 18 commercial email-marketing platforms — see lib.rs for the current list.

Performance

benches/clean.rs covers the HTML cleaner across realistic email sizes plus the sender heuristics + quote splitter. Measured with criterion 0.8 on Apple Silicon (M-series), cargo bench, release profile.

Operation	Input size	Median	Throughput
`clean_email_html` (short body)	60 B	~20 µs	—
`clean_email_html` (short marketing)	560 B	~56 µs	~10 MB/s
`clean_email_html` (5 KB marketing)	5.6 KB	~336 µs	~17 MB/s
`clean_email_html` (50 KB worst-case)	56 KB	~2.5 ms	~22 MB/s
`detect_bulk_sender` (with List-Id)	—	~60 ns	—
`is_automated_sender("no-reply@…")`	—	~50 ns	—
`is_automated_sender(human address)`	—	~50 ns	—
`split_quoted_content` (typical reply)	—	~970 ns	—

The cost is dominated by scraper parser construction + tree walk; the per-byte rate is roughly flat across input sizes once parser startup is amortized. For workloads that process every inbound message at SMTP time, this means a 50 KB worst-case marketing email costs ~2.5 ms of CPU on top of the DATA-stage I/O.

Run with cargo bench -p mailrs-clean. See tests/perf_gate.rs for the regression budgets.

What this is NOT

Not a general HTML sanitizer — for that, use ammonia. We make email-specific assumptions (single-document message, no script execution, structured-output text).
Not a parser for MIME bodies — bring your own (mail-parser, lettre, etc.); feed us the text/html part.
Not opinionated about how you score senders — the heuristics return booleans, the policy decisions are yours.

License

Licensed under either of Apache License 2.0 or MIT license at your option.