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 10from sphinx.errors import NoUri 11import re 12from itertools import chain 13 14from kernel_abi import get_kernel_abi 15 16# 17# Regex nastiness. Of course. 18# Try to identify "function()" that's not already marked up some 19# other way. Sphinx doesn't like a lot of stuff right after a 20# :c:func: block (i.e. ":c:func:`mmap()`s" flakes out), so the last 21# bit tries to restrict matches to things that won't create trouble. 22# 23RE_function = re.compile(r'\b(([a-zA-Z_]\w+)\(\))', flags=re.ASCII) 24 25# 26# Sphinx 3 uses a different C role for each one of struct, union, enum and 27# typedef 28# 29RE_struct = re.compile(r'\b(struct)\s+([a-zA-Z_]\w+)', flags=re.ASCII) 30RE_union = re.compile(r'\b(union)\s+([a-zA-Z_]\w+)', flags=re.ASCII) 31RE_enum = re.compile(r'\b(enum)\s+([a-zA-Z_]\w+)', flags=re.ASCII) 32RE_typedef = re.compile(r'\b(typedef)\s+([a-zA-Z_]\w+)', flags=re.ASCII) 33 34# 35# Detects a reference to a documentation page of the form Documentation/... with 36# an optional extension 37# 38RE_doc = re.compile(r'(\bDocumentation/)?((\.\./)*[\w\-/]+)\.(rst|txt)') 39RE_abi_file = re.compile(r'(\bDocumentation/ABI/[\w\-/]+)') 40RE_abi_symbol = re.compile(r'(\b/(sys|config|proc)/[\w\-/]+)') 41 42RE_namespace = re.compile(r'^\s*..\s*c:namespace::\s*(\S+)\s*$') 43 44# 45# Reserved C words that we should skip when cross-referencing 46# 47Skipnames = [ 'for', 'if', 'register', 'sizeof', 'struct', 'unsigned' ] 48 49# 50# Common English words that should not be recognized as C identifiers 51# when following struct/union/enum/typedef keywords. 52# Example: "a simple struct that" in workqueue.rst should not be marked as code. 53# 54Skipidentifiers = [ 'that', 'which', 'where', 'whose' ] 55 56# 57# Many places in the docs refer to common system calls. It is 58# pointless to try to cross-reference them and, as has been known 59# to happen, somebody defining a function by these names can lead 60# to the creation of incorrect and confusing cross references. So 61# just don't even try with these names. 62# 63Skipfuncs = [ 'open', 'close', 'read', 'write', 'fcntl', 'mmap', 64 'select', 'poll', 'fork', 'execve', 'clone', 'ioctl', 65 'socket' ] 66 67c_namespace = '' 68 69# 70# Detect references to commits. 71# 72RE_git = re.compile(r'commit\s+(?P<rev>[0-9a-f]{12,40})(?:\s+\(".*?"\))?', 73 flags=re.IGNORECASE | re.DOTALL) 74 75def markup_refs(docname, app, node): 76 t = node.astext() 77 done = 0 78 repl = [ ] 79 # 80 # Associate each regex with the function that will markup its matches 81 # 82 83 markup_func = {RE_doc: markup_doc_ref, 84 RE_abi_file: markup_abi_file_ref, 85 RE_abi_symbol: markup_abi_ref, 86 RE_function: markup_func_ref_sphinx3, 87 RE_struct: markup_c_ref, 88 RE_union: markup_c_ref, 89 RE_enum: markup_c_ref, 90 RE_typedef: markup_c_ref, 91 RE_git: markup_git} 92 93 match_iterators = [regex.finditer(t) for regex in markup_func] 94 # 95 # Sort all references by the starting position in text 96 # 97 sorted_matches = sorted(chain(*match_iterators), key=lambda m: m.start()) 98 for m in sorted_matches: 99 # 100 # Include any text prior to match as a normal text node. 101 # 102 if m.start() > done: 103 repl.append(nodes.Text(t[done:m.start()])) 104 105 # 106 # Call the function associated with the regex that matched this text and 107 # append its return to the text 108 # 109 repl.append(markup_func[m.re](docname, app, m)) 110 111 done = m.end() 112 if done < len(t): 113 repl.append(nodes.Text(t[done:])) 114 return repl 115 116# 117# Keep track of cross-reference lookups that failed so we don't have to 118# do them again. 119# 120failed_lookups = { } 121def failure_seen(target): 122 return (target) in failed_lookups 123def note_failure(target): 124 failed_lookups[target] = True 125 126# 127# In sphinx3 we can cross-reference to C macro and function, each one with its 128# own C role, but both match the same regex, so we try both. 129# 130def markup_func_ref_sphinx3(docname, app, match): 131 base_target = match.group(2) 132 target_text = nodes.Text(match.group(0)) 133 possible_targets = [base_target] 134 # Check if this document has a namespace, and if so, try 135 # cross-referencing inside it first. 136 if c_namespace: 137 possible_targets.insert(0, c_namespace + "." + base_target) 138 139 if base_target not in Skipnames: 140 for target in possible_targets: 141 if (target not in Skipfuncs) and not failure_seen(target): 142 lit_text = nodes.literal(classes=['xref', 'c', 'c-func']) 143 lit_text += target_text 144 xref = add_and_resolve_xref(app, docname, 'c', 'function', 145 target, contnode=lit_text) 146 if xref: 147 return xref 148 note_failure(target) 149 150 return target_text 151 152def markup_c_ref(docname, app, match): 153 class_str = {RE_struct: 'c-struct', 154 RE_union: 'c-union', 155 RE_enum: 'c-enum', 156 RE_typedef: 'c-type', 157 } 158 reftype_str = {RE_struct: 'struct', 159 RE_union: 'union', 160 RE_enum: 'enum', 161 RE_typedef: 'type', 162 } 163 164 base_target = match.group(2) 165 target_text = nodes.Text(match.group(0)) 166 possible_targets = [base_target] 167 # Check if this document has a namespace, and if so, try 168 # cross-referencing inside it first. 