xref: /aosp_15_r20/external/fonttools/Lib/fontTools/subset/svg.py (revision e1fe3e4ad2793916b15cccdc4a7da52a7e1dd0e9)
1from __future__ import annotations
2
3import re
4from functools import lru_cache
5from itertools import chain, count
6from typing import Dict, Iterable, Iterator, List, Optional, Set, Tuple
7
8try:
9    from lxml import etree
10except ImportError:
11    # lxml is required for subsetting SVG, but we prefer to delay the import error
12    # until subset_glyphs() is called (i.e. if font to subset has an 'SVG ' table)
13    etree = None
14
15from fontTools import ttLib
16from fontTools.subset.util import _add_method
17from fontTools.ttLib.tables.S_V_G_ import SVGDocument
18
19
20__all__ = ["subset_glyphs"]
21
22
23GID_RE = re.compile(r"^glyph(\d+)$")
24
25NAMESPACES = {
26    "svg": "http://www.w3.org/2000/svg",
27    "xlink": "http://www.w3.org/1999/xlink",
28}
29XLINK_HREF = f'{{{NAMESPACES["xlink"]}}}href'
30
31
32# TODO(antrotype): Replace with functools.cache once we are 3.9+
33@lru_cache(maxsize=None)
34def xpath(path):
35    # compile XPath upfront, caching result to reuse on multiple elements
36    return etree.XPath(path, namespaces=NAMESPACES)
37
38
39def group_elements_by_id(tree: etree.Element) -> Dict[str, etree.Element]:
40    # select all svg elements with 'id' attribute no matter where they are
41    # including the root element itself:
42    # https://github.com/fonttools/fonttools/issues/2548
43    return {el.attrib["id"]: el for el in xpath("//svg:*[@id]")(tree)}
44
45
46def parse_css_declarations(style_attr: str) -> Dict[str, str]:
47    # https://developer.mozilla.org/en-US/docs/Web/SVG/Attribute/style
48    # https://developer.mozilla.org/en-US/docs/Web/CSS/Syntax#css_declarations
49    result = {}
50    for declaration in style_attr.split(";"):
51        if declaration.count(":") == 1:
52            property_name, value = declaration.split(":")
53            property_name = property_name.strip()
54            result[property_name] = value.strip()
55        elif declaration.strip():
56            raise ValueError(f"Invalid CSS declaration syntax: {declaration}")
57    return result
58
59
60def iter_referenced_ids(tree: etree.Element) -> Iterator[str]:
61    # Yield all the ids that can be reached via references from this element tree.
62    # We currently support xlink:href (as used by <use> and gradient templates),
63    # and local url(#...) links found in fill or clip-path attributes
64    # TODO(anthrotype): Check we aren't missing other supported kinds of reference
65    find_svg_elements_with_references = xpath(
66        ".//svg:*[ "
67        "starts-with(@xlink:href, '#') "
68        "or starts-with(@fill, 'url(#') "
69        "or starts-with(@clip-path, 'url(#') "
70        "or contains(@style, ':url(#') "
71        "]",
72    )
73    for el in chain([tree], find_svg_elements_with_references(tree)):
74        ref_id = href_local_target(el)
75        if ref_id is not None:
76            yield ref_id
77
78        attrs = el.attrib
79        if "style" in attrs:
80            attrs = {**dict(attrs), **parse_css_declarations(el.attrib["style"])}
81        for attr in ("fill", "clip-path"):
82            if attr in attrs:
83                value = attrs[attr]
84                if value.startswith("url(#") and value.endswith(")"):
85                    ref_id = value[5:-1]
86                    assert ref_id
87                    yield ref_id
88
89
90def closure_element_ids(
91    elements: Dict[str, etree.Element], element_ids: Set[str]
92) -> None:
93    # Expand the initial subset of element ids to include ids that can be reached
94    # via references from the initial set.
95    unvisited = element_ids
96    while unvisited:
97        referenced: Set[str] = set()
98        for el_id in unvisited:
99            if el_id not in elements:
100                # ignore dangling reference; not our job to validate svg
101                continue
102            referenced.update(iter_referenced_ids(elements[el_id]))
103        referenced -= element_ids
104        element_ids.update(referenced)
105        unvisited = referenced
106
107
108def subset_elements(el: etree.Element, retained_ids: Set[str]) -> bool:
109    # Keep elements if their id is in the subset, or any of their children's id is.
110    # Drop elements whose id is not in the subset, and either have no children,
111    # or all their children are being dropped.
112    if el.attrib.get("id") in retained_ids:
113        # if id is in the set, don't recurse; keep whole subtree
114        return True
115    # recursively subset all the children; we use a list comprehension instead
116    # of a parentheses-less generator expression because we don't want any() to
117    # short-circuit, as our function has a side effect of dropping empty elements.
