texdiff 0.2.1__tar.gz → 0.2.2__tar.gz

{texdiff-0.2.1/src/texdiff.egg-info → texdiff-0.2.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: texdiff
-Version: 0.2.1
+Version: 0.2.2
 Summary: AST-driven semantic diff for LaTeX documents
 Author: GoBobr
 License: MIT License

{texdiff-0.2.1 → texdiff-0.2.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "texdiff"
-version = "0.2.1"
+version = "0.2.2"
 description = "AST-driven semantic diff for LaTeX documents"
 readme = "README.md"
 license = { file = "LICENSE" }

{texdiff-0.2.1 → texdiff-0.2.2}/src/texdiff/__init__.py

@@ -42,4 +42,4 @@ __all__ = [
     "word_diff",
 ]
 
-__version__ = "0.1.0"
+__version__ = "0.2.2"

{texdiff-0.2.1 → texdiff-0.2.2}/src/texdiff/align.py

@@ -834,9 +834,13 @@ def _table_keyed_pair_rescue(edits: "list[Edit]", _pair) -> "list[Edit]":
     by key with a LATER insert's table rows, convert the pair into a
     Modify so the emit layer's inline row markup applies.
 
-    Row keys compare by last path component: HDF-EOS grids get
-    renamed (NPP_Grid_IMG_2D -> VIIRS_Grid_IMG_2D) without touching
-    the data fields.
+    Row keys compare by the FULL first cell: an HDF-EOS grid
+    rename (NPP_Grid_IMG_2D -> VIIRS_Grid_IMG_2D) changes every
+    row's path prefix, and the resulting Modify would render as two
+    wholesale blocks (all rows struck, then all re-added) inside
+    one table - worse than leaving the pair as a clean whole-table
+    retire + reintroduction. Regenerated tables keep their row
+    keys; only wholesale renames are left alone.
     """
     import re
 
@@ -845,8 +849,7 @@ def _table_keyed_pair_rescue(edits: "list[Edit]", _pair) -> "list[Edit]":
             return set()
         keys = set()
         for m in re.finditer(r"^\s*([^&%\n]+?)\s*&", node.text or "", re.M):
-            key = m.group(1).strip().replace("\\_", "_")
-            keys.add(key.rsplit("/", 1)[-1])
+            keys.add(m.group(1).strip().replace("\\_", "_"))
         return keys
 
     dels = [

{texdiff-0.2.1 → texdiff-0.2.2}/src/texdiff/emit.py

@@ -162,7 +162,9 @@ def render(edits: list[Edit], markup: LatexdiffMarkup = LatexdiffMarkup()) -> st
             else:
                 out.append(_wrap_node(edit.old, markup, added=False))
                 out.append(_wrap_node(edit.new, markup, added=True))
-    return _normalize_endmark_colours("".join(out))
+    return _collapse_repeated_hlines(
+        _normalize_endmark_colours("".join(out))
+    )
 
 
 # inline end-marker tokens of longtable HEAD material; NOT preceded
@@ -225,6 +227,55 @@ def _normalize_endmark_colours(text: str) -> str:
     return _ENDMARK_COLOR_RE.sub(_insert_reset, text)
 
