s3-docs: Fix typos.
[Samba.git] / buildtools / wafsamba / symbols.py
blob0d0af79d06638c29dd0bc069b5e7d402782ceee5
1 # a waf tool to extract symbols from object files or libraries
2 # using nm, producing a set of exposed defined/undefined symbols
4 import Utils, Build, subprocess, Logs
5 from samba_wildcard import fake_build_environment
6 from samba_utils import *
8 # these are the data structures used in symbols.py:
10 # bld.env.symbol_map : dictionary mapping public symbol names to list of
11 # subsystem names where that symbol exists
13 # t.in_library : list of libraries that t is in
15 # bld.env.public_symbols: set of public symbols for each subsystem
16 # bld.env.used_symbols : set of used symbols for each subsystem
18 # bld.env.syslib_symbols: dictionary mapping system library name to set of symbols
19 # for that library
21 # LOCAL_CACHE(bld, 'TARGET_TYPE') : dictionary mapping subsystem name to target type
23 def symbols_extract(objfiles, dynamic=False):
24 '''extract symbols from objfile, returning a dictionary containing
25 the set of undefined and public symbols for each file'''
27 ret = {}
29 cmd = ["nm"]
30 if dynamic:
31 # needed for some .so files
32 cmd.append("-D")
33 cmd.extend(objfiles)
35 nmpipe = subprocess.Popen(cmd, stdout=subprocess.PIPE).stdout
36 if len(objfiles) == 1:
37 filename = objfiles[0]
38 ret[filename] = { "PUBLIC": set(), "UNDEFINED" : set()}
40 for line in nmpipe:
41 line = line.strip()
42 if line.endswith(':'):
43 filename = line[:-1]
44 ret[filename] = { "PUBLIC": set(), "UNDEFINED" : set() }
45 continue
46 cols = line.split(" ")
47 if cols == ['']:
48 continue
49 # see if the line starts with an address
50 if len(cols) == 3:
51 symbol_type = cols[1]
52 symbol = cols[2]
53 else:
54 symbol_type = cols[0]
55 symbol = cols[1]
56 if symbol_type in "BDGTRVWSi":
57 # its a public symbol
58 ret[filename]["PUBLIC"].add(symbol)
59 elif symbol_type in "U":
60 ret[filename]["UNDEFINED"].add(symbol)
62 return ret
65 def real_name(name):
66 if name.find(".objlist") != -1:
67 name = name[:-8]
68 return name
71 def find_syslib_path(bld, libname, deps):
72 '''find the path to the syslib we will link against'''
73 # the strategy is to use the targets that depend on the library, and run ldd
74 # on it to find the real location of the library that is used
76 linkpath = deps[0].link_task.outputs[0].abspath(bld.env)
78 if libname == "python":
79 libname += bld.env.PYTHON_VERSION
81 ret = None
83 lddpipe = subprocess.Popen(['ldd', linkpath], stdout=subprocess.PIPE).stdout
84 for line in lddpipe:
85 line = line.strip()
86 cols = line.split(" ")
87 if len(cols) < 3 or cols[1] != "=>":
88 continue
89 if cols[0].startswith("lib%s." % libname.lower()):
90 ret = cols[2]
91 if cols[0].startswith("libc."):
92 # save this one too
93 bld.env.libc_path = cols[2]
94 return ret
97 def build_symbol_sets(bld, tgt_list):
98 '''build the public_symbols and undefined_symbols attributes for each target'''
100 if bld.env.public_symbols:
101 return
103 objlist = [] # list of object file
104 objmap = {} # map from object filename to target (subsystem) name
106 for t in tgt_list:
107 t.public_symbols = set()
108 t.undefined_symbols = set()
109 t.used_symbols = set()
110 for tsk in getattr(t, 'compiled_tasks', []):
111 for output in tsk.outputs:
112 objpath = output.abspath(bld.env)
113 objlist.append(objpath)
114 objmap[objpath] = t
116 symbols = symbols_extract(objlist)
117 for obj in objlist:
118 t = objmap[obj]
119 t.public_symbols = t.public_symbols.union(symbols[obj]["PUBLIC"])
120 t.undefined_symbols = t.undefined_symbols.union(symbols[obj]["UNDEFINED"])
121 t.used_symbols = t.used_symbols.union(symbols[obj]["UNDEFINED"])
123 t.undefined_symbols = t.undefined_symbols.difference(t.public_symbols)
125 # and the reverse map of public symbols to subsystem name
