Draw the Word tables the way the PDF prints them

scripts/sdlc-delivery/convert-to-docx.py

@@ -20,6 +20,7 @@ Usage:
     convert-to-docx.py SOURCE_DIRECTORY DELIVERY_DIRECTORY [BAND_DIRECTORY]
 """
 import json
+import re
 import subprocess
 import sys
 import tempfile
@@ -39,6 +40,8 @@ BODY_TOP_MM = 32
 FOOTER_FROM_BOTTOM_MM = 24.2     # footer rule at 272.8mm on a 297mm page
 BAND_WIDTH_MM = 210
 STANDARD = 'ISO/IEC 29110-4-1:2018'
+RULE_SIZE = '4'                   # .5pt table rule, as printed
+RULE_COLOUR = '595959'            # --brn-line from sdlc-delivery.css
 THAI_FONT = 'Leelawadee UI'       # the face embedded in the PDF as "BRN Thai"
 INK = RGBColor(0x11, 0x11, 0x11)
 
@@ -90,6 +93,78 @@ def strip_reconstructed_furniture(document):
     return removed
 
 
+DECIMAL = re.compile(r'^-?\d+\.\d+$')
+BORDER_EDGES = {'top', 'bottom', 'left', 'right', 'insideH', 'insideV', 'tl2br', 'tr2bl'}
+
+
+def normalise_ooxml(document):
+    """Repair the measurements the converter writes outside the schema.
+
+    Widths, sizes and border weights are whole units in OOXML, and a colour is
+    six hex digits with no leading hash. The converter writes the numbers it
+    measured off the page instead - "5.599999999999909", "#585858" - which Word
+    is free to interpret as it likes, so borders come out at weights nobody
+    asked for.
+    """
+    fixed = 0
+    for element in document.element.body.iter():
+        for name, value in list(element.attrib.items()):
+            local = name.rpartition('}')[2]
+            if DECIMAL.match(value):
+                element.set(name, str(round(float(value))))
+                fixed += 1
+            elif local in ('color', 'fill') and value.startswith('#'):
+                element.set(name, value[1:])
+                fixed += 1
+    return fixed
+
+
+def match_rule_weight(document):
+    """Draw every table rule at the .5pt the PDF is printed with."""
+    edges = 0
+    for borders in document.element.body.iter():
+        if borders.tag not in (qn('w:tblBorders'), qn('w:tcBorders')):
+            continue
+        for edge in borders:
+            if edge.tag.rpartition('}')[2] not in BORDER_EDGES:
+                continue
+            if edge.get(qn('w:val')) in (None, 'nil', 'none'):
+                continue
+            edge.set(qn('w:sz'), RULE_SIZE)
+            edge.set(qn('w:color'), RULE_COLOUR)
+            edges += 1
+    return edges
+
+
+def drop_padding(document):
+    """Remove the blank paragraphs the converter pads each page out with.
+
+    A run of them at the foot of a page is pure whitespace - the section break
+    already ends the page - and a run between two blocks only ever needs one.
+    """
+    body = document.element.body
+    children = list(body)
+    removed = 0
+    index = 0
+
+    while index < len(children):
+        if not is_blank(children[index]):
+            index += 1
+            continue
+        start = index
+        while index < len(children) and is_blank(children[index]):
+            index += 1
+        following = children[index] if index < len(children) else None
+        ends_page = following is None or (
+            following.tag == qn('w:p') and following.find('.//' + qn('w:sectPr')) is not None
+        )
+        for element in children[start + (0 if ends_page else 1):index]:
+            body.remove(element)
+            removed += 1
+
+    return removed
+
+
 R_EMBED = '{http://schemas.openxmlformats.org/officeDocument/2006/relationships}embed'
 LOGO_MIN_PX = 100
 
@@ -179,6 +254,9 @@ def convert(pdf, band, footer_tag):
     document = Document(str(docx))
     strip_reconstructed_furniture(document)
     strip_logo_images(document)
+    drop_padding(document)
+    normalise_ooxml(document)
+    match_rule_weight(document)
     install_letterhead(document, band, footer_tag)
     document.save(str(docx))
     return docx