 
+# invisible-to-the-typesetter material that can sit between two
+# \hline rules: blank lines, %DIFDELCMD-style comments and the
+# no-op FL markers of flush-left diff regions
+_INVISIBLE_BETWEEN_RULES = (
+    r"(?:[ \t]*%[^\n]*\n?|[ \t]*\n|"
+    r"\\DIF(?:add|del)(?:begin|end)FL\n?|\\DIF(?:add|del)(?:begin|end)\n?)*"
+)
+# two consecutive \hline rules separated only by invisible material
+# (a deleted row's tail rule + the skeleton rule of the material
+# that follows): both draw the SAME grid line - longtable stacks
+# them with zero offset and the boundary prints twice as thick.
+# The pair is CONSUMED and replaced by a single rule (a lookahead
+# replacement would be a no-op: group 1 is identical to the match).
+# Iteration in _collapse_repeated_hlines handles longer runs.
+_REPEATED_HLINE_RE = re.compile(
+    r"\\hline" + _INVISIBLE_BETWEEN_RULES + r"[ \t]*\\hline"
+)
+# same-line adjacency: rules glued to one line by the row split
+# ("\hline\hline") or separated only by spaces/tabs draw on top of
+# each other just the same
+_REPEATED_HLINE_INLINE_RE = re.compile(r"\\hline[ \t]*\\hline")
+
+
+def _collapse_repeated_hlines(text: str) -> str:
+    """Merge grid rules that land on top of each other.
+
+    Row-region emitters keep skeleton ``\\hline``s on both sides of
+    their material: a retired (deleted) row region ends with its
+    trailing rule while the following region's lead already carries
+    one, and blank lines / ``%DIFDELCMD`` comments / no-op ``FL``
+    markers sit between them. Two rules with only invisible
+    material between them draw the same grid line - longtable
+    typesets them stacked at zero offset, printing a border twice
+    as thick as every other rule in the table.
+
+    One rule of each such pair is dropped. Visible material
+    (content, colour declarations) between two rules keeps both -
+    they delimit a real row.
+    """
+    if "\\hline" not in text:
+        return text
+    while True:
+        new = _REPEATED_HLINE_INLINE_RE.sub(r"\\hline", text)
+        new = _REPEATED_HLINE_RE.sub(r"\\hline", new)
+        if new == text:
+            return new
+        text = new
+
+
 def _hoist_retired_tables(edits: list[Edit]) -> list[Edit]:
     """Reorder ``Insert(new table) ... Delete(old table)`` pairs.
 

{texdiff-0.2.1 → texdiff-0.2.2}/src/texdiff/tables.py

@@ -355,12 +355,27 @@ def rows_pair_by_key(old_text: str, new_text: str, min_frac: float = 0.5) -> boo
     generation changes type names, shapes and every attribute), so
     word similarity collapses and the pair looks mutually
     unrecognizable - while the logical row structure (one row per
-    variable: ``lat``, ``lon``, ``crs``, ...) is intact. When at least
-    ``min_frac`` of the smaller side's rows have a first cell that
-    also appears as a first cell on the other side, the table is the
+    variable: ``lat``, ``lon``, ``crs``, ...) is intact. When at
+    least ``min_frac`` of the smaller side's rows have a first cell
+    that also appears as a first cell on the other side, the table is the
     SAME table with edited cells: inline row markup (red struck cell,
     blue replacement right after) reads far better than retiring the
     whole table and reintroducing it blue.
+
+    The FULL key must match - a whole-prefix rename (an HDF-EOS
+    grid renamed ``NPP_Grid_IMG_2D`` -> ``VIIRS_Grid_IMG_2D``)
+    changes every row's path without making the rows unrelated, but
+    the row-level machinery (_row_key_of / _segment_ids) keys on the
+    full first cell: with near-zero full-key overlap the pairing
+    never happens and the inline path degrades to two wholesale
+    blocks (all rows struck, then all rows re-added) in one table.
+    Such near-total renames read better as one clean whole-table
+    retire and one blue reintroduction, which the caller
+    (merge_tables / render_restructured) provides - so they are
+    rejected here even though their LAST path components still
+    match. Regenerated grouped tables keep ~75% full-key overlap
+    (the attribute keys repeat per group) and stay above the gate;
+    renamed-grid tables sit at ~25% or below.
     """
     lo = [
         _first_cell(r)
@@ -377,15 +392,8 @@ def rows_pair_by_key(old_text: str, new_text: str, min_frac: float = 0.5) -> boo
     # structural repeats (header rows repeated via \endhead) must not
     # count twice per side
     lo, ln = set(lo), set(ln)
-    # grid renames (NPP_Grid_IMG_2D -> VIIRS_Grid_IMG_2D) change the
-    # path prefix of every HDF-EOS field row without touching the
-    # data field itself: rows also pair when their last path
-    # component matches
-    def _last(k: str) -> str:
-        return k.rsplit("/", 1)[-1] if "/" in k else k
-    lo_last, ln_last = {_last(k) for k in lo}, {_last(k) for k in ln}
-    shared = len(lo_last & ln_last)
-    return shared >= min_frac * min(len(lo_last), len(ln_last)) and shared >= 2
+    shared = len(lo & ln)
+    return shared >= min_frac * min(len(lo), len(ln)) and shared >= 2
 
 
 _SPEC_RE = re.compile(r"\\begin\{longtable\*?\}\s*\{")

{texdiff-0.2.1 → texdiff-0.2.2/src/texdiff.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: texdiff
-Version: 0.2.1
+Version: 0.2.2
 Summary: AST-driven semantic diff for LaTeX documents
 Author: GoBobr
 License: MIT License