meatproxy.rsLinks to "Let Me MeatProxy That For You" (https://lmmptfy.com/), made by CT from the Jev Discord: a site like the old LMGTFY that hands a question to a chat LLM. Offered where Jev cannot take the question and an LLM would serve better.
The link format lives here and nowhere else. It was verified on
2026-10-05 by reading the site's own JavaScript (fetched raw): one deep
link per provider, in the URL fragment (so it is never sent to a server),
https://lmmptfy.com/#v3/<provider>?prompt=<encoded>. The site's decoder
re-encodes what it reads and refuses a mismatch, so encode must be
exact: encodeURIComponent, then each of ! ' ( ) * as %XX (upper
case), and a final . ~ _ - as %XX too. The whole fragment may be at
most 7959 characters. A change to the site's format shows up as a
decoder refusing our links; the vectors in the tests came from running
the site's decoder against a JavaScript twin of encode.
The ending also shows what the site's front page looks like when pasted
into Discord or Slack, as a "Provided by" card. The card is built the way
those do: the Worker fetches BASE itself (head), reads the <head>'s
og: tags (extract), and shows the title, description and the card
image the site chooses. Nothing else is ever fetched: the one origin is
BASE; the image is taken only if it is https on that same host, at
most MAX_IMAGE bytes of image/*; the page is read to MAX_HTML bytes
at most; both give up after TIMEOUT_MS. The image is served from our
own /meatproxy/card (image) and never linked from the visitor's
browser, which would hand every visitor's address and browser to the
other site. That route takes nothing from the request: it re-derives the
image address from the kept head, so it is no proxy for anything else.
The head is kept an hour per isolate (a failure is not kept, so an outage
is asked again by the next ending), and a failed or empty read shows the
links without the card, never a broken one. No visitor text is ever in a
request or a log line here.
The parsing and the markup are pure (the tests run natively); only
head and image touch worker.
37use std::cell::RefCell;
The site.
46pub const BASE: &str = "https://lmmptfy.com/";
The one host anything is fetched from.
49const HOST: &str = "lmmptfy.com";
The providers the site has a deep link for: the fragment's name, the text
of the link, and its logo, shuqikhor's pixel icon as vendored (MIT, in
third-party/pixel-icons), embedded unchanged (card::pixel, which the card is drawn from too).
What is offered, ahead of the links.
61pub const LEAD: &str = "Please choose an LLM instead:";
What is said above the card.
64pub const PROVIDED: &str = "Provided by:";
How much of the page is read: the <head> is first.
67const MAX_HTML: usize = 64 * 1024;
The largest card image taken.
69const MAX_IMAGE: usize = 1_500_000;
How long a fetch, body included, may take.
71const TIMEOUT_MS: u64 = 2_500;
How long an isolate keeps a good head.
73const KEPT_MS: f64 = 3_600_000.0;
Sent with every fetch.
75const AGENT: &str = "lmjtfy-preview (+https://lmjtfy.fun)";
Our own address for the card image.
77pub const CARD_PATH: &str = "/meatproxy/card";
The site's own limit on the fragment, # included.
84const SITE_MAX_FRAGMENT: usize = 7959;
Ours, well under the site's, so a provider's longer name or a change in the format cannot push a link over.
88const MAX_FRAGMENT: usize = 7000;
text as the site's decoder expects a prompt.
91pub fn encode(text: &str) -> String { 92 let mut out = String::with_capacity(text.len() * 3); 93 for byte in text.bytes() { 94 match byte { 95 // `encodeURIComponent` leaves these, and `! ' ( ) *`, alone; 96 // the site wants the last five written out. 97 b'A'..=b'Z' | b'a'..=b'z' | b'0'..=b'9' | b'-' | b'_' | b'.' | b'~' => out.push(byte as char), 98 _ => out.push_str(&format!("%{byte:02X}")), 99 } 100 } 101 // A trailing `. ~ _ -` is written out too. 102 if let Some(last @ ('.' | '~' | '_' | '-')) = out.chars().last() { 103 out.pop(); 104 out.push_str(&format!("%{:02X}", last as u8)); 105 } 106 out 107}
The link to provider carrying prompt, or nothing when the prompt is
empty or would make a fragment the site refuses. It is never shortened:
a cut question is another question.
