1# SPDX-License-Identifier: GPL-2.0
2# Copyright 2019 Jonathan Corbet <corbet@lwn.net>
3#
4# Apply kernel-specific tweaks after the initial document processing
5# has been done.
6#
7from docutils import nodes
8import sphinx
9from sphinx import addnodes
10if sphinx.version_info[0] < 2 or \
11   sphinx.version_info[0] == 2 and sphinx.version_info[1] < 1:
12    from sphinx.environment import NoUri
13else:
14    from sphinx.errors import NoUri
15import re
16from itertools import chain
17
18#
19# Python 2 lacks re.ASCII...
20#
21try:
22    ascii_p3 = re.ASCII
23except AttributeError:
24    ascii_p3 = 0
25
26#
27# Regex nastiness.  Of course.
28# Try to identify "function()" that's not already marked up some
29# other way.  Sphinx doesn't like a lot of stuff right after a
30# :c:func: block (i.e. ":c:func:`mmap()`s" flakes out), so the last
31# bit tries to restrict matches to things that won't create trouble.
32#
33RE_function = re.compile(r'\b(([a-zA-Z_]\w+)\(\))', flags=ascii_p3)
34
35#
36# Sphinx 2 uses the same :c:type role for struct, union, enum and typedef
37#
38RE_generic_type = re.compile(r'\b(struct|union|enum|typedef)\s+([a-zA-Z_]\w+)',
39                             flags=ascii_p3)
40
41#
42# Sphinx 3 uses a different C role for each one of struct, union, enum and
43# typedef
44#
45RE_struct = re.compile(r'\b(struct)\s+([a-zA-Z_]\w+)', flags=ascii_p3)
46RE_union = re.compile(r'\b(union)\s+([a-zA-Z_]\w+)', flags=ascii_p3)
47RE_enum = re.compile(r'\b(enum)\s+([a-zA-Z_]\w+)', flags=ascii_p3)
48RE_typedef = re.compile(r'\b(typedef)\s+([a-zA-Z_]\w+)', flags=ascii_p3)
49
50#
51# Detects a reference to a documentation page of the form Documentation/... with
52# an optional extension
53#
54RE_doc = re.compile(r'\bDocumentation(/[\w\-_/]+)(\.\w+)*')
55
56#
57# Reserved C words that we should skip when cross-referencing
58#
59Skipnames = [ 'for', 'if', 'register', 'sizeof', 'struct', 'unsigned' ]
60
61
62#
63# Many places in the docs refer to common system calls.  It is
64# pointless to try to cross-reference them and, as has been known
65# to happen, somebody defining a function by these names can lead
66# to the creation of incorrect and confusing cross references.  So
67# just don't even try with these names.
68#
69Skipfuncs = [ 'open', 'close', 'read', 'write', 'fcntl', 'mmap',
70              'select', 'poll', 'fork', 'execve', 'clone', 'ioctl',
71              'socket' ]
72
73def markup_refs(docname, app, node):
74    t = node.astext()
75    done = 0
76    repl = [ ]
77    #
78    # Associate each regex with the function that will markup its matches
79    #
80    markup_func_sphinx2 = {RE_doc: markup_doc_ref,
81                           RE_function: markup_c_ref,
82                           RE_generic_type: markup_c_ref}
83
84    markup_func_sphinx3 = {RE_doc: markup_doc_ref,
85                           RE_function: markup_func_ref_sphinx3,
86                           RE_struct: markup_c_ref,
87                           RE_union: markup_c_ref,
88                           RE_enum: markup_c_ref,
89                           RE_typedef: markup_c_ref}
90
91    if sphinx.version_info[0] >= 3:
92        markup_func = markup_func_sphinx3
93    else:
94        markup_func = markup_func_sphinx2
95
96    match_iterators = [regex.finditer(t) for regex in markup_func]
97    #
98    # Sort all references by the starting position in text
99    #
100    sorted_matches = sorted(chain(*match_iterators), key=lambda m: m.start())
101    for m in sorted_matches:
102        #
103        # Include any text prior to match as a normal text node.
104        #
105        if m.start() > done:
106            repl.append(nodes.Text(t[done:m.start()]))
107
108        #
109        # Call the function associated with the regex that matched this text and
110        # append its return to the text
111        #
112        repl.append(markup_func[m.re](docname, app, m))
113
114        done = m.end()
115    if done < len(t):
116        repl.append(nodes.Text(t[done:]))
117    return repl
118
119#
120# In sphinx3 we can cross-reference to C macro and function, each one with its
121# own C role, but both match the same regex, so we try both.