126 bld.env.symbol_map = {}
128 for t in tgt_list:
129 for s in t.public_symbols:
130 if not s in bld.env.symbol_map:
131 bld.env.symbol_map[s] = []
132 bld.env.symbol_map[s].append(real_name(t.sname))
134 targets = LOCAL_CACHE(bld, 'TARGET_TYPE')
136 bld.env.public_symbols = {}
137 for t in tgt_list:
138 name = real_name(t.sname)
139 if name in bld.env.public_symbols:
140 bld.env.public_symbols[name] = bld.env.public_symbols[name].union(t.public_symbols)
141 else:
142 bld.env.public_symbols[name] = t.public_symbols
143 if t.samba_type == 'LIBRARY':
144 for dep in t.add_objects:
145 t2 = bld.name_to_obj(dep, bld.env)
146 bld.ASSERT(t2 is not None, "Library '%s' has unknown dependency '%s'" % (name, dep))
147 bld.env.public_symbols[name] = bld.env.public_symbols[name].union(t2.public_symbols)
149 bld.env.used_symbols = {}
150 for t in tgt_list:
151 name = real_name(t.sname)
152 if name in bld.env.used_symbols:
153 bld.env.used_symbols[name] = bld.env.used_symbols[name].union(t.used_symbols)
154 else:
155 bld.env.used_symbols[name] = t.used_symbols
156 if t.samba_type == 'LIBRARY':
157 for dep in t.add_objects:
158 t2 = bld.name_to_obj(dep, bld.env)
159 bld.ASSERT(t2 is not None, "Library '%s' has unknown dependency '%s'" % (name, dep))
160 bld.env.used_symbols[name] = bld.env.used_symbols[name].union(t2.used_symbols)
163 def build_syslib_sets(bld, tgt_list):
164 '''build the public_symbols for all syslibs'''
166 if bld.env.syslib_symbols:
167 return
169 # work out what syslibs we depend on, and what targets those are used in
170 syslibs = {}
171 objmap = {}
172 for t in tgt_list:
173 if getattr(t, 'uselib', []) and t.samba_type in [ 'LIBRARY', 'BINARY', 'PYTHON' ]:
174 for lib in t.uselib:
175 if lib in ['PYEMBED', 'PYEXT']:
176 lib = "python"
177 if not lib in syslibs:
178 syslibs[lib] = []
179 syslibs[lib].append(t)
181 # work out the paths to each syslib
182 syslib_paths = []
183 for lib in syslibs:
184 path = find_syslib_path(bld, lib, syslibs[lib])
185 if path is None:
186 Logs.warn("Unable to find syslib path for %s" % lib)
187 if path is not None:
188 syslib_paths.append(path)
189 objmap[path] = lib.lower()
191 # add in libc
192 syslib_paths.append(bld.env.libc_path)
193 objmap[bld.env.libc_path] = 'c'
195 symbols = symbols_extract(syslib_paths, dynamic=True)
197 # keep a map of syslib names to public symbols
198 bld.env.syslib_symbols = {}
199 for lib in symbols:
200 bld.env.syslib_symbols[lib] = symbols[lib]["PUBLIC"]
202 # add to the map of symbols to dependencies
203 for lib in symbols:
204 for sym in symbols[lib]["PUBLIC"]:
205 if not sym in bld.env.symbol_map:
206 bld.env.symbol_map[sym] = []
207 bld.env.symbol_map[sym].append(objmap[lib])
209 # keep the libc symbols as well, as these are useful for some of the
210 # sanity checks
211 bld.env.libc_symbols = symbols[bld.env.libc_path]["PUBLIC"]
213 # add to the combined map of dependency name to public_symbols
214 for lib in bld.env.syslib_symbols:
215 bld.env.public_symbols[objmap[lib]] = bld.env.syslib_symbols[lib]
218 def build_autodeps(bld, t):
219 '''build the set of dependencies for a target'''
220 deps = set()
221 name = real_name(t.sname)
223 targets = LOCAL_CACHE(bld, 'TARGET_TYPE')
225 for sym in t.undefined_symbols:
226 if sym in t.public_symbols:
227 continue
228 if sym in bld.env.symbol_map:
229 depname = bld.env.symbol_map[sym]
230 if depname == [ name ]:
231 # self dependencies aren't interesting
232 continue
233 if t.in_library == depname:
234 # no need to depend on the library we are part of
235 continue
236 if depname[0] in ['c', 'python']:
237 # these don't go into autodeps
238 continue
239 if targets[depname[0]] in [ 'SYSLIB' ]:
240 deps.add(depname[0])
241 continue
242 t2 = bld.name_to_obj(depname[0], bld.env)
243 if len(t2.in_library) != 1:
244 deps.add(depname[0])
245 continue
