mirror of
https://github.com/samba-team/samba.git
synced 2024-12-23 17:34:34 +03:00
35c7061f97
This avoids errors like the following: ‘RuntimeError: No active exception to reraise’ Signed-off-by: Joseph Sutton <josephsutton@catalyst.net.nz> Reviewed-by: Andrew Bartlett <abartlet@samba.org>
755 lines
24 KiB
Python
755 lines
24 KiB
Python
# a waf tool to add autoconf-like macros to the configure section
|
|
# and for SAMBA_ macros for building libraries, binaries etc
|
|
|
|
import errno
|
|
import os, sys, re, fnmatch, shlex, inspect
|
|
from optparse import SUPPRESS_HELP
|
|
from waflib import Build, Options, Utils, Task, Logs, Configure, Errors, Context
|
|
from waflib import Scripting
|
|
from waflib.TaskGen import feature, before, after
|
|
from waflib.Configure import ConfigurationContext
|
|
from waflib.Logs import debug
|
|
from waflib import ConfigSet
|
|
from waflib.Build import CACHE_SUFFIX
|
|
|
|
# TODO: make this a --option
|
|
LIB_PATH="shared"
|
|
|
|
|
|
PY3 = sys.version_info[0] == 3
|
|
|
|
if PY3:
|
|
|
|
# helper function to get a string from a variable that maybe 'str' or
|
|
# 'bytes' if 'bytes' then it is decoded using 'utf8'. If 'str' is passed
|
|
# it is returned unchanged
|
|
# Using this function is PY2/PY3 code should ensure in most cases
|
|
# the PY2 code runs unchanged in PY2 whereas the code in PY3 possibly
|
|
# decodes the variable (see PY2 implementation of this function below)
|
|
def get_string(bytesorstring):
|
|
tmp = bytesorstring
|
|
if isinstance(bytesorstring, bytes):
|
|
tmp = bytesorstring.decode('utf8')
|
|
elif not isinstance(bytesorstring, str):
|
|
raise ValueError('Expected byte of string for %s:%s' % (type(bytesorstring), bytesorstring))
|
|
return tmp
|
|
|
|
else:
|
|
|
|
# Helper function to return string.
|
|
# if 'str' or 'unicode' passed in they are returned unchanged
|
|
# otherwise an exception is generated
|
|
# Using this function is PY2/PY3 code should ensure in most cases
|
|
# the PY2 code runs unchanged in PY2 whereas the code in PY3 possibly
|
|
# decodes the variable (see PY3 implementation of this function above)
|
|
def get_string(bytesorstring):
|
|
tmp = bytesorstring
|
|
if not(isinstance(bytesorstring, str) or isinstance(bytesorstring, unicode)):
|
|
raise ValueError('Expected str or unicode for %s:%s' % (type(bytesorstring), bytesorstring))
|
|
return tmp
|
|
|
|
# sigh, python octal constants are a mess
|
|
MODE_644 = int('644', 8)
|
|
MODE_744 = int('744', 8)
|
|
MODE_755 = int('755', 8)
|
|
MODE_777 = int('777', 8)
|
|
|
|
def conf(f):
|
|
# override in order to propagate the argument "mandatory"
|
|
def fun(*k, **kw):
|
|
mandatory = True
|
|
if 'mandatory' in kw:
|
|
mandatory = kw['mandatory']
|
|
del kw['mandatory']
|
|
|
|
try:
|
|
return f(*k, **kw)
|
|
except Errors.ConfigurationError:
|
|
if mandatory:
|
|
raise
|
|
|
|
fun.__name__ = f.__name__
|
|
if 'mandatory' in inspect.getsource(f):
|
|
fun = f
|
|
|
|
setattr(Configure.ConfigurationContext, f.__name__, fun)
