build: duplicate symbols between system libraries aren't interesting
[samba.git] / buildtools / wafsamba / symbols.py
index dfa3a8c61d40742329a7acb059f0c634e44a0cab..0408d61b297eb8b5484edc07a1372201ed810ac0 100644 (file)
@@ -5,6 +5,21 @@ import Utils, Build, subprocess, Logs
 from samba_wildcard import fake_build_environment
 from samba_utils import *
 
+# these are the data structures used in symbols.py:
+#
+# bld.env.symbol_map : dictionary mapping public symbol names to list of
+#                      subsystem names where that symbol exists
+#
+# t.in_library       : list of libraries that t is in
+#
+# bld.env.public_symbols: set of public symbols for each subsystem
+# bld.env.used_symbols  : set of used symbols for each subsystem
+#
+# bld.env.syslib_symbols: dictionary mapping system library name to set of symbols
+#                         for that library
+#
+# LOCAL_CACHE(bld, 'TARGET_TYPE') : dictionary mapping subsystem name to target type
+
 def symbols_extract(objfiles, dynamic=False):
     '''extract symbols from objfile, returning a dictionary containing
        the set of undefined and public symbols for each file'''
@@ -38,7 +53,7 @@ def symbols_extract(objfiles, dynamic=False):
         else:
             symbol_type = cols[0]
             symbol = cols[1]
-        if symbol_type in "BDGTRVWS":
+        if symbol_type in "BDGTRVWSi":
             # its a public symbol
             ret[filename]["PUBLIC"].add(symbol)
         elif symbol_type in "U":
@@ -82,13 +97,16 @@ def find_syslib_path(bld, libname, deps):
 def build_symbol_sets(bld, tgt_list):
     '''build the public_symbols and undefined_symbols attributes for each target'''
 
-    objlist = []  # list of object file
-    objmap = {}   # map from object filename to target
+    if bld.env.public_symbols:
+        return
 
+    objlist = []  # list of object file
+    objmap = {}   # map from object filename to target (subsystem) name
 
     for t in tgt_list:
         t.public_symbols = set()
         t.undefined_symbols = set()
+        t.used_symbols = set()
         for tsk in getattr(t, 'compiled_tasks', []):
             for output in tsk.outputs:
                 objpath = output.abspath(bld.env)
@@ -99,7 +117,8 @@ def build_symbol_sets(bld, tgt_list):
     for obj in objlist:
         t = objmap[obj]
         t.public_symbols = t.public_symbols.union(symbols[obj]["PUBLIC"])
-        t.undefined_symbols = t.public_symbols.union(symbols[obj]["UNDEFINED"])
+        t.undefined_symbols = t.undefined_symbols.union(symbols[obj]["UNDEFINED"])
+        t.used_symbols = t.used_symbols.union(symbols[obj]["UNDEFINED"])
 
     t.undefined_symbols = t.undefined_symbols.difference(t.public_symbols)
 
@@ -108,9 +127,44 @@ def build_symbol_sets(bld, tgt_list):
 
     for t in tgt_list:
         for s in t.public_symbols:
-            bld.env.symbol_map[s] = real_name(t.sname)
+            if not s in bld.env.symbol_map:
+                bld.env.symbol_map[s] = []
+            bld.env.symbol_map[s].append(real_name(t.sname))
+
+    targets = LOCAL_CACHE(bld, 'TARGET_TYPE')
+
+    bld.env.public_symbols = {}
+    for t in tgt_list:
+        name = real_name(t.sname)
+        if name in bld.env.public_symbols:
+            bld.env.public_symbols[name] = bld.env.public_symbols[name].union(t.public_symbols)
+        else:
+            bld.env.public_symbols[name] = t.public_symbols
+        if t.samba_type == 'LIBRARY':
+            for dep in t.add_objects:
+                t2 = bld.name_to_obj(dep, bld.env)
+                bld.ASSERT(t2 is not None, "Library '%s' has unknown dependency '%s'" % (name, dep))
+                bld.env.public_symbols[name] = bld.env.public_symbols[name].union(t2.public_symbols)
+
+    bld.env.used_symbols = {}
+    for t in tgt_list:
+        name = real_name(t.sname)
+        if name in bld.env.used_symbols:
+            bld.env.used_symbols[name] = bld.env.used_symbols[name].union(t.used_symbols)
+        else:
+            bld.env.used_symbols[name] = t.used_symbols
+        if t.samba_type == 'LIBRARY':
+            for dep in t.add_objects:
+                t2 = bld.name_to_obj(dep, bld.env)
+                bld.ASSERT(t2 is not None, "Library '%s' has unknown dependency '%s'" % (name, dep))
+                bld.env.used_symbols[name] = bld.env.used_symbols[name].union(t2.used_symbols)
+
+
+def build_syslib_sets(bld, tgt_list):
+    '''build the public_symbols for all syslibs'''
 