246 if t2.in_library == t.in_library:
247 # if we're part of the same library, we don't need to autodep
248 continue
249 deps.add(t2.in_library[0])
250 t.autodeps = deps
253 def build_library_names(bld, tgt_list):
254 '''add a in_library attribute to all targets that are part of a library'''
256 if bld.env.done_build_library_names:
257 return
259 for t in tgt_list:
260 t.in_library = []
262 for t in tgt_list:
263 if t.samba_type in [ 'LIBRARY' ]:
264 for obj in t.samba_deps_extended:
265 t2 = bld.name_to_obj(obj, bld.env)
266 if t2 and t2.samba_type in [ 'SUBSYSTEM', 'ASN1' ]:
267 if not t.sname in t2.in_library:
268 t2.in_library.append(t.sname)
269 bld.env.done_build_library_names = True
272 def check_library_deps(bld, t):
273 '''check that all the autodeps that have mutual dependency of this
274 target are in the same library as the target'''
276 name = real_name(t.sname)
278 if len(t.in_library) > 1:
279 Logs.warn("WARNING: Target '%s' in multiple libraries: %s" % (t.sname, t.in_library))
281 for dep in t.autodeps:
282 t2 = bld.name_to_obj(dep, bld.env)
283 if t2 is None:
284 continue
285 for dep2 in t2.autodeps:
286 if dep2 == name and t.in_library != t2.in_library:
287 Logs.warn("WARNING: mutual dependency %s <=> %s" % (name, real_name(t2.sname)))
288 Logs.warn("Libraries should match. %s != %s" % (t.in_library, t2.in_library))
289 # raise Utils.WafError("illegal mutual dependency")
292 def check_syslib_collisions(bld, tgt_list):
293 '''check if a target has any symbol collisions with a syslib
295 We do not want any code in Samba to use a symbol name from a
296 system library. The chance of that causing problems is just too
297 high. Note that libreplace uses a rep_XX approach of renaming
298 symbols via macros
301 has_error = False
302 for t in tgt_list:
303 for lib in bld.env.syslib_symbols:
304 common = t.public_symbols.intersection(bld.env.syslib_symbols[lib])
305 if common:
306 Logs.error("ERROR: Target '%s' has symbols '%s' which is also in syslib '%s'" % (t.sname, common, lib))
307 has_error = True
308 if has_error:
309 raise Utils.WafError("symbols in common with system libraries")
312 def check_dependencies(bld, t):
313 '''check for depenencies that should be changed'''
315 if bld.name_to_obj(t.sname + ".objlist", bld.env):
316 return
318 targets = LOCAL_CACHE(bld, 'TARGET_TYPE')
320 remaining = t.undefined_symbols.copy()
321 remaining = remaining.difference(t.public_symbols)
323 sname = real_name(t.sname)
325 deps = set(t.samba_deps)
326 for d in t.samba_deps:
327 if targets[d] in [ 'EMPTY', 'DISABLED', 'SYSLIB' ]:
328 continue
329 bld.ASSERT(d in bld.env.public_symbols, "Failed to find symbol list for dependency '%s'" % d)
330 diff = remaining.intersection(bld.env.public_symbols[d])
331 if not diff and targets[sname] != 'LIBRARY':
332 Logs.info("Target '%s' has no dependency on %s" % (sname, d))
333 else:
334 remaining = remaining.difference(diff)
336 t.unsatisfied_symbols = set()
337 needed = {}
338 for sym in remaining:
339 if sym in bld.env.symbol_map:
340 dep = bld.env.symbol_map[sym]
341 if not dep[0] in needed:
342 needed[dep[0]] = set()
343 needed[dep[0]].add(sym)
344 else:
345 t.unsatisfied_symbols.add(sym)
347 for dep in needed:
348 Logs.info("Target '%s' should add dep '%s' for symbols %s" % (sname, dep, " ".join(needed[dep])))
352 def check_syslib_dependencies(bld, t):
353 '''check for syslib depenencies'''
355 if bld.name_to_obj(t.sname + ".objlist", bld.env):
356 return
358 sname = real_name(t.sname)
360 remaining = set()
362 features = TO_LIST(t.features)
363 if 'pyembed' in features or 'pyext' in features:
364 t.unsatisfied_symbols = t.unsatisfied_symbols.difference(bld.env.public_symbols['python'])
366 needed = {}
367 for sym in t.unsatisfied_symbols:
368 if sym in bld.env.symbol_map:
369 dep = bld.env.symbol_map[sym][0]
