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
56RE_namespace = re.compile(r'^\s*..\s*c:namespace::\s*(\S+)\s*$')
57
58#
59# Reserved C words that we should skip when cross-referencing
60#
61Skipnames = [ 'for', 'if', 'register', 'sizeof', 'struct', 'unsigned' ]
62
63
64#
65# Many places in the docs refer to common system calls.  It is
66# pointless to try to cross-reference them and, as has been known
67# to happen, somebody defining a function by these names can lead
68# to the creation of incorrect and confusing cross references.  So
69# just don't even try with these names.
70#
71Skipfuncs = [ 'open', 'close', 'read', 'write', 'fcntl', 'mmap',
72              'select', 'poll', 'fork', 'execve', 'clone', 'ioctl',
73              'socket' ]
74
75c_namespace = ''
76
77def markup_refs(docname, app, node):
78    t = node.astext()
79    done = 0
80    repl = [ ]
81    #
82    # Associate each regex with the function that will markup its matches
83    #
84    markup_func_sphinx2 = {RE_doc: markup_doc_ref,
85                           RE_function: markup_c_ref,
86                           RE_generic_type: markup_c_ref}
87
88    markup_func_sphinx3 = {RE_doc: markup_doc_ref,
89                           RE_function: markup_func_ref_sphinx3,
90                           RE_struct: markup_c_ref,
91                           RE_union: markup_c_ref,
92                           RE_enum: markup_c_ref,
93                           RE_typedef: markup_c_ref}
94
95    if sphinx.version_info[0] >= 3:
96        markup_func = markup_func_sphinx3
97    else:
98        markup_func = markup_func_sphinx2
99
100    match_iterators = [regex.finditer(t) for regex in markup_func]
101    #
102    # Sort all references by the starting position in text
103    #
104    sorted_matches = sorted(chain(*match_iterators), key=lambda m: m.start())
105    for m in sorted_matches:
106        #
107        # Include any text prior to match as a normal text node.
108        #
109        if m.start() > done:
110            repl.append(nodes.Text(t[done:m.start()]))
111
112        #
113        # Call the function associated with the regex that matched this text and
114        # append its return to the text
115        #
116        repl.append(markup_func[m.re](docname, app, m))
117
118        done = m.end()
119    if done < len(t):
120        repl.append(nodes.Text(t[done:]))
121    return repl
122
123#
124# In sphinx3 we can cross-reference to C macro and function, each one with its
125# own C role, but both match the same regex, so we try both.
126#
127def markup_func_ref_sphinx3(docname, app, match):
128    class_str = ['c-func', 'c-macro']
129    reftype_str = ['function', 'macro']
130
131    cdom = app.env.domains['c']
132    #
133    # Go through the dance of getting an xref out of the C domain
134    #
135    base_target = match.group(2)
136    target_text = nodes.Text(match.group(0))
137    xref = None
138    possible_targets = [base_target]
139    # Check if this document has a namespace, and if so, try
140    # cross-referencing inside it first.
141    if c_namespace:
142        possible_targets.insert(0, c_namespace + "." + base_target)
143
144    if base_target not in Skipnames:
145        for target in possible_targets:
146            if target not in Skipfuncs:
147                for class_s, reftype_s in zip(class_str, reftype_str):
148                    lit_text = nodes.literal(classes=['xref', 'c', class_s])
149                    lit_text += target_text
150                    pxref = addnodes.pending_xref('', refdomain = 'c',
151                                                  reftype = reftype_s,
152                                                  reftarget = target, modname = None,
153                                                  classname = None)
154                    #
155                    # XXX The Latex builder will throw NoUri exceptions here,
156                    # work around that by ignoring them.