118    if any([subset_elements(e, retained_ids) for e in el]):
119        return True
120    assert len(el) == 0
121    parent = el.getparent()
122    if parent is not None:
123        parent.remove(el)
124    return False
125
126
127def remap_glyph_ids(
128    svg: etree.Element, glyph_index_map: Dict[int, int]
129) -> Dict[str, str]:
130    # Given {old_gid: new_gid} map, rename all elements containing id="glyph{gid}"
131    # special attributes
132    elements = group_elements_by_id(svg)
133    id_map = {}
134    for el_id, el in elements.items():
135        m = GID_RE.match(el_id)
136        if not m:
137            continue
138        old_index = int(m.group(1))
139        new_index = glyph_index_map.get(old_index)
140        if new_index is not None:
141            if old_index == new_index:
142                continue
143            new_id = f"glyph{new_index}"
144        else:
145            # If the old index is missing, the element correspond to a glyph that was
146            # excluded from the font's subset.
147            # We rename it to avoid clashes with the new GIDs or other element ids.
148            new_id = f".{el_id}"
149            n = count(1)
150            while new_id in elements:
151                new_id = f"{new_id}.{next(n)}"
152
153        id_map[el_id] = new_id
154        el.attrib["id"] = new_id
155
156    return id_map
157
158
159def href_local_target(el: etree.Element) -> Optional[str]:
160    if XLINK_HREF in el.attrib:
161        href = el.attrib[XLINK_HREF]
162        if href.startswith("#") and len(href) > 1:
163            return href[1:]  # drop the leading #
164    return None
165
166
167def update_glyph_href_links(svg: etree.Element, id_map: Dict[str, str]) -> None:
168    # update all xlink:href="#glyph..." attributes to point to the new glyph ids
169    for el in xpath(".//svg:*[starts-with(@xlink:href, '#glyph')]")(svg):
170        old_id = href_local_target(el)
171        assert old_id is not None
172        if old_id in id_map:
173            new_id = id_map[old_id]
174            el.attrib[XLINK_HREF] = f"#{new_id}"
175
176
177def ranges(ints: Iterable[int]) -> Iterator[Tuple[int, int]]:
178    # Yield sorted, non-overlapping (min, max) ranges of consecutive integers
179    sorted_ints = iter(sorted(set(ints)))
180    try:
181        start = end = next(sorted_ints)
182    except StopIteration:
183        return
184    for v in sorted_ints:
185        if v - 1 == end:
186            end = v
187        else:
188            yield (start, end)
189            start = end = v
190    yield (start, end)
191
192
193@_add_method(ttLib.getTableClass("SVG "))
194def subset_glyphs(self, s) -> bool:
195    if etree is None:
196        raise ImportError("No module named 'lxml', required to subset SVG")
197
198    # glyph names (before subsetting)
199    glyph_order: List[str] = s.orig_glyph_order
200    # map from glyph names to original glyph indices
201    rev_orig_glyph_map: Dict[str, int] = s.reverseOrigGlyphMap
202    # map from original to new glyph indices (after subsetting)
203    glyph_index_map: Dict[int, int] = s.glyph_index_map
204
205    new_docs: List[SVGDocument] = []
206    for doc in self.docList:
207        glyphs = {
208            glyph_order[i] for i in range(doc.startGlyphID, doc.endGlyphID + 1)
209        }.intersection(s.glyphs)
210        if not glyphs:
211            # no intersection: we can drop the whole record
212            continue
213
214        svg = etree.fromstring(
215            # encode because fromstring dislikes xml encoding decl if input is str.
216            # SVG xml encoding must be utf-8 as per OT spec.
217            doc.data.encode("utf-8"),
218            parser=etree.XMLParser(
219                # Disable libxml2 security restrictions to support very deep trees.
220                # Without this we would get an error like this:
221                # `lxml.etree.XMLSyntaxError: internal error: Huge input lookup`
222                # when parsing big fonts e.g. noto-emoji-picosvg.ttf.
223                huge_tree=True,
224                # ignore blank text as it's not meaningful in OT-SVG; it also prevents
225                # dangling tail text after removing an element when pretty_print=True
226                remove_blank_text=True,
227                # don't replace entities; we don't expect any in OT-SVG and they may
228                # be abused for XXE attacks
229                resolve_entities=False,
230            ),
231        )
232
233        elements = group_elements_by_id(svg)
234        gids = {rev_orig_glyph_map[g] for g in glyphs}
235        element_ids = {f"glyph{i}" for i in gids}
236        closure_element_ids(elements, element_ids)
237
238        if not subset_elements(svg, element_ids):
239            continue
240
241        if not s.options.retain_gids:
242            id_map = remap_glyph_ids(svg, glyph_index_map)
243            update_glyph_href_links(svg, id_map)
244
245        new_doc = etree.tostring(svg, pretty_print=s.options.pretty_svg).decode("utf-8")
246
247        new_gids = (glyph_index_map[i] for i in gids)
248        for start, end in ranges(new_gids):
249            new_docs.append(SVGDocument(new_doc, start, end, doc.compressed))
250
251    self.docList = new_docs
252
253    return bool(self.docList)
254