112pub fn link(provider: &str, prompt: &str) -> Option<String> { 113 if prompt.trim().is_empty() { 114 return None; 115 } 116 let fragment = format!("#v3/{provider}?prompt={}", encode(prompt)); 117 debug_assert!(MAX_FRAGMENT < SITE_MAX_FRAGMENT); 118 (fragment.len() <= MAX_FRAGMENT).then(|| format!("{BASE}{fragment}")) 119}
What a page says about itself in its <head>, as untrusted text with
its lengths already capped.
The suggestion, with a link per provider and, if the site's head could be read, the card of its front page; empty when the question cannot be carried in a link.
135pub fn suggestion(question: &str, card: Option<&Head>) -> Markup { 136 let links: Vec<(String, &str, &str)> = 137 PROVIDERS.iter().filter_map(|(provider, name, mark)| link(provider, question).map(|href| (href, *name, *mark))).collect(); 138 html! { 139 @if !links.is_empty() { 140 div .meatproxy { 141 p .why { (LEAD) } 142 p .llms { 143 @for (href, name, mark) in &links { 144 a .llm href=(href) target="_blank" rel="noopener noreferrer" { span .llm-mark aria-hidden="true" { (PreEscaped(*mark)) } span { (name) } } 145 } 146 } 147 @if let Some(head) = card { 148 p .why { (PROVIDED) } 149 a .linkcard href=(BASE) target="_blank" rel="noopener noreferrer" { 150 @if head.image.is_some() { img .linkcard-image src=(CARD_PATH) alt="" loading="lazy"; } 151 span .linkcard-text { 152 @if let Some(site) = &head.site_name { span .linkcard-site { (site) } } 153 span .linkcard-title { (head.title) } 154 @if let Some(description) = &head.description { span .linkcard-description { (description) } } 155 span .linkcard-host { (HOST) } 156 } 157 } 158 } 159 } 160 } 161 } 162}
---- reading a head ------------------------------------------------------
The head of html, the way a link preview reads it: og:title,
og:description (else description), og:site_name and og:image,
else <title>. None if there is no title to show. The image is kept
only if it resolves, against page, to an https address on HOST.
Only the first MAX_HTML bytes are read.
171pub fn extract(html: &str, page: &str) -> Option<Head> { 172 let html = cut(html, MAX_HTML); 173 let mut metas: Vec<(String, String)> = Vec::new(); 174 let mut title = None; 175 let lower = html.to_ascii_lowercase(); 176 let mut at = 0; 177 while let Some(found) = lower[at..].find('<') { 178 let open = at + found; 179 let rest = &lower[open..]; 180 if rest.starts_with("<!--") { 181 at = lower[open + 4..].find("-->").map_or(lower.len(), |end| open + 4 + end + 3); 182 } else if rest.starts_with("</head") || starts_tag(rest, "<body") { 183 break; 184 } else if starts_tag(rest, "<script") || starts_tag(rest, "<style") { 185 let close = if starts_tag(rest, "<script") { "</script" } else { "</style" }; 186 at = lower[open + 1..].find(close).map_or(lower.len(), |end| open + 1 + end + 1); 187 } else if starts_tag(rest, "<meta") { 188 let end = tag_end(&html[open..]); 189 let attributes = attributes(&html[open + 5..open + end]); 190 let get = |name: &str| attributes.iter().find(|(key, _)| key == name).map(|(_, value)| value.clone()); 191 if let (Some(key), Some(content)) = (get("property").or_else(|| get("name")), get("content")) { 192 metas.push((key.trim().to_ascii_lowercase(), decode(&content))); 193 } 194 at = open + end; 195 } else if starts_tag(rest, "<title") && title.is_none() { 196 let begin = open + tag_end(&html[open..]); 197 let end = lower[begin..].find("</title").map_or(lower.len(), |end| begin + end); 198 title = Some(decode(&html[begin.min(end)..end])); 199 at = end; 200 } else { 201 at = open + 1; 202 } 203 } 204 let first = |key: &str| metas.iter().find(|(name, _)| name == key).map(|(_, value)| tidy(value)).filter(|value| !value.is_empty()); 205 let title = first("og:title").or_else(|| title.map(|title| tidy(&title)).filter(|title| !title.is_empty()))?; 206 Some(Head { 207 title: capped(&title, MAX_TITLE), 208 description: first("og:description").or_else(|| first("description")).map(|text| capped(&text, MAX_DESCRIPTION)), 209 site_name: first("og:site_name").map(|text| capped(&text, MAX_SITE)), 210 image: first("og:image").and_then(|image| same_site(&image, page)), 211 }) 212}
html cut to at most max bytes, on a character boundary.
