Clean up svgDoc compression

This is based on bungeman's https://github.com/fonttools/fonttools/pull/2627

Previously, an entire `SVG ` table would be marked as compressed if any
of the decoded SVG documents in it were compressed. Then on encoding all
SVG documents would be considered for compression. The XML format had no
means to indicate if compression was desired.

Instead, mark each svgDoc with its compression status. When decoding
mark the svgDoc as compressed if the data was compressed. When encoding
try to compress the svgDoc if it is marked as compressed. In the XML
format the data itself is always uncompressed, but allow an optional
`compressed` boolean attribute (defaults to false) to indicate the
svgDoc should be compressed when encoded.

We also try to make sure that older code that relies on docList containing
sequences of three items (doc, startGID, endGID) will continue to work
without modification.

Lib/fontTools/subset/svg.py

@@ -14,6 +14,7 @@ except ModuleNotFoundError:
 
 from fontTools import ttLib
 from fontTools.subset.util import _add_method
+from fontTools.ttLib.tables.S_V_G_ import SVGDocument
 
 
 __all__ = ["subset_glyphs"]
@@ -201,10 +202,12 @@ def subset_glyphs(self, s) -> bool:
     # map from original to new glyph indices (after subsetting)
     glyph_index_map: Dict[int, int] = s.glyph_index_map
 
-    new_docs: List[Tuple[bytes, int, int]] = []
+    new_docs: List[SVGDocument] = []
-    for doc, start, end in self.docList:
+    for doc in self.docList:
 
-        glyphs = {glyph_order[i] for i in range(start, end + 1)}.intersection(s.glyphs)
+        glyphs = {
+            glyph_order[i] for i in range(doc.startGlyphID, doc.endGlyphID + 1)
+        }.intersection(s.glyphs)
         if not glyphs:
             # no intersection: we can drop the whole record
             continue
@@ -212,7 +215,7 @@ def subset_glyphs(self, s) -> bool:
         svg = etree.fromstring(
             # encode because fromstring dislikes xml encoding decl if input is str.
             # SVG xml encoding must be utf-8 as per OT spec.
-            doc.encode("utf-8"),
+            doc.data.encode("utf-8"),
             parser=etree.XMLParser(
                 # Disable libxml2 security restrictions to support very deep trees.
                 # Without this we would get an error like this:
@@ -241,7 +244,7 @@ def subset_glyphs(self, s) -> bool:
 
         new_gids = (glyph_index_map[i] for i in gids)
         for start, end in ranges(new_gids):
-            new_docs.append((new_doc, start, end))
+            new_docs.append(SVGDocument(new_doc, start, end, doc.compressed))
 
     self.docList = new_docs
 

Lib/fontTools/ttLib/tables/S_V_G_.py

@@ -17,9 +17,11 @@ The XML format is:
 	</SVG>
 """
 
-from fontTools.misc.textTools import bytesjoin, strjoin, tobytes, tostr
+from fontTools.misc.textTools import bytesjoin, safeEval, strjoin, tobytes, tostr
 from fontTools.misc import sstruct
 from . import DefaultTable
+from collections.abc import Sequence
+from dataclasses import dataclass, astuple
 from io import BytesIO
 import struct
 import logging
@@ -75,15 +77,18 @@ class table_S_V_G_(DefaultTable.DefaultTable):
 				start = entry.svgDocOffset + subTableStart
 				end = start + entry.svgDocLength
 				doc = data[start:end]
+				compressed = False
 				if doc.startswith(b"\x1f\x8b"):
 					import gzip
 					bytesIO = BytesIO(doc)
 					with gzip.GzipFile(None, "r", fileobj=bytesIO) as gunzipper:
 						doc = gunzipper.read()
-					self.compressed = True
 					del bytesIO
+					compressed = True
 				doc = tostr(doc, "utf_8")
-				self.docList.append( [doc, entry.startGlyphID, entry.endGlyphID] )
+				self.docList.append(
+					SVGDocument(doc, entry.startGlyphID, entry.endGlyphID, compressed)
+				)
 
