ElyPDF: the edit tool checks a page's text blocks in one pass (P-04)

crates/ely-pdf-engine/shim/ely_text.c

@@ -29,6 +29,12 @@ typedef struct
 	int *c;
 	int n;
 	int hit;
+	/* Page-wide: each target's block, and hits per block. */
+	int *owner;
+	int *hits;
+	/* The glyph each block last counted, so once each. */
+	int *seen;
+	int glyph;
 } targets;
 
 /* The glyphs of `block` with their user-space origins. */
@@ -58,19 +64,32 @@ static int take_glyph(fz_context *ctx, void *opaque, int *ucs, int ucslen, fz_ma
 {
 	targets *t = opaque;
 	fz_point at = fz_transform_point(fz_transform_point(fz_make_point(0, 0), trm), ctm);
-	int i, k;
+	int i, k, found = 0;
+	t->glyph++;
 	for (i = 0; i < t->n; i++)
 	{
+		int match = 0;
 		if (fabsf(at.x - t->at[i].x) > 0.5f || fabsf(at.y - t->at[i].y) > 0.5f)
 			continue;
-		for (k = 0; k < ucslen; k++)
-			if (ucs[k] == t->c[i] || t->c[i] == FZ_REPLACEMENT_CHARACTER)
-			{
-				t->hit++;
-				return 1;
-			}
+		for (k = 0; k < ucslen && !match; k++)
+			match = ucs[k] == t->c[i] || t->c[i] == FZ_REPLACEMENT_CHARACTER;
+		if (!match)
+			continue;
+		if (!t->owner)
+		{
+			t->hit++;
+			return 1;
+		}
+		/* Every block this glyph matches counts it once. */
+		if (t->seen[t->owner[i]] != t->glyph)
+		{
+			t->seen[t->owner[i]] = t->glyph;
+			t->hits[t->owner[i]]++;
+		}
+		found = 1;
 	}
-	return 0;
+	t->hit += found;
+	return found;
 }
 
 /* Runs the glyph filter over the page; `dry` changes nothing. */
@@ -99,7 +118,7 @@ fz_matrix ely_to_user(fz_context *ctx, pdf_page *page)
 /* Whether all of `block` sits in the page's own content. */
 static int reachable(fz_context *ctx, pdf_document *pdf, pdf_page *page, fz_stext_block *block)
 {
-	targets t = { NULL, NULL, 0, 0 };
+	targets t = { NULL, NULL, 0, 0, NULL, NULL, NULL, 0 };
 	int ok = 0;
 	fz_var(t.at);
 	fz_var(t.c);
@@ -122,6 +141,70 @@ static int reachable(fz_context *ctx, pdf_document *pdf, pdf_page *page, fz_stex
 	return ok;
 }
 