Does rest open the tag name (lower case, with its <)?
Where the tag at the start of from ends, just past its >, with quotes
respected; the end of the text if it never does.
A tag's attributes, names in lower case, values as written (undecoded): double-quoted, single-quoted, bare, or none.
245fn attributes(tag: &str) -> Vec<(String, String)> { 246 let tag = tag.trim_end_matches('>').trim_end_matches('/'); 247 let bytes = tag.as_bytes(); 248 let (mut found, mut at) = (Vec::new(), 0); 249 while at < bytes.len() { 250 while at < bytes.len() && (bytes[at].is_ascii_whitespace() || bytes[at] == b'/') { 251 at += 1; 252 } 253 let begin = at; 254 while at < bytes.len() && !bytes[at].is_ascii_whitespace() && !matches!(bytes[at], b'=' | b'/' | b'>') { 255 at += 1; 256 } 257 let name = tag[begin..at].to_ascii_lowercase(); 258 while at < bytes.len() && bytes[at].is_ascii_whitespace() { 259 at += 1; 260 } 261 let mut value = String::new(); 262 if bytes.get(at) == Some(&b'=') { 263 at += 1; 264 while at < bytes.len() && bytes[at].is_ascii_whitespace() { 265 at += 1; 266 } 267 match bytes.get(at) { 268 Some("e @ (b'"' | b'\'')) => { 269 let end = tag[at + 1..].find(quote as char).map_or(tag.len(), |end| at + 1 + end); 270 value = tag[at + 1..end].to_owned(); 271 at = (end + 1).min(tag.len()); 272 } 273 _ => { 274 let begin = at; 275 while at < bytes.len() && !bytes[at].is_ascii_whitespace() { 276 at += 1; 277 } 278 value = tag[begin..at].to_owned(); 279 } 280 } 281 } 282 if name.is_empty() { 283 at += 1; 284 } else { 285 found.push((name, value)); 286 } 287 } 288 found 289}
text with the character references HTML text and attributes use read:
the five named ones and numbers, decimal and hex. Anything else stays.
293fn decode(text: &str) -> String { 294 let mut out = String::with_capacity(text.len()); 295 let mut rest = text; 296 while let Some(amp) = rest.find('&') { 297 out.push_str(&rest[..amp]); 298 rest = &rest[amp..]; 299 let read = rest.find(';').filter(|end| *end <= 10).and_then(|end| { 300 let name = &rest[1..end]; 301 let c = match name { 302 "amp" => Some('&'), 303 "lt" => Some('<'), 304 "gt" => Some('>'), 305 "quot" => Some('"'), 306 "apos" => Some('\''), 307 _ => name 308 .strip_prefix('#') 309 .and_then(|n| match n.strip_prefix(['x', 'X']) { 310 Some(hex) => u32::from_str_radix(hex, 16).ok(), 311 None => n.parse().ok(), 312 }) 313 .and_then(char::from_u32), 314 }; 315 c.map(|c| (c, end + 1)) 316 }); 317 match read { 318 Some((c, used)) => { 319 out.push(c); 320 rest = &rest[used..]; 321 } 322 None => { 323 out.push('&'); 324 rest = &rest[1..]; 325 } 326 } 327 } 328 out.push_str(rest); 329 out 330}
Runs of whitespace and control characters as one space, trimmed.
At most max characters, with an ellipsis if it was cut.
image as an absolute https address, resolved against page, if and
only if its host is HOST itself: no other host, no userinfo, no port.
350fn same_site(image: &str, page: &str) -> Option<String> { 351 let image = image.trim(); 352 if image.is_empty() || image.chars().any(|c| c.is_control() || c.is_whitespace() || c == '\\') { 353 return None; 354 } 355 let absolute = if let Some(rest) = image.strip_prefix("//") { 356 format!("https://{rest}") 357 } else if image.starts_with("https://") { 358 image.to_owned() 359 } else if image.split('/').next().is_some_and(|first| first.contains(':')) { 360 // Another scheme, or `http`. 361 return None; 362 } else if image.starts_with('/') { 363 format!("{}{image}", origin_of(page)?) 364 } else { 365 let base = page.split(['?', '#']).next()?; 366 let directory = &base[..=base.rfind('/')?]; 367 format!("{directory}{}", image.strip_prefix("./").unwrap_or(image)) 368 }; 369 let rest = absolute.strip_prefix("https://")?; 370 let authority = rest.split(['/', '?', '#']).next()?; 371 (authority.eq_ignore_ascii_case(HOST) && !rest[authority.len()..].split(['?', '#']).next()?.split('/').any(|part| part == ".." || part == ".")).then_some(absolute) 372}
https://host of page.