|
|
setattr(Build.BuildContext, f.__name__, fun)
|
|
return f
|
|
Configure.conf = conf
|
|
Configure.conftest = conf
|
|
|
|
@conf
|
|
def SET_TARGET_TYPE(ctx, target, value):
|
|
'''set the target type of a target'''
|
|
cache = LOCAL_CACHE(ctx, 'TARGET_TYPE')
|
|
if target in cache and cache[target] != 'EMPTY':
|
|
Logs.error("ERROR: Target '%s' in directory %s re-defined as %s - was %s" % (target, ctx.path.abspath(), value, cache[target]))
|
|
sys.exit(1)
|
|
LOCAL_CACHE_SET(ctx, 'TARGET_TYPE', target, value)
|
|
debug("task_gen: Target '%s' created of type '%s' in %s" % (target, value, ctx.path.abspath()))
|
|
return True
|
|
|
|
|
|
def GET_TARGET_TYPE(ctx, target):
|
|
'''get target type from cache'''
|
|
cache = LOCAL_CACHE(ctx, 'TARGET_TYPE')
|
|
if not target in cache:
|
|
return None
|
|
return cache[target]
|
|
|
|
|
|
def ADD_LD_LIBRARY_PATH(path):
|
|
'''add something to LD_LIBRARY_PATH'''
|
|
if 'LD_LIBRARY_PATH' in os.environ:
|
|
oldpath = os.environ['LD_LIBRARY_PATH']
|
|
else:
|
|
oldpath = ''
|
|
newpath = oldpath.split(':')
|
|
if not path in newpath:
|
|
newpath.append(path)
|
|
os.environ['LD_LIBRARY_PATH'] = ':'.join(newpath)
|
|
|
|
|
|
def needs_private_lib(bld, target):
|
|
'''return True if a target links to a private library'''
|
|
for lib in getattr(target, "final_libs", []):
|
|
t = bld.get_tgen_by_name(lib)
|
|
if t and getattr(t, 'private_library', False):
|
|
return True
|
|
return False
|
|
|
|
|
|
def install_rpath(target):
|
|
'''the rpath value for installation'''
|
|
bld = target.bld
|
|
bld.env['RPATH'] = []
|
|
ret = set()
|
|
if bld.env.RPATH_ON_INSTALL:
|
|
ret.add(bld.EXPAND_VARIABLES(bld.env.LIBDIR))
|
|
if bld.env.RPATH_ON_INSTALL_PRIVATE and needs_private_lib(bld, target):
|
|
ret.add(bld.EXPAND_VARIABLES(bld.env.PRIVATELIBDIR))
|
|
return list(ret)
|
|
|
|
|
|
def build_rpath(bld):
|
|
'''the rpath value for build'''
|
|
rpaths = [os.path.normpath('%s/%s' % (bld.env.BUILD_DIRECTORY, d)) for d in ("shared", "shared/private")]
|
|
bld.env['RPATH'] = []
|
|
if bld.env.RPATH_ON_BUILD:
|
|
return rpaths
|
|
for rpath in rpaths:
|
|
ADD_LD_LIBRARY_PATH(rpath)
|
|
return []
|
|
|
|
|
|
@conf
|
|
def LOCAL_CACHE(ctx, name):
|
|
'''return a named build cache dictionary, used to store
|
|
state inside other functions'''
|
|
if name in ctx.env:
|
|
return ctx.env[name]
|
|
ctx.env[name] = {}
|
|
return ctx.env[name]
|
|
|
|
|
|
@conf
|
|
def LOCAL_CACHE_SET(ctx, cachename, key, value):
|
|
'''set a value in a local cache'''
|
|
cache = LOCAL_CACHE(ctx, cachename)
|
|
cache[key] = value
|
|
|
|
|
|
@conf
|
|
def ASSERT(ctx, expression, msg):
|
|
'''a build assert call'''
|
|
if not expression:
|
|
raise Errors.WafError("ERROR: %s\n" % msg)
|
|
Build.BuildContext.ASSERT = ASSERT
|
|
|
|
|
|
def SUBDIR(bld, subdir, list):
|
|
'''create a list of files by pre-pending each with a subdir name'''
|
|
ret = ''
|
|
for l in TO_LIST(list):
|
|
ret = ret + os.path.normpath(os.path.join(subdir, l)) + ' '
