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\-/]+)\.(rst|txt)')
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# Keep track of cross-reference lookups that failed so we don't have to
125# do them again.
126#
127failed_lookups = { }
128def failure_seen(target):
129    return (target) in failed_lookups
130def note_failure(target):
131    failed_lookups[target] = True
132
133#
134# In sphinx3 we can cross-reference to C macro and function, each one with its
135# own C role, but both match the same regex, so we try both.
136#
137def markup_func_ref_sphinx3(docname, app, match):
138    cdom = app.env.domains['c']
139    #
140    # Go through the dance of getting an xref out of the C domain
141    #
142    base_target = match.group(2)
143    target_text = nodes.Text(match.group(0))
144    xref = None
145    possible_targets = [base_target]
146    # Check if this document has a namespace, and if so, try
147    # cross-referencing inside it first.
148    if c_namespace:
149        possible_targets.insert(0, c_namespace + "." + base_target)
150
151    if base_target not in Skipnames:
152        for target in possible_targets:
153            if (target not in Skipfuncs) and not failure_seen(target):
154                lit_text = nodes.literal(classes=['xref', 'c', 'c-func'])
155                lit_text += target_text
156                pxref = addnodes.pending_xref('', refdomain = 'c',
157                                              reftype = 'function',
158                                              reftarget = target,
159                                              modname = None,
160                                              classname = None)
161                #
162                # XXX The Latex builder will throw NoUri exceptions here,
163                # work around that by ignoring them.
164                #
165                try:
166                    xref = cdom.resolve_xref(app.env, docname, app.builder,
167                                             'function', target, pxref,
168                                             lit_text)
169                except NoUri:
170                    xref = None
171
172                if xref:
173                    return xref
174                note_failure(target)
175
176    return target_text
177
178def markup_c_ref(docname, app, match):
179    class_str = {# Sphinx 2 only
180                 RE_function: 'c-func',
181                 RE_generic_type: 'c-type',
182                 # Sphinx 3+ only
183                 RE_struct: 'c-struct',
184                 RE_union: 'c-union',
185                 RE_enum: 'c-enum',
186                 RE_typedef: 'c-type',
187                 }
188    reftype_str = {# Sphinx 2 only
189                   RE_function: 'function',
190                   RE_generic_type: 'type',
191                   # Sphinx 3+ only
192                   RE_struct: 'struct',
193                   RE_union: 'union',
194                   RE_enum: 'enum',
195                   RE_typedef: 'type',
196                   }
197
198    cdom = app.env.domains['c']
199    #
200    # Go through the dance of getting an xref out of the C domain
201    #
202    base_target = match.group(2)
203    target_text = nodes.Text(match.group(0))
204    xref = None
205    possible_targets = [base_target]
206    # Check if this document has a namespace, and if so, try
207    # cross-referencing inside it first.
208    if c_namespace:
209        possible_targets.insert(0, c_namespace + "." + base_target)
210
211    if base_target not in Skipnames:
212        for target in possible_targets:
213            if not (match.re == RE_function and target in Skipfuncs):
214                lit_text = nodes.literal(classes=['xref', 'c', class_str[match.re]])
215                lit_text += target_text
216                pxref = addnodes.pending_xref('', refdomain = 'c',
217                                              reftype = reftype_str[match.re],
218                                              reftarget = target, modname = None,
219                                              classname = None)
220                #
221                # XXX The Latex builder will throw NoUri exceptions here,
222                # work around that by ignoring them.
223                #
224                try:
225                    xref = cdom.resolve_xref(app.env, docname, app.builder,
226                                             reftype_str[match.re], target, pxref,
227                                             lit_text)
228                except NoUri:
229                    xref = None
230
231                if xref:
232                    return xref
233
234    return target_text
235
236#
237# Try to replace a documentation reference of the form Documentation/... with a
238# cross reference to that page
239#
240def markup_doc_ref(docname, app, match):
241    stddom = app.env.domains['std']
242    #
243    # Go through the dance of getting an xref out of the std domain
244    #
245    absolute = match.group(1)
246    target = match.group(2)
247    if absolute:
248       target = "/" + target
249    xref = None
250    pxref = addnodes.pending_xref('', refdomain = 'std', reftype = 'doc',
251                                  reftarget = target, modname = None,
252                                  classname = None, refexplicit = False)
253    #
254    # XXX The Latex builder will throw NoUri exceptions here,
255    # work around that by ignoring them.
256    #
257    try:
258        xref = stddom.resolve_xref(app.env, docname, app.builder, 'doc',
259                                   target, pxref, None)
260    except NoUri:
261        xref = None
262    #
263    # Return the xref if we got it; otherwise just return the plain text.
264    #
265    if xref:
266        return xref
267    else:
268        return nodes.Text(match.group(0))
269
270def get_c_namespace(app, docname):
271    source = app.env.doc2path(docname)
272    with open(source) as f:
273        for l in f:
274            match = RE_namespace.search(l)
275            if match:
276                return match.group(1)
277    return ''
278
279def auto_markup(app, doctree, name):
280    global c_namespace
281    c_namespace = get_c_namespace(app, name)
282    def text_but_not_a_reference(node):
283        # The nodes.literal test catches ``literal text``, its purpose is to
284        # avoid adding cross-references to functions that have been explicitly
285        # marked with cc:func:.
286        if not isinstance(node, nodes.Text) or isinstance(node.parent, nodes.literal):
287            return False
288
289        child_of_reference = False
290        parent = node.parent
291        while parent:
292            if isinstance(parent, nodes.Referential):
293                child_of_reference = True
294                break
295            parent = parent.parent
296        return not child_of_reference
297
298    #
299    # This loop could eventually be improved on.  Someday maybe we
300    # want a proper tree traversal with a lot of awareness of which
301    # kinds of nodes to prune.  But this works well for now.
302    #
303    for para in doctree.traverse(nodes.paragraph):
304        for node in para.traverse(condition=text_but_not_a_reference):
305            node.parent.replace(node, markup_refs(name, app, node))
306
307def setup(app):
308    app.connect('doctree-resolved', auto_markup)
309    return {
310        'parallel_read_safe': True,
311        'parallel_write_safe': True,
312        }
313