---- fetching ------------------------------------------------------------
382thread_local! {
The head, and when it was read: only a good one.
384 static HEAD: RefCell<Option<(f64, Head)>> = const { RefCell::new(None) };
The image of that head, by its address, and when it was read.
The head of BASE, kept an hour per isolate. Only a success is kept, so
an outage is asked again by the next ending, and None is shown as no
card. Never fails the answer.
392pub async fn head() -> Option<Head> { 393 let now = js_sys::Date::now(); 394 if let Some(kept) = HEAD.with(|kept| kept.borrow().as_ref().filter(|(at, _)| now - at < KEPT_MS).map(|(_, head)| head.clone())) { 395 return Some(kept); 396 } 397 let (_, body) = get(BASE, MAX_HTML, "text/html").await?; 398 let head = extract(&String::from_utf8_lossy(&body), BASE)?; 399 HEAD.with(|kept| *kept.borrow_mut() = Some((now, head.clone()))); 400 Some(head) 401}
Where the card image is, if the head names one: the only thing the route fetches, so with no head there is nothing to serve.
GET /meatproxy/card: the card image the site's head names, from our own
address. Nothing in the request is read: the address comes from the head,
which is BASE's and only ever an https address on HOST.
412pub async fn image() -> Response<Body> { 413 let gone = |status| Response::builder().status(status).header(header::CACHE_CONTROL, "no-store").body(Body::empty()).expect("static headers are valid"); 414 let head = head().await; 415 let Some(url) = card_address(head.as_ref()) else { return gone(404) }; 416 let url = url.to_owned(); 417 let now = js_sys::Date::now(); 418 let kept = IMAGE.with(|kept| kept.borrow().as_ref().filter(|(at, seen, _, _)| now - at < KEPT_MS && *seen == url).map(|(_, _, bytes, kind)| (bytes.clone(), kind.clone()))); 419 let (bytes, kind) = match kept { 420 Some(kept) => kept, 421 None => { 422 let Some((kind, bytes)) = get(&url, MAX_IMAGE, "image/*").await else { return gone(502) }; 423 IMAGE.with(|kept| *kept.borrow_mut() = Some((now, url, bytes.clone(), kind.clone()))); 424 (bytes, kind) 425 } 426 }; 427 Response::builder() 428 .header(header::CONTENT_TYPE, kind) 429 .header("x-content-type-options", "nosniff") 430 .header(header::CACHE_CONTROL, "public, max-age=3600") 431 .body(Body::from(bytes)) 432 .expect("static headers are valid") 433}
A GET of url (which is BASE or an address same_site passed), its
content type and at most max bytes; None for a redirect, a status
other than 200, a type that is not accept, a body over max (for an
image: an image cut short is no image), or taking over TIMEOUT_MS. The
page is cut at max, which is what is wanted of a head.
440async fn get(url: &str, max: usize, accept: &str) -> Option<(String, Vec<u8>)> { 441 let work = async { 442 let headers = worker::Headers::new(); 443 headers.set("user-agent", AGENT).ok()?; 444 headers.set("accept", accept).ok()?; 445 let mut init = worker::RequestInit::new(); 446 init.with_headers(headers).with_redirect(worker::RequestRedirect::Manual); 447 let request = worker::Request::new_with_init(url, &init).ok()?; 448 let mut response = worker::Fetch::Request(request).send().await.ok()?; 449 if response.status_code() != 200 { 450 return None; 451 } 452 let kind = response.headers().get("content-type").ok().flatten()?.split(';').next()?.trim().to_ascii_lowercase(); 453 let wanted = match accept.strip_suffix("/*") { 454 Some(group) => kind.starts_with(&format!("{group}/")) && kind.len() > group.len() + 1, 455 None => kind == accept, 456 }; 457 if !wanted { 458 return None; 459 } 460 let truncate = accept != "image/*"; 461 let mut body = Vec::new(); 462 let mut stream = response.stream().ok()?; 463 while let Some(chunk) = stream.next().await { 464 body.extend_from_slice(&chunk.ok()?); 465 if body.len() > max { 466 if truncate { 467 body.truncate(max); 468 break; 469 } 470 return None; 471 } 472 } 473 Some((kind, body)) 474 }; 475 futures_util::pin_mut!(work); 476 let timeout = worker::Delay::from(std::time::Duration::from_millis(TIMEOUT_MS)); 477 match futures_util::future::select(work, timeout).await { 478 futures_util::future::Either::Left((done, _)) => done, 479 futures_util::future::Either::Right(_) => None, 480 } 481}
Each from the site's own decoder: the JavaScript twin of encode
round-tripped every one (2026-10-05).