|
|
return ret
|
|
Build.BuildContext.SUBDIR = SUBDIR
|
|
|
|
|
|
def dict_concat(d1, d2):
|
|
'''concatenate two dictionaries d1 += d2'''
|
|
for t in d2:
|
|
if t not in d1:
|
|
d1[t] = d2[t]
|
|
|
|
def ADD_COMMAND(opt, name, function):
|
|
'''add a new top level command to waf'''
|
|
Context.g_module.__dict__[name] = function
|
|
opt.name = function
|
|
Options.OptionsContext.ADD_COMMAND = ADD_COMMAND
|
|
|
|
|
|
@feature('c', 'cc', 'cshlib', 'cprogram')
|
|
@before('apply_core','exec_rule')
|
|
def process_depends_on(self):
|
|
'''The new depends_on attribute for build rules
|
|
allow us to specify a dependency on output from
|
|
a source generation rule'''
|
|
if getattr(self , 'depends_on', None):
|
|
lst = self.to_list(self.depends_on)
|
|
for x in lst:
|
|
y = self.bld.get_tgen_by_name(x)
|
|
self.bld.ASSERT(y is not None, "Failed to find dependency %s of %s" % (x, self.name))
|
|
y.post()
|
|
if getattr(y, 'more_includes', None):
|
|
self.includes += " " + y.more_includes
|
|
|
|
|
|
def unique_list(seq):
|
|
'''return a uniquified list in the same order as the existing list'''
|
|
seen = {}
|
|
result = []
|
|
for item in seq:
|
|
if item in seen: continue
|
|
seen[item] = True
|
|
result.append(item)
|
|
return result
|
|
|
|
|
|
def TO_LIST(str, delimiter=None):
|
|
'''Split a list, preserving quoted strings and existing lists'''
|
|
if str is None:
|
|
return []
|
|
if isinstance(str, list):
|
|
# we need to return a new independent list...
|
|
return list(str)
|
|
if len(str) == 0:
|
|
return []
|
|
lst = str.split(delimiter)
|
|
# the string may have had quotes in it, now we
|
|
# check if we did have quotes, and use the slower shlex
|
|
# if we need to
|
|
for e in lst:
|
|
if e[0] == '"':
|
|
return shlex.split(str)
|
|
return lst
|
|
|
|
|
|
def subst_vars_error(string, env):
|
|
'''substitute vars, throw an error if a variable is not defined'''
|
|
lst = re.split(r'(\$\{\w+\})', string)
|
|
out = []
|
|
for v in lst:
|
|
if re.match(r'\$\{\w+\}', v):
|
|
vname = v[2:-1]
|
|
if not vname in env:
|
|
raise KeyError("Failed to find variable %s in %s in env %s <%s>" % (vname, string, env.__class__, str(env)))
|
|
v = env[vname]
|
|
if isinstance(v, list):
|
|
v = ' '.join(v)
|
|
out.append(v)
|
|
return ''.join(out)
|
|
|
|
|
|
@conf
|
|
def SUBST_ENV_VAR(ctx, varname):
|
|
'''Substitute an environment variable for any embedded variables'''
|
|
return subst_vars_error(ctx.env[varname], ctx.env)
|
|
Build.BuildContext.SUBST_ENV_VAR = SUBST_ENV_VAR
|
|
|
|
|
|
def recursive_dirlist(dir, relbase, pattern=None):
|
|
'''recursive directory list'''
|
|
ret = []
|
|
for f in os.listdir(dir):
|
|
f2 = dir + '/' + f
|
|
if os.path.isdir(f2):
|
|
ret.extend(recursive_dirlist(f2, relbase))
|
|
else:
|
|
if pattern and not fnmatch.fnmatch(f, pattern):
|
|
continue
|
|
ret.append(os.path.relpath(f2, relbase))
|
|
return ret
|
|
|
|
|
|
def symlink(src, dst, force=True):
|
|
"""Can create symlink by force"""