157                    #
158                    try:
159                        xref = cdom.resolve_xref(app.env, docname, app.builder,
160                                                 reftype_s, target, pxref,
161                                                 lit_text)
162                    except NoUri:
163                        xref = None
164
165                    if xref:
166                        return xref
167
168    return target_text
169
170def markup_c_ref(docname, app, match):
171    class_str = {# Sphinx 2 only
172                 RE_function: 'c-func',
173                 RE_generic_type: 'c-type',
174                 # Sphinx 3+ only
175                 RE_struct: 'c-struct',
176                 RE_union: 'c-union',
177                 RE_enum: 'c-enum',
178                 RE_typedef: 'c-type',
179                 }
180    reftype_str = {# Sphinx 2 only
181                   RE_function: 'function',
182                   RE_generic_type: 'type',
183                   # Sphinx 3+ only
184                   RE_struct: 'struct',
185                   RE_union: 'union',
186                   RE_enum: 'enum',
187                   RE_typedef: 'type',
188                   }
189
190    cdom = app.env.domains['c']
191    #
192    # Go through the dance of getting an xref out of the C domain
193    #
194    base_target = match.group(2)
195    target_text = nodes.Text(match.group(0))
196    xref = None
197    possible_targets = [base_target]
198    # Check if this document has a namespace, and if so, try
199    # cross-referencing inside it first.
200    if c_namespace:
201        possible_targets.insert(0, c_namespace + "." + base_target)
202
203    if base_target not in Skipnames:
204        for target in possible_targets:
205            if not (match.re == RE_function and target in Skipfuncs):
206                lit_text = nodes.literal(classes=['xref', 'c', class_str[match.re]])
207                lit_text += target_text
208                pxref = addnodes.pending_xref('', refdomain = 'c',
209                                              reftype = reftype_str[match.re],
210                                              reftarget = target, modname = None,
211                                              classname = None)
212                #
213                # XXX The Latex builder will throw NoUri exceptions here,
214                # work around that by ignoring them.
215                #
216                try:
217                    xref = cdom.resolve_xref(app.env, docname, app.builder,
218                                             reftype_str[match.re], target, pxref,
219                                             lit_text)
220                except NoUri:
221                    xref = None
222
223                if xref:
224                    return xref
225
226    return target_text
227
228#
229# Try to replace a documentation reference of the form Documentation/... with a
230# cross reference to that page
231#
232def markup_doc_ref(docname, app, match):
233    stddom = app.env.domains['std']
234    #
235    # Go through the dance of getting an xref out of the std domain
236    #
237    target = match.group(1)
238    xref = None
239    pxref = addnodes.pending_xref('', refdomain = 'std', reftype = 'doc',
240                                  reftarget = target, modname = None,
241                                  classname = None, refexplicit = False)
242    #
243    # XXX The Latex builder will throw NoUri exceptions here,
244    # work around that by ignoring them.
245    #
246    try:
247        xref = stddom.resolve_xref(app.env, docname, app.builder, 'doc',
248                                   target, pxref, None)
249    except NoUri:
250        xref = None
251    #
252    # Return the xref if we got it; otherwise just return the plain text.
253    #
254    if xref:
255        return xref
256    else:
257        return nodes.Text(match.group(0))
258
259def get_c_namespace(app, docname):
260    source = app.env.doc2path(docname)
261    with open(source) as f:
262        for l in f:
263            match = RE_namespace.search(l)
264            if match:
265                return match.group(1)
266    return ''
267
268def auto_markup(app, doctree, name):
269    global c_namespace
270    c_namespace = get_c_namespace(app, name)
271    #
272    # This loop could eventually be improved on.  Someday maybe we
273    # want a proper tree traversal with a lot of awareness of which
274    # kinds of nodes to prune.  But this works well for now.
275    #
276    # The nodes.literal test catches ``literal text``, its purpose is to
277    # avoid adding cross-references to functions that have been explicitly
278    # marked with cc:func:.
279    #
280    for para in doctree.traverse(nodes.paragraph):
281        for node in para.traverse(nodes.Text):
282            if not isinstance(node.parent, nodes.literal):
283                node.parent.replace(node, markup_refs(name, app, node))
284
285def setup(app):
286    app.connect('doctree-resolved', auto_markup)
287    return {
288        'parallel_read_safe': True,
289        'parallel_write_safe': True,
290        }
291