fix(form_fit): let closing punctuation after a space start a line

CHANGELOG.md

@@ -102,6 +102,9 @@
 
 - FormFit과 쪽 수 추정에서 `!`의 반 em을 공백처럼 배치 단위(1/1800인치)로 내림한다. 9·11·13pt처럼 글자 크기의 4분의 1이
   홀수인 크기에서 `!`가 4 넓게 잡혀, 그 뒤 탭 폭이 한/글보다 4 좁았다.
+- FormFit과 쪽 수 추정이 공백 뒤의 닫는 문장부호(`!`·`?`·`)`·`%`·`」`·`·` 등 줄 머리에 오지 않는 문자 모두)로 줄을
+  시작한다. 한/글이 그렇게 놓는다. 전에는 말줄임표(`…`)만 그랬고, 다른 문자는 공백 앞 낱말까지 다음 줄로 넘겨 한 줄에
+  낱말 하나가 덜 들어갔다. 앞 글자에 붙은 문장부호는 전처럼 그 글자와 함께 다음 줄로 간다.
 - `.hwp`를 HWPX로 읽을 때 변경 추적 설정(`hh:trackchageConfig@flags`)을 설정 레코드의 첫 값 그대로 쓴다. 한컴오피스 2024가
   쓰는 값이다(대개 56). 전에는 레코드가 있으면 늘 4를 썼다. `.hwp`로 쓸 때도 HWPX의 flags를 그 자리에 쓴다(없으면 56).
 - 메모 필드(`notes.add_memo`)와 `.hwp`에서 읽은 필드의 `hp:fieldBegin`에 한/글처럼 `metaTag=""`를 단다(메일 머지

src/hwpx/form_fit/measure.py

@@ -129,8 +129,8 @@
 _EMOJI_BMP = frozenset("✅❌❎❗❓❔❕➕➖➗⏰⏳⛔✋✨⭐⬛⬜")
 #: The variation selectors and the zero-width joiner: no advance.
 _ZERO_WIDTH = frozenset("\ufe0e\ufe0f\u200d")
-#: Closing punctuation that never starts a line; it moves down with the
-#: character before it; but the ellipsis starts a line after a space.
+#: Closing punctuation that never starts a line right after a character: it
+#: moves down with that character. After a space it may start a line.
 _NO_LINE_START = frozenset("!%),.:;?]}¢°’”‰′″℃〉》」』】〕…·、。")
 #: Opening punctuation that never ends a line.
 _NO_LINE_END = frozenset("([{‘“〈《「『【〔")
@@ -1608,9 +1608,9 @@ def hancom_line_starts(
     do. Once the line's text reaches the margin (its 자간 included), its spaces
     hang and the next word starts the next line however far the spaces could
     shrink: only the word that crosses the margin shrinks them. The line then
-    ends at the last break opportunity that fits — never before a closing or
-    after an opening punctuation mark — or mid-word when no opportunity is
-    left. *sizes*, when given, holds each character's size in pt (runs of
+    ends at the last break opportunity that fits — never before a closing
+    punctuation mark attached to the character before it, nor after an opening
+    one — or mid-word when no opportunity is left. *sizes*, when given, holds each character's size in pt (runs of
     several sizes): every character, and every space that shrinks, then
     takes its own size.
     *styles*, when given, holds each character's own style (runs of several
@@ -1678,7 +1678,7 @@ def hancom_line_starts(
             options = [
                 index for index in breaks
                 if start < index <= end
-                and (text[index] not in _NO_LINE_START or text[index] == "…" and text[index - 1] in _HANGING_SPACES)
+                and (text[index] not in _NO_LINE_START or text[index - 1] in _HANGING_SPACES)
                 and text[index - 1] not in _NO_LINE_END
             ]
             start = max(options) if options else end

tests/test_layout_page_estimate.py

@@ -212,6 +212,10 @@
     # Tables of short sentences, one cell holding "… " among its words: Hancom starts a line with the ellipsis after
     # a space (attached to the character before it, it keeps to that one).
     "pages_ellipsis_starts_a_cell_line": 4,
+    # Paragraphs of sentences among which " 」" or " %" stands after a space: Hancom starts a line with the mark
+    # there, as with any closing punctuation after a space (attached to a character, it moves down with it).
+    "pages_closing_bracket_after_a_space_starts_a_line": 2,
+    "pages_percent_after_a_space_starts_a_line": 2,
     "pages_no_break_spaces": 2,  # rows of "가나" and a no-break space: each is half an em and keeps the row
                                   # one word
     "pages_typed_ideographic_spaces": 5,  # cells of "가나다라마" and a U+3000 typed in the text, a full em
@@ -1079,6 +1083,26 @@ def test_without_line_caches_formfit_breaks_the_lines_the_same(name: str) -> Non
     _assert_like_hancom(estimate_pages(_without_caches(data)), data, HANCOM_PAGES[name])
 
 
+@pytest.mark.parametrize("name", ["pages_closing_bracket_after_a_space_starts_a_line",
+                                  "pages_percent_after_a_space_starts_a_line"])
+def test_a_closing_mark_after_a_space_may_start_a_line(name: str) -> None:
+    # The line count alone does not show where a line starts: each paragraph's lines start where Hancom's do.
+    root = HwpxDocument.open(io.BytesIO((FIXTURES / f"{name}.hwpx").read_bytes()))._root
+    measure = page_layout._Measure(root)
+    page = page_layout._page(root.sections[0].element)
+
+    for paragraph in root.sections[0].element.findall(f"{HP}p")[1:]:
+        runs = paragraph.findall(f"{HP}run")
+        size, refs, _ = page_layout._text_size(measure, runs)
+        style = measure.style(paragraph.get("paraPrIDRef"), refs, paragraph)
+        shape = measure.shape(paragraph.get("paraPrIDRef"))
+        widths = page_layout._indented(page.column_width - shape.left - shape.right, shape, style)
+
+        starts = measure.line_starts(page_layout._run_text(runs), widths, size, style)
+
+        assert starts == [int(segment.get("textpos")) for segment in paragraph.findall(f"{HP}linesegarray/{HP}lineseg")]
+
+
 @pytest.mark.parametrize(
     "name",
     [