|
|
try:
|
|
os.symlink(src, dst)
|
|
except OSError as exc:
|
|
if exc.errno == errno.EEXIST and force:
|
|
os.remove(dst)
|
|
os.symlink(src, dst)
|
|
else:
|
|
raise
|
|
|
|
|
|
def mkdir_p(dir):
|
|
'''like mkdir -p'''
|
|
if not dir:
|
|
return
|
|
if dir.endswith("/"):
|
|
mkdir_p(dir[:-1])
|
|
return
|
|
if os.path.isdir(dir):
|
|
return
|
|
mkdir_p(os.path.dirname(dir))
|
|
os.mkdir(dir)
|
|
|
|
|
|
def SUBST_VARS_RECURSIVE(string, env):
|
|
'''recursively expand variables'''
|
|
if string is None:
|
|
return string
|
|
limit=100
|
|
while (string.find('${') != -1 and limit > 0):
|
|
string = subst_vars_error(string, env)
|
|
limit -= 1
|
|
return string
|
|
|
|
|
|
@conf
|
|
def EXPAND_VARIABLES(ctx, varstr, vars=None):
|
|
'''expand variables from a user supplied dictionary
|
|
|
|
This is most useful when you pass vars=locals() to expand
|
|
all your local variables in strings
|
|
'''
|
|
|
|
if isinstance(varstr, list):
|
|
ret = []
|
|
for s in varstr:
|
|
ret.append(EXPAND_VARIABLES(ctx, s, vars=vars))
|
|
return ret
|
|
|
|
if not isinstance(varstr, str):
|
|
return varstr
|
|
|
|
env = ConfigSet.ConfigSet()
|
|
ret = varstr
|
|
# substitute on user supplied dict if available
|
|
if vars is not None:
|
|
for v in vars.keys():
|
|
env[v] = vars[v]
|
|
ret = SUBST_VARS_RECURSIVE(ret, env)
|
|
|
|
# if anything left, subst on the environment as well
|
|
if ret.find('${') != -1:
|
|
ret = SUBST_VARS_RECURSIVE(ret, ctx.env)
|
|
# make sure there is nothing left. Also check for the common
|
|
# typo of $( instead of ${
|
|
if ret.find('${') != -1 or ret.find('$(') != -1:
|
|
Logs.error('Failed to substitute all variables in varstr=%s' % ret)
|
|
sys.exit(1)
|
|
return ret
|
|
Build.BuildContext.EXPAND_VARIABLES = EXPAND_VARIABLES
|
|
|
|
|
|
def RUN_COMMAND(cmd,
|
|
env=None,
|
|
shell=False):
|
|
'''run a external command, return exit code or signal'''
|
|
if env:
|
|
cmd = SUBST_VARS_RECURSIVE(cmd, env)
|
|
|
|
status = os.system(cmd)
|
|
if os.WIFEXITED(status):
|
|
return os.WEXITSTATUS(status)
|
|
if os.WIFSIGNALED(status):
|
|
return - os.WTERMSIG(status)
|
|
Logs.error("Unknown exit reason %d for command: %s" % (status, cmd))
|
|
return -1
|
|
|
|
|
|
def RUN_PYTHON_TESTS(testfiles, pythonpath=None, extra_env=None):
|
|
env = LOAD_ENVIRONMENT()
|
|
if pythonpath is None:
|
|
pythonpath = os.path.join(Context.g_module.out, 'python')
|
|
result = 0
|
|
for interp in env.python_interpreters:
|
|
if not isinstance(interp, str):
|
|
interp = ' '.join(interp)
|
|
for testfile in testfiles:
|
|
cmd = "PYTHONPATH=%s %s %s" % (pythonpath, interp, testfile)
|
|
if extra_env:
|
|
for key, value in extra_env.items():
|
|
cmd = "%s=%s %s" % (key, value, cmd)
|
|
print('Running Python test with %s: %s' % (interp, testfile))
|
|
ret = RUN_COMMAND(cmd)
|
|
if ret:
|
|
print('Python test failed: %s' % cmd)
|
|
result = ret
|
|
return result
|
|
|
|
|
|
# make sure we have md5. some systems don't have it
|
|
try:
|
|
from hashlib import md5
|
|
# Even if hashlib.md5 exists, it may be unusable.
|
|
# Try to use MD5 function. In FIPS mode this will cause an exception
|
|
# and we'll get to the replacement code
|
|
foo = md5(b'abcd')
|
|
except:
|
|
try:
|
|
import md5
|
|
# repeat the same check here, mere success of import is not enough.
|
|
# Try to use MD5 function. In FIPS mode this will cause an exception
|
|
foo = md5.md5(b'abcd')
|
|
except:
|
|
Context.SIG_NIL = hash('abcd')
|
|
class replace_md5(object):
|
|
def __init__(self):
|
|
self.val = None
|
|
def update(self, val):
|
|
self.val = hash((self.val, val))
|
|
def digest(self):
|
|
return str(self.val)
|
|
def hexdigest(self):
|
|
return self.digest().encode('hex')
|
|
def replace_h_file(filename):
|
|
f = open(filename, 'rb')
|
|
m = replace_md5()
|
|
while (filename):
|
|
filename = f.read(100000)
|
|
m.update(filename)
|
|
f.close()
|
|
return m.digest()
|
|
Utils.md5 = replace_md5
|
|
Task.md5 = replace_md5
|
|
Utils.h_file = replace_h_file
|
|
|
|
|
|
def LOAD_ENVIRONMENT():
|
|
'''load the configuration environment, allowing access to env vars
|
|
from new commands'''
|
|
env = ConfigSet.ConfigSet()
|
|
try:
|
|
p = os.path.join(Context.g_module.out, 'c4che/default'+CACHE_SUFFIX)
|
|
env.load(p)
|
|
except (OSError, IOError):
|
|
pass
|
|
return env
|
|
|
|
|
|
def IS_NEWER(bld, file1, file2):
|
|
'''return True if file1 is newer than file2'''
|
|
curdir = bld.path.abspath()
|
|
t1 = os.stat(os.path.join(curdir, file1)).st_mtime
|
|
t2 = os.stat(os.path.join(curdir, file2)).st_mtime
|
|
return t1 > t2
|
|
Build.BuildContext.IS_NEWER = IS_NEWER
|
|
|
|
|
|
@conf
|
|
def RECURSE(ctx, directory):
|
|
'''recurse into a directory, relative to the curdir or top level'''
|
|
try:
|
|
visited_dirs = ctx.visited_dirs
|
|
except AttributeError:
|
|
visited_dirs = ctx.visited_dirs = set()
|
|
d = os.path.join(ctx.path.abspath(), directory)
|
|
if os.path.exists(d):
|
|
abspath = os.path.abspath(d)
|
|
else:
|
|
abspath = os.path.abspath(os.path.join(Context.g_module.top, directory))
|
|
ctxclass = ctx.__class__.__name__
|
|
key = ctxclass + ':' + abspath
|
|
if key in visited_dirs:
|
|
# already done it
|
|
return
|
|
visited_dirs.add(key)
|
|
relpath = os.path.relpath(abspath, ctx.path.abspath())
|
|
if ctxclass in ['OptionsContext',
|
|
'ConfigurationContext',
|
|
'BuildContext',
|
|
'CleanContext',
|
|
'InstallContext',
|
|