370 if dep == 'c':
371 continue
372 if not dep in needed:
373 needed[dep] = set()
374 needed[dep].add(sym)
375 else:
376 remaining.add(sym)
378 for dep in needed:
379 Logs.info("Target '%s' should add syslib dep '%s' for symbols %s" % (sname, dep, " ".join(needed[dep])))
381 if remaining:
382 debug("deps: Target '%s' has unsatisfied symbols: %s" % (sname, " ".join(remaining)))
386 def symbols_symbolcheck(task):
387 '''check the internal dependency lists'''
388 bld = task.env.bld
389 tgt_list = get_tgt_list(bld)
391 build_symbol_sets(bld, tgt_list)
392 build_library_names(bld, tgt_list)
394 for t in tgt_list:
395 t.autodeps = set()
396 if getattr(t, 'source', ''):
397 build_autodeps(bld, t)
399 for t in tgt_list:
400 check_dependencies(bld, t)
402 for t in tgt_list:
403 check_library_deps(bld, t)
405 def symbols_syslibcheck(task):
406 '''check the syslib dependencies'''
407 bld = task.env.bld
408 tgt_list = get_tgt_list(bld)
410 build_syslib_sets(bld, tgt_list)
411 check_syslib_collisions(bld, tgt_list)
413 for t in tgt_list:
414 check_syslib_dependencies(bld, t)
417 def symbols_whyneeded(task):
418 """check why 'target' needs to link to 'subsystem'"""
419 bld = task.env.bld
420 tgt_list = get_tgt_list(bld)
422 why = Options.options.WHYNEEDED.split(":")
423 if len(why) != 2:
424 raise Utils.WafError("usage: WHYNEEDED=TARGET:DEPENDENCY")
425 target = why[0]
426 subsystem = why[1]
428 build_symbol_sets(bld, tgt_list)
429 build_library_names(bld, tgt_list)
430 build_syslib_sets(bld, tgt_list)
432 Logs.info("Checking why %s needs to link to %s" % (target, subsystem))
433 if not target in bld.env.used_symbols:
434 Logs.warn("unable to find target '%s' in used_symbols dict" % target)
435 return
436 if not subsystem in bld.env.public_symbols:
437 Logs.warn("unable to find subsystem '%s' in public_symbols dict" % subsystem)
438 return
439 overlap = bld.env.used_symbols[target].intersection(bld.env.public_symbols[subsystem])
440 if not overlap:
441 Logs.info("target '%s' doesn't use any public symbols from '%s'" % (target, subsystem))
442 else:
443 Logs.info("target '%s' uses symbols %s from '%s'" % (target, overlap, subsystem))
447 def symbols_dupcheck(task):
448 '''check for symbols defined in two different subsystems'''
449 bld = task.env.bld
450 tgt_list = get_tgt_list(bld)
452 targets = LOCAL_CACHE(bld, 'TARGET_TYPE')
454 Logs.info("Checking for duplicate symbols")
455 for sym in bld.env.symbol_map:
456 subsystems = set(bld.env.symbol_map[sym])
457 if len(subsystems) == 1:
458 continue
460 if sym in ['main', '_init', '_fini', 'init_samba_module', 'samba_init_module', 'ldb_init_module' ]:
461 # these are expected to be in many subsystems
462 continue
464 # if all of them are in system libraries, we can ignore them. This copes
465 # with the duplication between libc, libpthread and libattr
466 all_syslib = True
467 for s in subsystems:
468 if s != 'c' and (not s in targets or targets[s] != 'SYSLIB'):
469 all_syslib = False
470 if all_syslib:
471 continue
472 Logs.info("symbol %s appears in %s" % (sym, subsystems))
475 def SYMBOL_CHECK(bld):
476 '''check our dependency lists'''
477 if Options.options.SYMBOLCHECK:
478 bld.SET_BUILD_GROUP('symbolcheck')
479 task = bld(rule=symbols_symbolcheck, always=True, name='symbol checking')
480 task.env.bld = bld
482 bld.SET_BUILD_GROUP('syslibcheck')
483 task = bld(rule=symbols_syslibcheck, always=True, name='syslib checking')
484 task.env.bld = bld
486 bld.SET_BUILD_GROUP('syslibcheck')
487 task = bld(rule=symbols_dupcheck, always=True, name='symbol duplicate checking')
488 task.env.bld = bld
490 if Options.options.WHYNEEDED:
491 bld.SET_BUILD_GROUP('syslibcheck')
492 task = bld(rule=symbols_whyneeded, always=True, name='check why a dependency is needed')
493 task.env.bld = bld
496 Build.BuildContext.SYMBOL_CHECK = SYMBOL_CHECK