-    # now do the same for syslibs
+    if bld.env.syslib_symbols:
+        return
 
     # work out what syslibs we depend on, and what targets those are used in
     syslibs = {}
@@ -129,10 +183,10 @@ def build_symbol_sets(bld, tgt_list):
     for lib in syslibs:
         path = find_syslib_path(bld, lib, syslibs[lib])
         if path is None:
-            print("Unable to find syslib path for %s used by %s" % lib)
+            Logs.warn("Unable to find syslib path for %s" % lib)
         if path is not None:
             syslib_paths.append(path)
-            objmap[path] = lib
+            objmap[path] = lib.lower()
 
     # add in libc
     syslib_paths.append(bld.env.libc_path)
@@ -148,21 +202,17 @@ def build_symbol_sets(bld, tgt_list):
     # add to the map of symbols to dependencies
     for lib in symbols:
         for sym in symbols[lib]["PUBLIC"]:
-            bld.env.symbol_map[sym] = objmap[lib].lower()
+            if not sym in bld.env.symbol_map:
+                bld.env.symbol_map[sym] = []
+            bld.env.symbol_map[sym].append(objmap[lib])
 
     # keep the libc symbols as well, as these are useful for some of the
     # sanity checks
     bld.env.libc_symbols = symbols[bld.env.libc_path]["PUBLIC"]
 
-    # a combined map of dependency name to public_symbols
-    bld.env.all_symbols = {}
+    # add to the combined map of dependency name to public_symbols
     for lib in bld.env.syslib_symbols:
-        bld.env.all_symbols[lib] = bld.env.syslib_symbols[lib]
-    for t in tgt_list:
-        name = real_name(t.sname)
-        if bld.name_to_obj(t.sname + '.objlist', bld.env):
-            continue
-        bld.env.all_symbols[name] = t.public_symbols
+        bld.env.public_symbols[objmap[lib]] = bld.env.syslib_symbols[lib]
 
 
 def build_autodeps(bld, t):
@@ -177,32 +227,35 @@ def build_autodeps(bld, t):
             continue
         if sym in bld.env.symbol_map:
             depname = bld.env.symbol_map[sym]
-            if depname == name:
+            if depname == [ name ]:
                 # self dependencies aren't interesting
                 continue
-            if t.in_library == [depname]:
+            if t.in_library == depname:
                 # no need to depend on the library we are part of
                 continue
-            if depname in ['c', 'python']:
+            if depname[0] in ['c', 'python']:
                 # these don't go into autodeps
                 continue
-            if targets[depname] in [ 'SYSLIB' ]:
-                deps.add(depname)
+            if targets[depname[0]] in [ 'SYSLIB' ]:
+                deps.add(depname[0])
                 continue
-            t2 = bld.name_to_obj(depname, bld.env)
+            t2 = bld.name_to_obj(depname[0], bld.env)
             if len(t2.in_library) != 1:
-                deps.add(depname)
+                deps.add(depname[0])
                 continue
             if t2.in_library == t.in_library:
                 # if we're part of the same library, we don't need to autodep
                 continue
-            print("adding library %s for symbol %s" % (t2.in_library[0], sym))
             deps.add(t2.in_library[0])
     t.autodeps = deps
 
 
 def build_library_names(bld, tgt_list):
     '''add a in_library attribute to all targets that are part of a library'''
+
+    if bld.env.done_build_library_names:
+        return
+
     for t in tgt_list:
         t.in_library = []
 