 	def compile(self, ttFont):
 		version = 0
@@ -96,9 +101,13 @@ class table_S_V_G_(DefaultTable.DefaultTable):
 		entryList.append(datum)
 		curOffset = len(datum) + doc_index_entry_format_0Size*numEntries
 		seenDocs = {}
-		for doc, startGlyphID, endGlyphID in self.docList:
+		allCompressed = getattr(self, "compressed", False)
-			docBytes = tobytes(doc, encoding="utf_8")
+		for i, doc in enumerate(self.docList):
-			if getattr(self, "compressed", False) and not docBytes.startswith(b"\x1f\x8b"):
+			if isinstance(doc, (list, tuple)):
+				doc = SVGDocument(*doc)
+				self.docList[i] = doc
+			docBytes = tobytes(doc.data, encoding="utf_8")
+			if (allCompressed or doc.compressed) and not docBytes.startswith(b"\x1f\x8b"):
 				import gzip
 				bytesIO = BytesIO()
 				with gzip.GzipFile(None, "w", fileobj=bytesIO) as gzipper:
@@ -115,7 +124,7 @@ class table_S_V_G_(DefaultTable.DefaultTable):
 				curOffset += docLength
 				seenDocs[docBytes] = docOffset
 				docList.append(docBytes)
-			entry = struct.pack(">HHLL", startGlyphID, endGlyphID, docOffset, docLength)
+			entry = struct.pack(">HHLL", doc.startGlyphID, doc.endGlyphID, docOffset, docLength)
 			entryList.append(entry)
 		entryList.extend(docList)
 		svgDocData = bytesjoin(entryList)
@@ -127,10 +136,16 @@ class table_S_V_G_(DefaultTable.DefaultTable):
 		return data
 
 	def toXML(self, writer, ttFont):
-		for doc, startGID, endGID in self.docList:
+		for i, doc in enumerate(self.docList):
-			writer.begintag("svgDoc", startGlyphID=startGID, endGlyphID=endGID)
+			if isinstance(doc, (list, tuple)):
+				doc = SVGDocument(*doc)
+				self.docList[i] = doc
+			attrs = {"startGlyphID": doc.startGlyphID, "endGlyphID": doc.endGlyphID}
+			if doc.compressed:
+				attrs["compressed"] = 1
+			writer.begintag("svgDoc", **attrs)
 			writer.newline()
-			writer.writecdata(doc)
+			writer.writecdata(doc.data)
 			writer.newline()
 			writer.endtag("svgDoc")
 			writer.newline()
@@ -143,7 +158,8 @@ class table_S_V_G_(DefaultTable.DefaultTable):
 			doc = doc.strip()
 			startGID = int(attrs["startGlyphID"])
 			endGID = int(attrs["endGlyphID"])
-			self.docList.append( [doc, startGID, endGID] )
+			compressed = bool(safeEval(attrs.get("compressed", "0")))
+			self.docList.append(SVGDocument(doc, startGID, endGID, compressed))
 		else:
 			log.warning("Unknown %s %s", name, content)
 
@@ -157,3 +173,23 @@ class DocumentIndexEntry(object):
 
 	def __repr__(self):
 		return "startGlyphID: %s, endGlyphID: %s, svgDocOffset: %s, svgDocLength: %s" % (self.startGlyphID, self.endGlyphID, self.svgDocOffset, self.svgDocLength)
+
+
+@dataclass
+class SVGDocument(Sequence):
+	data: str
+	startGlyphID: int
+	endGlyphID: int
+	compressed: bool = False
+
+	# Previously, the SVG table's docList attribute contained a lists of 3 items:
+	# [doc, startGlyphID, endGlyphID]; later, we added a `compressed` attribute.
+	# For backward compatibility with code that depends of them being sequences of
+	# fixed length=3, we subclass the Sequence abstract base class and pretend only
+	# the first three items are present. 'compressed' is only accessible via named
+	# attribute lookup like regular dataclasses: i.e. `doc.compressed`, not `doc[3]`
+	def __getitem__(self, index):
+		return astuple(self)[:3][index]
+
+	def __len__(self):
+		return 3