489 const VECTORS: [(&str, &str); 6] = [ 490 ("how many sheep do androids dream of?", "how%20many%20sheep%20do%20androids%20dream%20of%3F"), 491 ("what's 2+2 (really)!", "what%27s%202%2B2%20%28really%29%21"), 492 ("ends with dot.", "ends%20with%20dot%2E"), 493 ("café ünïcode 🐑 ~", "caf%C3%A9%20%C3%BCn%C3%AFcode%20%F0%9F%90%91%20%7E"), 494 ("a_b-", "a_b%2D"), 495 ("100% sure & yes #1 ?x=1", "100%25%20sure%20%26%20yes%20%231%20%3Fx%3D1"), 496 ];
498 #[test] 499 fn the_encoder_matches_the_sites_decoder_vectors() { 500 for (text, encoded) in VECTORS { 501 assert_eq!(encode(text), encoded, "{text}"); 502 } 503 } 504 505 #[test] 506 fn an_encoded_prompt_is_only_what_the_decoder_accepts() { 507 for (text, _) in VECTORS { 508 assert!(encode(text).bytes().all(|b| b.is_ascii_alphanumeric() || b"-_.~%".contains(&b)), "{text}"); 509 assert!(!encode(text).ends_with(['.', '~', '_', '-']), "{text}"); 510 } 511 } 512 513 const PAGE: &str = "https://lmmptfy.com/"; 514 515 fn head_of(html: &str) -> Option<Head> { 516 extract(html, PAGE) 517 } 518 519 fn card() -> Head { 520 Head { 521 title: "Let Me MeatProxy that for <You>".into(), 522 description: Some("A \"description\" & more".into()), 523 site_name: Some("LMMPTFY".into()), 524 image: Some("https://lmmptfy.com/assets/og-card-1200x630-XXvjlRvI.png".into()), 525 } 526 } 527 528 #[test] 529 fn three_links_carry_the_question_in_the_fragment() { 530 let html = suggestion("what's 2+2 (really)!", None).into_string(); 531 for (provider, name, _) in PROVIDERS { 532 let href = format!("https://lmmptfy.com/#v3/{provider}?prompt=what%27s%202%2B2%20%28really%29%21"); 533 assert!(html.contains(&format!(r#"<a class="llm" href="{href}" target="_blank" rel="noopener noreferrer">"#)), "{html}"); 534 assert!(html.contains(&format!("<span>{name}</span>")), "{html}"); 535 } 536 assert!(html.contains("Please choose an LLM instead:"), "{html}"); 537 assert_eq!(html.matches(r#"rel="noopener noreferrer""#).count(), 3); 538 assert_eq!(html.matches("<svg").count(), 3); 539 // No card without a head. 540 assert!(!html.contains("Provided by") && !html.contains("linkcard"), "{html}"); 541 } 542 543 #[test] 544 fn a_card_is_built_from_the_head_and_escaped() { 545 let html = suggestion("why?", Some(&card())).into_string(); 546 assert!(html.contains("Provided by:"), "{html}"); 547 assert!(html.contains(r#"<a class="linkcard" href="https://lmmptfy.com/" target="_blank" rel="noopener noreferrer">"#), "{html}"); 548 // Our own address, never the other site's. 549 assert!(html.contains(r#"<img class="linkcard-image" src="/meatproxy/card" alt="" loading="lazy">"#), "{html}"); 550 assert!(!html.contains("og-card"), "{html}"); 551 assert!(html.contains("Let Me MeatProxy that for <You>"), "{html}"); 552 assert!(html.contains("A "description" & more"), "{html}"); 553 assert!(html.contains(r#"<span class="linkcard-host">lmmptfy.com</span>"#), "{html}"); 554 assert_eq!(html.matches(r#"rel="noopener noreferrer""#).count(), 4); 555 } 556 557 #[test] 558 fn a_card_without_an_image_has_no_image_element() { 559 let head = Head { image: None, description: None, site_name: None, ..card() }; 560 let html = suggestion("why?", Some(&head)).into_string(); 561 assert!(html.contains("linkcard-title") && !html.contains("<img") && !html.contains("linkcard-description"), "{html}"); 562 } 563 564 #[test] 565 fn an_overlong_or_blank_question_gets_no_links() { 566 assert!