+/* Each text block's glyphs found in page content, one pass. */
+static void reachable_all(fz_context *ctx, pdf_document *pdf, pdf_page *page, fz_stext_page *text, int n, int *ok)
+{
+	targets t = { NULL, NULL, 0, 0, NULL, NULL, NULL, 0 };
+	int *count = NULL, all = 0, b;
+	fz_matrix to_user = ely_to_user(ctx, page);
+	fz_stext_block *block;
+	fz_stext_line *line;
+	fz_stext_char *ch;
+	fz_var(t.at);
+	fz_var(t.c);
+	fz_var(t.owner);
+	fz_var(t.hits);
+	fz_var(t.seen);
+	fz_var(count);
+	pdf_begin_operation(ctx, pdf, "Look at text");
+	fz_try(ctx)
+	{
+		for (block = text->first_block; block; block = block->next)
+			if (block->type == FZ_STEXT_BLOCK_TEXT)
+				for (line = block->u.t.first_line; line; line = line->next)
+					for (ch = line->first_char; ch; ch = ch->next)
+						all++;
+		t.at = fz_malloc_array(ctx, all + 1, fz_point);
+		t.c = fz_malloc_array(ctx, all + 1, int);
+		t.owner = fz_malloc_array(ctx, all + 1, int);
+		t.hits = fz_calloc(ctx, n + 1, sizeof(int));
+		t.seen = fz_calloc(ctx, n + 1, sizeof(int));
+		count = fz_calloc(ctx, n + 1, sizeof(int));
+		b = 0;
+		for (block = text->first_block; block; block = block->next)
+		{
+			if (block->type != FZ_STEXT_BLOCK_TEXT)
+				continue;
+			for (line = block->u.t.first_line; line; line = line->next)
+				for (ch = line->first_char; ch; ch = ch->next)
+					if (!(ch->flags & FZ_STEXT_SYNTHETIC))
+					{
+						t.at[t.n] = fz_transform_point(ch->origin, to_user);
+						t.c[t.n] = ch->c;
+						t.owner[t.n++] = b;
+						count[b]++;
+					}
+			b++;
+		}
+		filter_glyphs(ctx, pdf, page, &t, 1);
+		for (b = 0; b < n; b++)
+			ok[b] = count[b] > 0 && t.hits[b] == count[b];
+	}
+	fz_always(ctx)
+	{
+		pdf_abandon_operation(ctx, pdf);
+		ely_forget_resources(ctx, pdf);
+		fz_free(ctx, t.at);
+		fz_free(ctx, t.c);
+		fz_free(ctx, t.owner);
+		fz_free(ctx, t.hits);
+		fz_free(ctx, t.seen);
+		fz_free(ctx, count);
+	}
+	fz_catch(ctx)
+		fz_rethrow(ctx);
+}
+
 /* The text of `block`, lines on lines. */
 static void block_text(fz_context *ctx, fz_buffer *buf, fz_stext_block *block)
 {
@@ -153,18 +236,24 @@ int ely_text_blocks(fz_context *ctx, fz_document *doc, int number, unsigned char
 	pdf_page *page = NULL;
 	fz_buffer *buf = NULL, *words = NULL;
 	fz_output *out = NULL;
+	int *reached = NULL;
 	fz_var(text);
 	fz_var(page);
 	fz_var(buf);
 	fz_var(words);
 	fz_var(out);
+	fz_var(reached);
 	fz_try(ctx)
 	{
 		pdf_document *pdf = ely_pdf_of(ctx, doc);
 		fz_stext_block *block;
-		int first = 1;
+		int first = 1, n = 0, at = 0;
 		text = read_text(ctx, doc, number);
 		page = pdf_load_page(ctx, pdf, number);
+		for (block = text->first_block; block; block = block->next)
+			n += block->type == FZ_STEXT_BLOCK_TEXT;
+		reached = fz_calloc(ctx, n + 1, sizeof(int));
+		reachable_all(ctx, pdf, page, text, n, reached);
 		buf = fz_new_buffer(ctx, 1024);
 		out = fz_new_output_with_buffer(ctx, buf);
 		fz_write_string(ctx, out, "[");
@@ -175,7 +264,8 @@ int ely_text_blocks(fz_context *ctx, fz_document *doc, int number, unsigned char
 			if (block->type != FZ_STEXT_BLOCK_TEXT)
 				continue;
 			ch = block->u.t.first_line->first_char;
-			editable = upright(block) && reachable(ctx, pdf, page, block);
+			editable = reached[at++];
+			editable = editable && upright(block);
 			fz_drop_buffer(ctx, words);
 			words = NULL;
 			words = fz_new_buffer(ctx, 256);
@@ -198,6 +288,7 @@ int ely_text_blocks(fz_context *ctx, fz_document *doc, int number, unsigned char
 		fz_drop_buffer(ctx, words);
 		fz_drop_page(ctx, (fz_page *)page);
 		fz_drop_stext_page(ctx, text);
+		fz_free(ctx, reached);
 	}
 	fz_catch(ctx)
 		return ely_fail(ctx);

crates/ely-pdf-engine/tests/content.rs

@@ -4,7 +4,7 @@ use std::path::Path;
 