@@ -211,7 +264,9 @@ def build_library_names(bld, tgt_list):
             for obj in t.samba_deps_extended:
                 t2 = bld.name_to_obj(obj, bld.env)
                 if t2 and t2.samba_type in [ 'SUBSYSTEM', 'ASN1' ]:
-                    t2.in_library.append(t.sname)
+                    if not t.sname in t2.in_library:
+                        t2.in_library.append(t.sname)
+    bld.env.done_build_library_names = True
 
 
 def check_library_deps(bld, t):
@@ -229,9 +284,9 @@ def check_library_deps(bld, t):
             continue
         for dep2 in t2.autodeps:
             if dep2 == name and t.in_library != t2.in_library:
-                Logs.error("Illegal mutual dependency %s <=> %s" % (name, real_name(t2.sname)))
-                Logs.error("Libraries must match. %s != %s" % (t.in_library, t2.in_library))
-                sys.exit(1)
+                Logs.warn("WARNING: mutual dependency %s <=> %s" % (name, real_name(t2.sname)))
+                Logs.warn("Libraries should match. %s != %s" % (t.in_library, t2.in_library))
+                # raise Utils.WafError("illegal mutual dependency")
 
 
 def check_syslib_collisions(bld, tgt_list):
@@ -254,55 +309,183 @@ def check_syslib_collisions(bld, tgt_list):
         raise Utils.WafError("symbols in common with system libraries")
 
 
-def check_dep_list(bld, t):
-    '''check for depenencies that can be removed'''
+def check_dependencies(bld, t):
+    '''check for depenencies that should be changed'''
+
     if bld.name_to_obj(t.sname + ".objlist", bld.env):
         return
+
+    targets = LOCAL_CACHE(bld, 'TARGET_TYPE')
+
+    remaining = t.undefined_symbols.copy()
+    remaining = remaining.difference(t.public_symbols)
+
+    sname = real_name(t.sname)
+
     deps = set(t.samba_deps)
-    diff = deps.difference(t.autodeps)
-    for d in ['replace']:
-        if d in diff:
-            diff.remove(d)
-    if diff:
-        Logs.info("Target '%s' could remove deps: %s" % (real_name(t.sname), " ".join(diff)))
-    diff = t.autodeps.difference(deps)
-    for d in diff:
-        Logs.info("Target '%s' should add dep '%s' for symbols %s" % (
-            real_name(t.sname), d, t.undefined_symbols.intersection(bld.env.all_symbols[d])))
-
-
-def symbols_autodep(task):
-    '''check the dependency lists'''
+    for d in t.samba_deps:
+        if targets[d] in [ 'EMPTY', 'DISABLED', 'SYSLIB' ]:
+            continue
+        bld.ASSERT(d in bld.env.public_symbols, "Failed to find symbol list for dependency '%s'" % d)
+        diff = remaining.intersection(bld.env.public_symbols[d])
+        if not diff and targets[sname] != 'LIBRARY':
+            Logs.info("Target '%s' has no dependency on %s" % (sname, d))
+        else:
+            remaining = remaining.difference(diff)
+
+    t.unsatisfied_symbols = set()
+    needed = {}
+    for sym in remaining:
+        if sym in bld.env.symbol_map:
+            dep = bld.env.symbol_map[sym]
+            if not dep[0] in needed:
+                needed[dep[0]] = set()
+            needed[dep[0]].add(sym)
+        else:
+            t.unsatisfied_symbols.add(sym)
+
+    for dep in needed:
+        Logs.info("Target '%s' should add dep '%s' for symbols %s" % (sname, dep, " ".join(needed[dep])))
+
+
+
+def check_syslib_dependencies(bld, t):
+    '''check for syslib depenencies'''
+
+    if bld.name_to_obj(t.sname + ".objlist", bld.env):
+        return
+
+    sname = real_name(t.sname)
+
+    remaining = set()
+
+    features = TO_LIST(t.features)
+    if 'pyembed' in features or 'pyext' in features:
+        t.unsatisfied_symbols = t.unsatisfied_symbols.difference(bld.env.public_symbols['python'])
+
+    needed = {}
+    for sym in t.unsatisfied_symbols:
+        if sym in bld.env.symbol_map:
+            dep = bld.env.symbol_map[sym][0]
+            if dep == 'c':
+                continue
+            if not dep in needed:
+                needed[dep] = set()
+            needed[dep].add(sym)
+        else:
+            remaining.add(sym)
+
+    for dep in needed:
+        Logs.info("Target '%s' should add syslib dep '%s' for symbols %s" % (sname, dep, " ".join(needed[dep])))
+
+    if remaining:
+        debug("deps: Target '%s' has unsatisfied symbols: %s" % (sname, " ".join(remaining)))
+
+
+
+def symbols_symbolcheck(task):
+    '''check the internal dependency lists'''
     bld = task.env.bld
     tgt_list = get_tgt_list(bld)
 
     build_symbol_sets(bld, tgt_list)
     build_library_names(bld, tgt_list)
 