(suggestion(&"a".repeat(6_900), None).into_string().contains("lmmptfy.com/#v3/claude")); 567 assert_eq!(suggestion(&"a".repeat(7_000), Some(&card())).into_string(), ""); 568 // Two bytes each, six encoded characters each. 569 assert_eq!(suggestion(&"é".repeat(1_200), None).into_string(), ""); 570 assert_eq!(suggestion(" ", Some(&card())).into_string(), ""); 571 } 572 573 #[test] 574 fn the_logos_are_all_there_distinct_and_in_their_upstream_colours() { 575 let marks: Vec<_> = PROVIDERS.iter().map(|(_, _, svg)| *svg).collect(); 576 for (mark, class) in marks.iter().zip(["pixelicon-claude", "pixelicon-chatgpt", "pixelicon-gemini"]) { 577 assert!(mark.starts_with("<svg") && mark.contains("viewBox=\"0 0 9 9\"") && mark.contains(class), "{mark}"); 578 } 579 for (a, b) in [(0, 1), (0, 2), (1, 2)] { 580 assert_ne!(marks[a], marks[b]); 581 } 582 let html = suggestion("why?", None).into_string(); 583 for colour in ["#D97757", "#343341", "#7C72C6", "#637FCE", "#9168C0", "#438DD7", "#1BA1E3"] { 584 assert!(html.contains(&format!("fill=\"{colour}\"")), "{colour}"); 585 } 586 // Not recoloured, and the caption names each. 587 assert!(!html.contains("currentColor")); 588 } 589 590 #[test] 591 fn og_tags_in_any_order_and_quote_style() { 592 let head = head_of( 593 r#"<!doctype html><html><head><META CONTENT='Let Me & MeatProxy' Property="og:title"> 594 <meta name=description content="plain "desc""> 595 <meta content="https://lmmptfy.com/a.png?x=1&y=2" property="og:image" /> 596 <meta property='og:site_name' content='LMMPTFY'></head><body><meta property="og:title" content="late"></body>"#, 597 ) 598 .unwrap(); 599 assert_eq!(head.title, "Let Me & MeatProxy"); 600 assert_eq!(head.description.as_deref(), Some(r#"plain "desc""#)); 601 assert_eq!(head.site_name.as_deref(), Some("LMMPTFY")); 602 assert_eq!(head.image.as_deref(), Some("https://lmmptfy.com/a.png?x=1&y=2")); 603 } 604 605 #[test] 606 fn og_description_beats_the_name_one_and_numeric_entities_decode() { 607 let head = head_of(r#"<head><meta name="description" content="b"><meta property="og:description" content="a 'q' 🐑 &bogus; &"><meta property="og:title" content="t"></head>"#).unwrap(); 608 assert_eq!(head.description.as_deref(), Some("a 'q' 🐑 &bogus; &")); 609 } 610 611 #[test] 612 fn a_missing_og_title_falls_back_to_the_title_element() { 613 let head = head_of("<head><title>\n Hello\n there & you </title><meta property=\"og:image\" content=\"/assets/c.png\"></head>").unwrap(); 614 assert_eq!(head.title, "Hello there & you"); 615 assert_eq!(head.description, None); 616 assert_eq!(head.image.as_deref(), Some("https://lmmptfy.com/assets/c.png")); 617 } 618 619 #[test] 620 fn a_relative_image_is_resolved_against_the_page() { 621 let at = |image: &str| same_site(image, "https://lmmptfy.com/deep/page.html?x#y"); 622 assert_eq!(at("/a.png").as_deref(), Some("https://lmmptfy.com/a.png")); 623 assert_eq!(at("a.png").as_deref(), Some("https://lmmptfy.com/deep/a.png")); 624 assert_eq!(at("./a.png").as_deref(), Some("https://lmmptfy.com/deep/a.png")); 625 assert_eq!(at("//lmmptfy.com/a.png").as_deref(), Some("https://lmmptfy.com/a.png")); 626 assert_eq!