 use ely_pdf_engine::{
     Doc, Worker, Writing, Wrote,
-    sample::{Extras, pdf_with},
+    sample::{Extras, one_page, pdf_with},
 };
 
 fn worker() -> Worker {
@@ -372,3 +372,64 @@ fn a_partial_to_unicode_falls_back_as_reading_does() {
     let doc = worker.open(&out).expect("reopens").doc;
     assert_eq!(page_text(&mut worker, doc), "BA", "saved and read again");
 }
+
+#[test]
+fn a_dense_page_lists_its_blocks_in_one_pass() {
+    let dir = tempfile::tempdir().expect("a temp dir");
+    let mut content = String::new();
+    for col in 0..10 {
+        for row in 0..100 {
+            content.push_str(&format!("BT /F1 2 Tf {} {} Td (Cell {col} {row}) Tj ET\n", 20 + col * 58, 780 - row * 7));
+        }
+    }
+    let font = "<< /Type /Font /Subtype /Type1 /BaseFont /Helvetica >>".to_string();
+    let path = dir.path().join("dense.pdf");
+    std::fs::write(&path, one_page("", " /Resources << /Font << /F1 5 0 R >> >>", &content, &[font])).expect("writes");
+    let mut worker = worker();
+    let doc = worker.open(&path).expect("opens").doc;
+    let started = std::time::Instant::now();
+    let blocks = worker.text_blocks(doc, 0).expect("blocks");
+    let took = started.elapsed();
+    assert_eq!((blocks.len(), blocks.iter().filter(|b| b.editable).count()), (1000, 1000));
+    // Per block took 1.5 s here; one pass 0.1 s.
+    assert!(took < std::time::Duration::from_millis(750), "{took:?}");
+}
+
+/// One page: `content`, font F1, and form Fm1 drawing `form`.
+fn layered(dir: &Path, content: &str, form: &str) -> (Worker, Doc) {
+    let objects = [
+        "<< /Type /Font /Subtype /Type1 /BaseFont /Helvetica >>".to_string(),
+        format!("<< /Type /XObject /Subtype /Form /BBox [0 0 612 792] /Resources << /Font << /F1 5 0 R >> >> /Length {} >>\nstream\n{form}\nendstream", form.len()),
+    ];
+    let page = " /Resources << /Font << /F1 5 0 R >> /XObject << /Fm1 6 0 R >> >>";
+    let path = dir.join("layered.pdf");
+    std::fs::write(&path, one_page("", page, content, &objects)).expect("writes");
+    let mut worker = worker();
+    let doc = worker.open(&path).expect("opens").doc;
+    (worker, doc)
+}
+
+fn editable(worker: &mut Worker, doc: Doc) -> Vec<(String, bool)> {
+    let blocks = worker.text_blocks(doc, 0).expect("blocks");
+    blocks.into_iter().map(|b| (b.text, b.editable)).collect()
+}
+
+#[test]
+fn turned_text_first_leaves_the_later_blocks_their_own_answer() {
+    let dir = tempfile::tempdir().expect("a temp dir");
+    let content = "BT /F1 18 Tf 0 1 -1 0 50 500 Tm (ROTATED) Tj ET q 1 0 0 1 300 400 cm /Fm1 Do Q BT /F1 18 Tf 72 200 Td (OWN) Tj ET";
+    let (mut worker, doc) = layered(dir.path(), content, "BT /F1 12 Tf 0 10 Td (STAMP) Tj ET");
+    let mut got = editable(&mut worker, doc);
+    got.sort();
+    assert_eq!(got, [("OWN".into(), true), ("ROTATED".into(), false), ("STAMP".into(), false)]);
+}
+
+#[test]
+fn blocks_that_share_a_glyph_origin_each_count_it() {
+    let dir = tempfile::tempdir().expect("a temp dir");
+    let content = "q /Fm1 Do Q BT /F1 18 Tf 72 300 Td (BREAK) Tj ET BT /F1 18 Tf 72 700 Td (AB) Tj ET";
+    let (mut worker, doc) = layered(dir.path(), content, "BT /F1 36 Tf 72 700 Td (AX) Tj ET");
+    let got = editable(&mut worker, doc);
+    assert!(got.contains(&("AB".into(), true)), "{got:?}");
+    assert!(got.contains(&("AX".into(), false)), "{got:?}");
+}

