code wiki / (root) / nx_html_extract_links.nx
`nx_html_extract_links.nx` source

↩ module page · 108 lines · 5164 B
1// nx_html_extract_links.nx -- enumerate <a href="..."> links in a
2// parsed HTML document, resolving each href against the page base URL.
3//
4// Arc B2 sibling of nx_html_extract_imgs.  Same shape, same guarantees:
5//   - script/style/textarea/title raw-text bodies are SKIPPED so they
6//     can't pollute the extracted list with literal "<a" tokens-inside-
7//     strings
8//   - URL resolution is RFC 3986 §5.2 via nx_url_resolve
9//   - caller-supplied flat URL buffer + parallel (offset, length)
10//     arrays; substrate-honest truncation
11//
12// This is the primitive the bits-up browser uses to discover
13// navigation targets in a fetched page.
14//
15// nx_safety_envelope:
16//   intended_use:       "Anchor-link enumeration for the bits-up browser."
17//   sil_target:         SIL1
18//   evidence:           [composes_b1_raw_text_mode,
19//                        composes_rfc3986_url_resolver]
20//   hazard_register:    [bug-tape-javascript-url-bypass-via-href,
21//                        bug-tape-data-url-bypass-via-href]
22//   residual_risk:      "Caller is responsible for scheme-allowlisting
23//                        (e.g. reject javascript:, data:, file:);
24//                        substrate emits absolute URLs verbatim."
25//   verdict:            NOT_YET_EVALUATED
26
27import "nx_syscalls.nx"
28import "nx_html_tokenizer.nx"
29import "nx_dom_query.nx"
30import "nx_url_resolve.nx"
31import "nx_url_canon.nx"      // an href/src is HTML TEXT: decode character references BEFORE resolving
32
33// Case-insensitive check that a tag name is exactly "a".
34func _link_name_eq_a(src: *u8, name_off: i64, name_len: i64) -> i64 {
35    if name_len != 1 { return 0 }
36    let c: i64 = src[name_off]
37    if c == 0x61 { return 1 }       // 'a'
38    if c == 0x41 { return 1 }       // 'A'
39    return 0
40}
41
42func nx_html_extract_links(html: *u8, html_len: i64,
43                           base_url: *u8, base_url_len: i64,
44                           url_buf: *u8, url_buf_cap: i64,
45                           offsets: *i64, lengths: *i64,
46                           max_links: i64) -> i64 {
47    let c: *HtmlCursor = sys_mmap(24) as *HtmlCursor
48    nx_html_cursor_init(c, html, html_len)
49    let tok: *HtmlToken = sys_mmap(56) as *HtmlToken
50    let href_attr: *u8 = "href" as *u8
51    let val_off_p: *i64 = sys_mmap(8) as *i64
52    let val_len_p: *i64 = sys_mmap(8) as *i64
53    let out_len_p: *i64 = sys_mmap(8) as *i64
54    // ONE scratch for the whole call, not one per element.
55    let canon_buf: *u8 = sys_mmap(UC_SCRATCH)
56
57    var buf_pos: i64 = 0
58    var count: i64 = 0
59
60    while count < max_links {
61        nx_html_next_token(c, tok)
62        if tok.kind == NX_HTML_TOK_EOF { return count }
63
64        if tok.kind == NX_HTML_TOK_START_TAG {
65            // Raw-text-mode tags: skip their bodies.
66            if nx_html_is_raw_text_tag(html, tok.name_off, tok.name_len) == 1 {
67                let tag_name_off: i64 = tok.name_off
68                let tag_name_len: i64 = tok.name_len
69                nx_html_consume_raw_text(c, html + tag_name_off, tag_name_len, tok)
70            } else {
71                // Regular <a> tag candidate.
72                if _link_name_eq_a(html, tok.name_off, tok.name_len) == 1 {
73                    let has: i64 = nx_dom_find_attr(html, tok.src_off, tok.src_len,
74                                                    href_attr, val_off_p, val_len_p)
75                    if has == 1 {
76                        let rel_off: i64 = val_off_p[0]
77                        let rel_len: i64 = val_len_p[0]
78                        if rel_len > 0 {
79                            let remaining: i64 = url_buf_cap - buf_pos
80                        // `&amp;` in an href is CORRECT html for a single `&`. Copying the
81                        // attribute verbatim produced a param literally named `amp;...`, so
82                        // every multi-param url we harvested was malformed. Canon REFUSES
83                        // (returns 0) rather than truncate, and we then fall back to the raw
84                        // bytes so a url too long for the scratch is still resolved exactly as
85                        // before -- a canonicaliser must never LOSE a link it cannot improve.
86                        var canon_rel: *u8 = html + rel_off
87                        var canon_len: i64 = rel_len
88                        let cl: i64 = nx_url_canon(html + rel_off, rel_len, canon_buf, UC_SCRATCH - 1)
89                        if cl > 0 { canon_rel = canon_buf; canon_len = cl }
90                            let rc: i64 = nx_url_resolve(
91                                base_url, base_url_len,
92                                canon_rel, canon_len,
93                                url_buf + buf_pos, remaining,
94                                out_len_p)
95                            if rc == NX_URL_RESOLVE_OK {
96                                offsets[count] = buf_pos
97                                lengths[count] = out_len_p[0]
98                                buf_pos = buf_pos + out_len_p[0]
99                                count = count + 1
100                            }
101                        }
102                    }
103                }
104            }
105        }
106    }
107    return count
108}
nx_html_extract_links.nx source

`nx_html_extract_links.nx` source