(at("https://LMMPTFY.com/a.png").as_deref(), Some("https://LMMPTFY.com/a.png")); 627 } 628 629 #[test] 630 fn an_image_off_the_site_is_dropped_and_the_card_kept() { 631 for image in [ 632 "https://evil.example/a.png", 633 "http://lmmptfy.com/a.png", 634 "//evil.example/a.png", 635 "https://lmmptfy.com.evil.example/a.png", 636 "https://evil.example/https://lmmptfy.com/a.png", 637 "https://lmmptfy.com@evil.example/a.png", 638 "https://evil.example@lmmptfy.com/a.png", 639 "https://lmmptfy.com:8443/a.png", 640 "https://lmmptfy.com/../a.png", 641 "https://lmmptfy.com\\@evil.example/a.png", 642 "data:image/png;base64,AAAA", 643 "javascript:alert(1)", 644 "ftp://lmmptfy.com/a.png", 645 "", 646 ] { 647 let html = format!(r#"<head><meta property="og:title" content="T"><meta property="og:image" content="{}"></head>"#, image.replace('"', """)); 648 let head = head_of(&html).unwrap(); 649 assert_eq!(head.image, None, "{image}"); 650 assert_eq!(head.title, "T"); 651 } 652 } 653 654 #[test] 655 fn a_huge_page_is_read_only_to_the_cap_and_text_is_capped() { 656 let wanted = format!(r#"<head><meta property="og:title" content="{}"><meta property="og:description" content="{}">"#, "t".repeat(500), "d".repeat(5_000)); 657 let late = format!("{wanted}{}<meta property=\"og:site_name\" content=\"too late\"></head>", " ".repeat(MAX_HTML)); 658 let head = head_of(&late).unwrap(); 659 assert_eq!(head.site_name, None); 660 assert_eq!(head.title.chars().count(), MAX_TITLE); 661 assert!(head.title.ends_with('…')); 662 assert_eq!(head.description.unwrap().chars().count(), MAX_DESCRIPTION); 663 // A cut inside a multibyte character does not panic. 664 let wide = format!("<head><title>x</title>{}</head>", "é".repeat(MAX_HTML)); 665 assert_eq!(head_of(&wide).unwrap().title, "x"); 666 } 667 668 #[test] 669 fn comments_scripts_and_the_body_are_not_the_head() { 670 let head = head_of( 671 r#"<head><!-- <meta property="og:title" content="comment"> --><script>var s = '<meta property="og:title" content="script">';</script> 672 <style>/* <title>style</title> */</style><meta property="og:title" content="real"></head>"#, 673 ) 674 .unwrap(); 675 assert_eq!(head.title, "real"); 676 } 677 678 #[test] 679 fn the_route_has_nothing_to_serve_without_a_head_or_an_image() { 680 assert_eq!(card_address(None), None); 681 assert_eq!(card_address(Some(&Head { image: None, ..card() })), None); 682 assert_eq!(card_address(Some(&card())), Some("https://lmmptfy.com/assets/og-card-1200x630-XXvjlRvI.png")); 683 } 684 685 #[test] 686 fn nothing_to_show_is_none() { 687 assert_eq!(head_of(""), None); 688 assert_eq!(head_of("<head><meta charset=utf-8><meta name=viewport content='width=1'></head>"), None); 689 assert_eq!(head_of(r#"<head><meta property="og:image" content="/a.png"><title> </title></head>"#), None); 690 assert_eq!(head_of("<head><meta property=og:title content="), None); 691 assert_eq!(head_of("<meta <<<>>> =\"= <title"), None); 692 }
Needs the network: the site's real front page, read as the Worker
would. Run with cargo test -p lmjtfy --lib -- --ignored real_site --nocapture.
696 #[test] 697 #[ignore] 698 fn real_site_head() { 699 let out = std::process::Command::new("curl").args(["-sSL", "-A", AGENT, "--max-time", "10", BASE]).output().expect("curl"); 700 let html = String::from_utf8_lossy(&out.stdout); 701 let head = extract(&html, BASE).expect("a head"); 702 println!("{head:#?}"); 703 assert!(!head.title.is_empty()); 704 assert!(head.image.as_deref().is_some_and(|image| image.starts_with("https://lmmptfy.com/"))); 705 } 706}