tasks/BUGS.md

@@ -36,7 +36,7 @@
 - [x] P-01 文件系统的 `PermissionDenied` 被拼成不存在的文案键，应用崩溃 — worker 的拒绝改为自有错误类型 Refused；R1 测试不跨平台、不看提示；组件库 f834f55 加 Toaster::shown；R2 PASS
 - [x] P-02 每次编辑落地清掉全部图块与缩略图，页面空白、刚画的墨迹消失 — 图块随 R2.6；缩略图在编辑后留到重画；R1、R2 测试未证实际绘制；R3 PASS
 - [x] P-03 “编辑文字和图片”先命中图片对象再命中文字；一点抖动就移动了背景图 — 包住文字块的对象让位于文字，移动须过 3 个窗口像素；R1 漏掉同点的小图；R2 PASS
-- [ ] P-04 `TEXT_BLOCKS` 对每个文字块重跑一遍整页内容过滤，编辑工具长时间无反应
+- [x] P-04 `TEXT_BLOCKS` 对每个文字块重跑一遍整页内容过滤，编辑工具长时间无反应 — 整页一遍干跑判定全部文字块（1000 块 1.5 s→0.1 s）；R1 转向文字后错位、共用原点只计一块；R2 PASS
 - [ ] P-05 连续快画或快点被拒：“文档在此间已改动，未执行”
 - [ ] P-06 默认字体“原字体”在 Type 1/Type 3 字体的文件上失败，只给引擎原话
 - [ ] P-07 含连字的文字块永远不可编辑

tasks/reviews/P-04-r1.md

@@ -0,0 +1,5 @@
+FAIL
+
+crates/ely-pdf-engine/shim/ely_text.c:250 — `at++` 位于 `&&` 右侧，旋转文字让 `upright(block)` 返回 0 后，后续文字块的 `reached` 索引错位，违反 PDF-03 的编辑要求。实测单页依次绘制 18 pt `ROTATED`（矩阵 `0 1 -1 0 50 500`）、位于 `(300,400)` 的 Form XObject 文字 `STAMP`、位于 `(72,200)` 的页面自身横排文字 `OWN`，返回的 editable 为 `[false,true,false]`；界面因此允许选择 STAMP，应用时引擎报 `this text is not in the page's own content`，可正常预览修改的 OWN 则无法选择。改动前返回 `[false,false,true]`，同一探针通过；当前代码重复失败。修复方向：每处理一个文字块都递增索引，使递增独立于横排判定，并加入旋转文字后混排 Form 与页面文字的回归测试。
+
+crates/ely-pdf-engine/shim/ely_text.c:74 — 页级匹配把一次字形命中只计入首个匹配目标的 owner，随后 `return 1`；不同文字块共享原点和字符时，较早块抢占较晚块的计数，使可编辑文字被标为不可编辑。实测先以 Form XObject 在 `(72,700)` 绘制 36 pt `AX`，再在页面 `(72,300)` 绘制 18 pt `BREAK`，最后在页面 `(72,700)` 绘制 18 pt `AB`；三块均横排，AB 的 A 被计入 AX，最终 AB 的 editable 为 false，界面无法选中编辑，违反 PDF-03。改动前同一文件的 AB 为 true，同一探针通过；当前代码重复失败。修复方向：在一次整页干跑中保留各块独立的匹配语义，对同一字形匹配到的每个块分别计数，每个块每个字形至多计一次，并加入共享原点的重叠块回归测试。

tasks/reviews/P-04-r2.md

@@ -0,0 +1 @@
+PASS