'UninstallContext',
|
|
'ListContext']:
|
|
return ctx.recurse(relpath)
|
|
if 'waflib.extras.compat15' in sys.modules:
|
|
return ctx.recurse(relpath)
|
|
raise Errors.WafError('Unknown RECURSE context class: {}'.format(ctxclass))
|
|
Options.OptionsContext.RECURSE = RECURSE
|
|
Build.BuildContext.RECURSE = RECURSE
|
|
|
|
|
|
def CHECK_MAKEFLAGS(options):
|
|
'''check for MAKEFLAGS environment variable in case we are being
|
|
called from a Makefile try to honor a few make command line flags'''
|
|
if not 'WAF_MAKE' in os.environ:
|
|
return
|
|
makeflags = os.environ.get('MAKEFLAGS')
|
|
if makeflags is None:
|
|
makeflags = ""
|
|
jobs_set = False
|
|
jobs = None
|
|
# we need to use shlex.split to cope with the escaping of spaces
|
|
# in makeflags
|
|
for opt in shlex.split(makeflags):
|
|
# options can come either as -x or as x
|
|
if opt[0:2] == 'V=':
|
|
options.verbose = Logs.verbose = int(opt[2:])
|
|
if Logs.verbose > 0:
|
|
Logs.zones = ['runner']
|
|
if Logs.verbose > 2:
|
|
Logs.zones = ['*']
|
|
elif opt[0].isupper() and opt.find('=') != -1:
|
|
# this allows us to set waf options on the make command line
|
|
# for example, if you do "make FOO=blah", then we set the
|
|
# option 'FOO' in Options.options, to blah. If you look in wafsamba/wscript
|
|
# you will see that the command line accessible options have their dest=
|
|
# set to uppercase, to allow for passing of options from make in this way
|
|
# this is also how "make test TESTS=testpattern" works, and
|
|
# "make VERBOSE=1" as well as things like "make SYMBOLCHECK=1"
|
|
loc = opt.find('=')
|
|
setattr(options, opt[0:loc], opt[loc+1:])
|
|
elif opt[0] != '-':
|
|
for v in opt:
|
|
if re.search(r'j[0-9]*$', v):
|
|
jobs_set = True
|
|
jobs = opt.strip('j')
|
|
elif v == 'k':
|
|
options.keep = True
|
|
elif re.search(r'-j[0-9]*$', opt):
|
|
jobs_set = True
|
|
jobs = opt.strip('-j')
|
|
elif opt == '-k':
|
|
options.keep = True
|
|
if not jobs_set:
|
|
# default to one job
|
|
options.jobs = 1
|
|
elif jobs_set and jobs:
|
|
options.jobs = int(jobs)
|
|
|
|
waflib_options_parse_cmd_args = Options.OptionsContext.parse_cmd_args
|
|
def wafsamba_options_parse_cmd_args(self, _args=None, cwd=None, allow_unknown=False):
|
|
(options, commands, envvars) = \
|
|
waflib_options_parse_cmd_args(self,
|
|
_args=_args,
|
|
cwd=cwd,
|
|
allow_unknown=allow_unknown)
|
|
CHECK_MAKEFLAGS(options)
|
|
if options.jobs == 1:
|
|
#
|
|
# waflib.Runner.Parallel processes jobs inline if the possible number
|
|
# of jobs is just 1. But (at least in waf <= 2.0.12) it still calls
|
|
# create a waflib.Runner.Spawner() which creates a single
|
|
# waflib.Runner.Consumer() thread that tries to process jobs from the
|
|
# queue.
|
|
#
|
|
# This has strange effects, which are not noticed typically,
|
|
# but at least on AIX python has broken threading and fails
|
|
# in random ways.
|
|
#
|
|
# So we just add a dummy Spawner class.
|
|
class NoOpSpawner(object):
|
|
def __init__(self, master):
|
|
return
|
|
from waflib import Runner
|
|
Runner.Spawner = NoOpSpawner
|
|
return options, commands, envvars
|
|
Options.OptionsContext.parse_cmd_args = wafsamba_options_parse_cmd_args
|
|
|
|
option_groups = {}
|
|
|
|
def option_group(opt, name):
|
|
'''find or create an option group'''
|
|
global option_groups
|
|
if name in option_groups:
|
|
return option_groups[name]
|
|
gr = opt.add_option_group(name)
|
|
option_groups[name] = gr
|
|
return gr
|
|
Options.OptionsContext.option_group = option_group
|
|
|
|
|
|
def save_file(filename, contents, create_dir=False):
|
|
'''save data to a file'''
|
|
if create_dir:
|
|
mkdir_p(os.path.dirname(filename))
|
|
try:
|
|
f = open(filename, 'w')
|
|
f.write(contents)
|
|
f.close()
|
|
except:
|
|
return False
|
|
return True
|
|
|
|
|
|
def load_file(filename):
|
|
'''return contents of a file'''
|
|
try:
|
|
f = open(filename, 'r')
|
|
r = f.read()
|
|
f.close()
|
|
except:
|
|
return None
|
|
return r
|
|
|
|
|
|
def reconfigure(ctx):
|
|
'''rerun configure if necessary'''
|
|
if not os.path.exists(os.environ.get('WAFLOCK', '.lock-wscript')):
|
|
raise Errors.WafError('configure has not been run')
|
|
import samba_wildcard
|
|
bld = samba_wildcard.fake_build_environment()
|
|
Configure.autoconfig = True
|
|
Scripting.check_configured(bld)
|
|
|
|
|
|
def map_shlib_extension(ctx, name, python=False):
|
|
'''map a filename with a shared library extension of .so to the real shlib name'''
|
|
if name is None:
|
|
return None
|
|
if name[-1:].isdigit():
|
|
# some libraries have specified versions in the wscript rule
|
|
return name
|
|
(root1, ext1) = os.path.splitext(name)
|
|
if python:
|
|
return ctx.env.pyext_PATTERN % root1
|
|
else:
|
|
(root2, ext2) = os.path.splitext(ctx.env.cshlib_PATTERN)
|
|
return root1+ext2
|
|
Build.BuildContext.map_shlib_extension = map_shlib_extension
|
|
|
|
def apply_pattern(filename, pattern):
|
|
'''apply a filename pattern to a filename that may have a directory component'''
|
|
dirname = os.path.dirname(filename)
|
|
if not dirname:
|
|
return pattern % filename
|
|
basename = os.path.basename(filename)
|
|
return os.path.join(dirname, pattern % basename)
|
|
|
|
def make_libname(ctx, name, nolibprefix=False, version=None, python=False):
|
|
"""make a library filename
|
|
Options:
|
|
nolibprefix: don't include the lib prefix
|
|
version : add a version number
|
|
python : if we should use python module name conventions"""
|
|
|
|
if python:
|
|
libname = apply_pattern(name, ctx.env.pyext_PATTERN)
|
|
else:
|
|
libname = apply_pattern(name, ctx.env.cshlib_PATTERN)
|
|
if nolibprefix and libname[0:3] == 'lib':
|
|
libname = libname[3:]
|
|
if version:
|
|
if version[0] == '.':
|
|
version = version[1:]
|
|
(root, ext) = os.path.splitext(libname)
|
|
if ext == ".dylib":
|
|
# special case - version goes before the prefix
|
|
libname = "%s.%s%s" % (root, version, ext)
|
|
else:
|
|
libname = "%s%s.%s" % (root, ext, version)
|
|
return libname
|
|
Build.BuildContext.make_libname = make_libname
|
|
|
|
|
|
def get_tgt_list(bld):
|
|
'''return a list of build objects for samba'''
|
|
|
|
targets = LOCAL_CACHE(bld, 'TARGET_TYPE')
|
|
|
|
# build a list of task generators we are interested in
|
|
tgt_list = []
|
|
for tgt in targets:
|
|
type = targets[tgt]
|
|
if not type in ['SUBSYSTEM', 'BUILTIN', 'MODULE', 'BINARY', 'LIBRARY', 'PLUGIN', 'ASN1', 'PYTHON']:
|
|
continue
|
|
t = bld.get_tgen_by_name(tgt)
|
|
if t is None:
|
|
Logs.error("Target %s of type %s has no task generator" % (tgt, type))
|
|
sys.exit(1)
|
|
tgt_list.append(t)
|
|
return tgt_list
|
|
|
|
from waflib.Context import WSCRIPT_FILE
|
|
def PROCESS_SEPARATE_RULE(self, rule):
|
|
''' cause waf to process additional script based on `rule'.