-    t = bld.name_to_obj('SERVICE_SMB', bld.env)
-    build_autodeps(bld, t)
-    check_dep_list(bld, t)
-    return
-
     for t in tgt_list:
         t.autodeps = set()
         if getattr(t, 'source', ''):
             build_autodeps(bld, t)
 
+    for t in tgt_list:
+        check_dependencies(bld, t)
+
     for t in tgt_list:
         check_library_deps(bld, t)
 
-    check_syslib_collisions(bld, tgt_list)
+def symbols_syslibcheck(task):
+    '''check the syslib dependencies'''
+    bld = task.env.bld
+    tgt_list = get_tgt_list(bld)
 
+    build_syslib_sets(bld, tgt_list)
+    check_syslib_collisions(bld, tgt_list)
 
     for t in tgt_list:
-        check_dep_list(bld, t)
+        check_syslib_dependencies(bld, t)
+
+
+def symbols_whyneeded(task):
+    """check why 'target' needs to link to 'subsystem'"""
+    bld = task.env.bld
+    tgt_list = get_tgt_list(bld)
+
+    why = Options.options.WHYNEEDED.split(":")
+    if len(why) != 2:
+        raise Utils.WafError("usage: WHYNEEDED=TARGET:DEPENDENCY")
+    target = why[0]
+    subsystem = why[1]
+
+    build_symbol_sets(bld, tgt_list)
+    build_library_names(bld, tgt_list)
+    build_syslib_sets(bld, tgt_list)
+
+    Logs.info("Checking why %s needs to link to %s" % (target, subsystem))
+    if not target in bld.env.used_symbols:
+        Logs.warn("unable to find target '%s' in used_symbols dict" % target)
+        return
+    if not subsystem in bld.env.public_symbols:
+        Logs.warn("unable to find subsystem '%s' in public_symbols dict" % subsystem)
+        return
+    overlap = bld.env.used_symbols[target].intersection(bld.env.public_symbols[subsystem])
+    if not overlap:
+        Logs.info("target '%s' doesn't use any public symbols from '%s'" % (target, subsystem))
+    else:
+        Logs.info("target '%s' uses symbols %s from '%s'" % (target, overlap, subsystem))
+
 
 
-def AUTODEP(bld):
+def symbols_dupcheck(task):
+    '''check for symbols defined in two different subsystems'''
+    bld = task.env.bld
+    tgt_list = get_tgt_list(bld)
+
+    targets = LOCAL_CACHE(bld, 'TARGET_TYPE')
+
+    Logs.info("Checking for duplicate symbols")
+    for sym in bld.env.symbol_map:
+        subsystems = bld.env.symbol_map[sym]
+        if len(subsystems) == 1:
+            continue
+        # if all of them are in system libraries, we can ignore them. This copes
+        # with the duplication between libc, libpthread and libattr
+        all_syslib = True
+        for s in subsystems:
+            if s != 'c' and (not s in targets or targets[s] != 'SYSLIB'):
+                all_syslib = False
+        if all_syslib:
+            continue
+        Logs.info("symbol %s appears in %s" % (sym, subsystems))
+
+
+def SYMBOL_CHECK(bld):
     '''check our dependency lists'''
-    if bld.env.DEVELOPER_MODE:
-        bld.SET_BUILD_GROUP('final')
-        task = bld(rule=symbols_autodep, always=True, name='Autodep')
+    if Options.options.SYMBOLCHECK:
+        bld.SET_BUILD_GROUP('symbolcheck')
+        task = bld(rule=symbols_symbolcheck, always=True, name='symbol checking')
+        task.env.bld = bld
+
+        bld.SET_BUILD_GROUP('syslibcheck')
+        task = bld(rule=symbols_syslibcheck, always=True, name='syslib checking')
+        task.env.bld = bld
+
+        bld.SET_BUILD_GROUP('syslibcheck')
+        task = bld(rule=symbols_dupcheck, always=True, name='symbol duplicate checking')
+        task.env.bld = bld
+
+    if Options.options.WHYNEEDED:
+        bld.SET_BUILD_GROUP('syslibcheck')
+        task = bld(rule=symbols_whyneeded, always=True, name='check why a dependency is needed')
         task.env.bld = bld
-Build.BuildContext.AUTODEP = AUTODEP
+
+
+Build.BuildContext.SYMBOL_CHECK = SYMBOL_CHECK