169 if c_namespace: 170 possible_targets.insert(0, c_namespace + "." + base_target) 171 172 # Skip common English words that match identifier pattern but are not C code. 173 if base_target in Skipidentifiers: 174 return target_text 175 176 if base_target not in Skipnames: 177 for target in possible_targets: 178 if not (match.re == RE_function and target in Skipfuncs): 179 lit_text = nodes.literal(classes=['xref', 'c', class_str[match.re]]) 180 lit_text += target_text 181 xref = add_and_resolve_xref(app, docname, 'c', 182 reftype_str[match.re], target, 183 contnode=lit_text) 184 if xref: 185 return xref 186 187 return target_text 188 189# 190# Try to replace a documentation reference of the form Documentation/... with a 191# cross reference to that page 192# 193def markup_doc_ref(docname, app, match): 194 absolute = match.group(1) 195 target = match.group(2) 196 if absolute: 197 target = "/" + target 198 199 xref = add_and_resolve_xref(app, docname, 'std', 'doc', target) 200 if xref: 201 return xref 202 else: 203 return nodes.Text(match.group(0)) 204 205# 206# Try to replace a documentation reference for ABI symbols and files 207# with a cross reference to that page 208# 209def markup_abi_ref(docname, app, match, warning=False): 210 kernel_abi = get_kernel_abi() 211 212 fname = match.group(1) 213 target = kernel_abi.xref(fname) 214 215 # Kernel ABI doesn't describe such file or symbol 216 if not target: 217 if warning: 218 kernel_abi.log.warning("%s not found", fname) 219 return nodes.Text(match.group(0)) 220 221 xref = add_and_resolve_xref(app, docname, 'std', 'ref', target) 222 if xref: 223 return xref 224 else: 225 return nodes.Text(match.group(0)) 226 227def add_and_resolve_xref(app, docname, domain, reftype, target, contnode=None): 228 # 229 # Go through the dance of getting an xref out of the corresponding domain 230 # 231 dom_obj = app.env.domains[domain] 232 pxref = addnodes.pending_xref('', refdomain = domain, reftype = reftype, 233 reftarget = target, modname = None, 234 classname = None, refexplicit = False) 235 236 # 237 # XXX The Latex builder will throw NoUri exceptions here, 238 # work around that by ignoring them. 239 # 240 try: 241 xref = dom_obj.resolve_xref(app.env, docname, app.builder, reftype, 242 target, pxref, contnode) 243 except NoUri: 244 xref = None 245 246 if xref: 247 return xref 248 # 249 # We didn't find the xref; if a container node was supplied, 250 # mark it as a broken xref 251 # 252 if contnode: 253 contnode['classes'].append("broken_xref") 254 return contnode 255 256# 257# Variant of markup_abi_ref() that warns when a reference is not found 258# 259def markup_abi_file_ref(docname, app, match): 260 return markup_abi_ref(docname, app, match, warning=True) 261 262 263def get_c_namespace(app, docname): 264 source = app.env.doc2path(docname) 265 with open(source) as f: 266 for l in f: 267 match = RE_namespace.search(l) 268 if match: 269 return match.group(1) 270 return '' 271 272def markup_git(docname, app, match): 273 # While we could probably assume that we are running in a git 274 # repository, we can't know for sure, so let's just mechanically 275 # turn them into git.kernel.org links without checking their 276 # validity. (Maybe we can do something in the future to warn about 277 # these references if this is explicitly requested.) 278 text = match.group(0) 279 rev = match.group('rev') 280 return nodes.reference('', nodes.Text(text), 281 refuri=f'https://git.kernel.org/torvalds/c/{rev}') 282 283def auto_markup(app, doctree, name): 284 global c_namespace 285 c_namespace = get_c_namespace(app, name) 286 def text_but_not_a_reference(node): 287 # The nodes.literal test catches ``literal text``, its purpose is to 288 # avoid adding cross-references to functions that have been explicitly 289 # marked with cc:func:. 290 if not isinstance(node, nodes.Text) or isinstance(node.parent, nodes.literal): 291 return False 292 293 child_of_reference = False 294 parent = node.parent 295 while parent: 296 if isinstance(parent, nodes.Referential): 297 child_of_reference = True 298 break 299 parent = parent.parent 300 return not child_of_reference 301 302 # 303 # This loop could eventually be improved on. Someday maybe we 304 # want a proper tree traversal with a lot of awareness of which 305 # kinds of nodes to prune. But this works well for now. 306 # 307 for para in doctree.traverse(nodes.paragraph): 308 for node in para.traverse(condition=text_but_not_a_reference): 309 node.parent.replace(node, markup_refs(name, app, node)) 310 311def setup(app): 312 app.connect('doctree-resolved', auto_markup) 313 return { 314 'parallel_read_safe': True, 315 'parallel_write_safe': True, 316 } 317