122#
123def markup_func_ref_sphinx3(docname, app, match):
124    class_str = ['c-func', 'c-macro']
125    reftype_str = ['function', 'macro']
126
127    cdom = app.env.domains['c']
128    #
129    # Go through the dance of getting an xref out of the C domain
130    #
131    target = match.group(2)
132    target_text = nodes.Text(match.group(0))
133    xref = None
134    if not (target in Skipfuncs or target in Skipnames):
135        for class_s, reftype_s in zip(class_str, reftype_str):
136            lit_text = nodes.literal(classes=['xref', 'c', class_s])
137            lit_text += target_text
138            pxref = addnodes.pending_xref('', refdomain = 'c',
139                                          reftype = reftype_s,
140                                          reftarget = target, modname = None,
141                                          classname = None)
142            #
143            # XXX The Latex builder will throw NoUri exceptions here,
144            # work around that by ignoring them.
145            #
146            try:
147                xref = cdom.resolve_xref(app.env, docname, app.builder,
148                                         reftype_s, target, pxref,
149                                         lit_text)
150            except NoUri:
151                xref = None
152
153            if xref:
154                return xref
155
156    return target_text
157
158def markup_c_ref(docname, app, match):
159    class_str = {# Sphinx 2 only
160                 RE_function: 'c-func',
161                 RE_generic_type: 'c-type',
162                 # Sphinx 3+ only
163                 RE_struct: 'c-struct',
164                 RE_union: 'c-union',
165                 RE_enum: 'c-enum',
166                 RE_typedef: 'c-type',
167                 }
168    reftype_str = {# Sphinx 2 only
169                   RE_function: 'function',
170                   RE_generic_type: 'type',
171                   # Sphinx 3+ only
172                   RE_struct: 'struct',
173                   RE_union: 'union',
174                   RE_enum: 'enum',
175                   RE_typedef: 'type',
176                   }
177
178    cdom = app.env.domains['c']
179    #
180    # Go through the dance of getting an xref out of the C domain
181    #
182    target = match.group(2)
183    target_text = nodes.Text(match.group(0))
184    xref = None
185    if not ((match.re == RE_function and target in Skipfuncs)
186            or (target in Skipnames)):
187        lit_text = nodes.literal(classes=['xref', 'c', class_str[match.re]])
188        lit_text += target_text
189        pxref = addnodes.pending_xref('', refdomain = 'c',
190                                      reftype = reftype_str[match.re],
191                                      reftarget = target, modname = None,
192                                      classname = None)
193        #
194        # XXX The Latex builder will throw NoUri exceptions here,
195        # work around that by ignoring them.
196        #
197        try:
198            xref = cdom.resolve_xref(app.env, docname, app.builder,
199                                     reftype_str[match.re], target, pxref,
200                                     lit_text)
201        except NoUri:
202            xref = None
203    #
204    # Return the xref if we got it; otherwise just return the plain text.
205    #
206    if xref:
207        return xref
208    else:
209        return target_text
210
211#
212# Try to replace a documentation reference of the form Documentation/... with a
213# cross reference to that page
214#
215def markup_doc_ref(docname, app, match):
216    stddom = app.env.domains['std']
217    #
218    # Go through the dance of getting an xref out of the std domain
219    #
220    target = match.group(1)
221    xref = None
222    pxref = addnodes.pending_xref('', refdomain = 'std', reftype = 'doc',
223                                  reftarget = target, modname = None,
224                                  classname = None, refexplicit = False)
225    #
226    # XXX The Latex builder will throw NoUri exceptions here,
227    # work around that by ignoring them.
228    #
229    try:
230        xref = stddom.resolve_xref(app.env, docname, app.builder, 'doc',
231                                   target, pxref, None)
232    except NoUri:
233        xref = None
234    #
235    # Return the xref if we got it; otherwise just return the plain text.
236    #
237    if xref:
238        return xref
239    else:
240        return nodes.Text(match.group(0))
241
242def auto_markup(app, doctree, name):
243    #
244    # This loop could eventually be improved on.  Someday maybe we
245    # want a proper tree traversal with a lot of awareness of which
246    # kinds of nodes to prune.  But this works well for now.
247    #
248    # The nodes.literal test catches ``literal text``, its purpose is to
249    # avoid adding cross-references to functions that have been explicitly
250    # marked with cc:func:.
251    #
252    for para in doctree.traverse(nodes.paragraph):
253        for node in para.traverse(nodes.Text):
254            if not isinstance(node.parent, nodes.literal):
255                node.parent.replace(node, markup_refs(name, app, node))
256
257def setup(app):
258    app.connect('doctree-resolved', auto_markup)
259    return {
260        'parallel_read_safe': True,
261        'parallel_write_safe': True,
262        }
263