|
|
You should have file named wscript_<stage>_rule in the current directory
|
|
where stage is either 'configure' or 'build'
|
|
'''
|
|
stage = ''
|
|
if isinstance(self, Configure.ConfigurationContext):
|
|
stage = 'configure'
|
|
elif isinstance(self, Build.BuildContext):
|
|
stage = 'build'
|
|
file_path = os.path.join(self.path.abspath(), WSCRIPT_FILE+'_'+stage+'_'+rule)
|
|
node = self.root.find_node(file_path)
|
|
if node:
|
|
try:
|
|
cache = self.recurse_cache
|
|
except AttributeError:
|
|
cache = self.recurse_cache = {}
|
|
if node not in cache:
|
|
cache[node] = True
|
|
self.pre_recurse(node)
|
|
try:
|
|
function_code = node.read('r', None)
|
|
exec(compile(function_code, node.abspath(), 'exec'), self.exec_dict)
|
|
finally:
|
|
self.post_recurse(node)
|
|
|
|
Build.BuildContext.PROCESS_SEPARATE_RULE = PROCESS_SEPARATE_RULE
|
|
ConfigurationContext.PROCESS_SEPARATE_RULE = PROCESS_SEPARATE_RULE
|
|
|
|
def AD_DC_BUILD_IS_ENABLED(self):
|
|
if self.CONFIG_SET('AD_DC_BUILD_IS_ENABLED'):
|
|
return True
|
|
return False
|
|
|
|
Build.BuildContext.AD_DC_BUILD_IS_ENABLED = AD_DC_BUILD_IS_ENABLED
|
|
|
|
@feature('cprogram', 'cshlib', 'cstaticlib')
|
|
@after('apply_lib_vars')
|
|
@before('apply_obj_vars')
|
|
def samba_before_apply_obj_vars(self):
|
|
"""before apply_obj_vars for uselib, this removes the standard paths"""
|
|
|
|
def is_standard_libpath(env, path):
|
|
normalized_path = os.path.normpath(path)
|
|
for _path in env.STANDARD_LIBPATH:
|
|
if _path == normalized_path:
|
|
return True
|
|
return False
|
|
|
|
v = self.env
|
|
|
|
for i in v['RPATH']:
|
|
if is_standard_libpath(v, i):
|
|
v['RPATH'].remove(i)
|
|
|
|
for i in v['LIBPATH']:
|
|
if is_standard_libpath(v, i):
|
|
v['LIBPATH'].remove(i)
|
|
|
|
# Samba options are mostly on by default (administrators and packagers
|
|
# specify features to remove, not add), which is why default=True
|
|
|
|
def samba_add_onoff_option(opt, option, help=(), dest=None, default=True,
|
|
with_name="with", without_name="without"):
|
|
if default is None:
|
|
default_str = "auto"
|
|
elif default is True:
|
|
default_str = "yes"
|
|
elif default is False:
|
|
default_str = "no"
|
|
else:
|
|
default_str = str(default)
|
|
|
|
if help == ():
|
|
help = ("Build with %s support (default=%s)" % (option, default_str))
|
|
if dest is None:
|
|
dest = "with_%s" % option.replace('-', '_')
|
|
|
|
with_val = "--%s-%s" % (with_name, option)
|
|
without_val = "--%s-%s" % (without_name, option)
|
|
|
|
opt.add_option(with_val, help=help, action="store_true", dest=dest,
|
|
default=default)
|
|
opt.add_option(without_val, help=SUPPRESS_HELP, action="store_false",
|
|
dest=dest)
|
|
Options.OptionsContext.samba_add_onoff_option = samba_add_onoff_option
|