Subst.py
上传用户:market2
上传日期:2018-11-18
资源大小:18786k
文件大小:33k
- """SCons.Subst
- SCons string substitution.
- """
- #
- # Copyright (c) 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 The SCons Foundation
- #
- # Permission is hereby granted, free of charge, to any person obtaining
- # a copy of this software and associated documentation files (the
- # "Software"), to deal in the Software without restriction, including
- # without limitation the rights to use, copy, modify, merge, publish,
- # distribute, sublicense, and/or sell copies of the Software, and to
- # permit persons to whom the Software is furnished to do so, subject to
- # the following conditions:
- #
- # The above copyright notice and this permission notice shall be included
- # in all copies or substantial portions of the Software.
- #
- # THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY
- # KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE
- # WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
- # NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
- # LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
- # OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
- # WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- #
- __revision__ = "src/engine/SCons/Subst.py 3057 2008/06/09 22:21:00 knight"
- import SCons.compat
- import re
- import string
- import types
- import UserList
- import UserString
- import SCons.Errors
- from SCons.Util import is_String, is_Sequence
- # Indexed by the SUBST_* constants below.
- _strconv = [SCons.Util.to_String_for_subst,
- SCons.Util.to_String_for_subst,
- SCons.Util.to_String_for_signature]
- AllowableExceptions = (IndexError, NameError)
- def SetAllowableExceptions(*excepts):
- global AllowableExceptions
- AllowableExceptions = filter(None, excepts)
- def raise_exception(exception, target, s):
- name = exception.__class__.__name__
- msg = "%s `%s' trying to evaluate `%s'" % (name, exception, s)
- if target:
- raise SCons.Errors.BuildError, (target[0], msg)
- else:
- raise SCons.Errors.UserError, msg
- class Literal:
- """A wrapper for a string. If you use this object wrapped
- around a string, then it will be interpreted as literal.
- When passed to the command interpreter, all special
- characters will be escaped."""
- def __init__(self, lstr):
- self.lstr = lstr
- def __str__(self):
- return self.lstr
- def escape(self, escape_func):
- return escape_func(self.lstr)
- def for_signature(self):
- return self.lstr
- def is_literal(self):
- return 1
- class SpecialAttrWrapper:
- """This is a wrapper for what we call a 'Node special attribute.'
- This is any of the attributes of a Node that we can reference from
- Environment variable substitution, such as $TARGET.abspath or
- $SOURCES[1].filebase. We implement the same methods as Literal
- so we can handle special characters, plus a for_signature method,
- such that we can return some canonical string during signature
- calculation to avoid unnecessary rebuilds."""
- def __init__(self, lstr, for_signature=None):
- """The for_signature parameter, if supplied, will be the
- canonical string we return from for_signature(). Else
- we will simply return lstr."""
- self.lstr = lstr
- if for_signature:
- self.forsig = for_signature
- else:
- self.forsig = lstr
- def __str__(self):
- return self.lstr
- def escape(self, escape_func):
- return escape_func(self.lstr)
- def for_signature(self):
- return self.forsig
- def is_literal(self):
- return 1
- def quote_spaces(arg):
- """Generic function for putting double quotes around any string that
- has white space in it."""
- if ' ' in arg or 't' in arg:
- return '"%s"' % arg
- else:
- return str(arg)
- class CmdStringHolder(UserString.UserString):
- """This is a special class used to hold strings generated by
- scons_subst() and scons_subst_list(). It defines a special method
- escape(). When passed a function with an escape algorithm for a
- particular platform, it will return the contained string with the
- proper escape sequences inserted.
- """
- def __init__(self, cmd, literal=None):
- UserString.UserString.__init__(self, cmd)
- self.literal = literal
- def is_literal(self):
- return self.literal
- def escape(self, escape_func, quote_func=quote_spaces):
- """Escape the string with the supplied function. The
- function is expected to take an arbitrary string, then
- return it with all special characters escaped and ready
- for passing to the command interpreter.
- After calling this function, the next call to str() will
- return the escaped string.
- """
- if self.is_literal():
- return escape_func(self.data)
- elif ' ' in self.data or 't' in self.data:
- return quote_func(self.data)
- else:
- return self.data
- def escape_list(list, escape_func):
- """Escape a list of arguments by running the specified escape_func
- on every object in the list that has an escape() method."""
- def escape(obj, escape_func=escape_func):
- try:
- e = obj.escape
- except AttributeError:
- return obj
- else:
- return e(escape_func)
- return map(escape, list)
- class NLWrapper:
- """A wrapper class that delays turning a list of sources or targets
- into a NodeList until it's needed. The specified function supplied
- when the object is initialized is responsible for turning raw nodes
- into proxies that implement the special attributes like .abspath,
- .source, etc. This way, we avoid creating those proxies just
- "in case" someone is going to use $TARGET or the like, and only
- go through the trouble if we really have to.
- In practice, this might be a wash performance-wise, but it's a little
- cleaner conceptually...
- """
-
- def __init__(self, list, func):
- self.list = list
- self.func = func
- def _return_nodelist(self):
- return self.nodelist
- def _gen_nodelist(self):
- list = self.list
- if list is None:
- list = []
- elif not is_Sequence(list):
- list = [list]
- # The map(self.func) call is what actually turns
- # a list into appropriate proxies.
- self.nodelist = SCons.Util.NodeList(map(self.func, list))
- self._create_nodelist = self._return_nodelist
- return self.nodelist
- _create_nodelist = _gen_nodelist
-
- class Targets_or_Sources(UserList.UserList):
- """A class that implements $TARGETS or $SOURCES expansions by in turn
- wrapping a NLWrapper. This class handles the different methods used
- to access the list, calling the NLWrapper to create proxies on demand.
- Note that we subclass UserList.UserList purely so that the
- is_Sequence() function will identify an object of this class as
- a list during variable expansion. We're not really using any
- UserList.UserList methods in practice.
- """
- def __init__(self, nl):
- self.nl = nl
- def __getattr__(self, attr):
- nl = self.nl._create_nodelist()
- return getattr(nl, attr)
- def __getitem__(self, i):
- nl = self.nl._create_nodelist()
- return nl[i]
- def __getslice__(self, i, j):
- nl = self.nl._create_nodelist()
- i = max(i, 0); j = max(j, 0)
- return nl[i:j]
- def __str__(self):
- nl = self.nl._create_nodelist()
- return str(nl)
- def __repr__(self):
- nl = self.nl._create_nodelist()
- return repr(nl)
- class Target_or_Source:
- """A class that implements $TARGET or $SOURCE expansions by in turn
- wrapping a NLWrapper. This class handles the different methods used
- to access an individual proxy Node, calling the NLWrapper to create
- a proxy on demand.
- """
- def __init__(self, nl):
- self.nl = nl
- def __getattr__(self, attr):
- nl = self.nl._create_nodelist()
- try:
- nl0 = nl[0]
- except IndexError:
- # If there is nothing in the list, then we have no attributes to
- # pass through, so raise AttributeError for everything.
- raise AttributeError, "NodeList has no attribute: %s" % attr
- return getattr(nl0, attr)
- def __str__(self):
- nl = self.nl._create_nodelist()
- if nl:
- return str(nl[0])
- return ''
- def __repr__(self):
- nl = self.nl._create_nodelist()
- if nl:
- return repr(nl[0])
- return ''
- def subst_dict(target, source):
- """Create a dictionary for substitution of special
- construction variables.
- This translates the following special arguments:
- target - the target (object or array of objects),
- used to generate the TARGET and TARGETS
- construction variables
- source - the source (object or array of objects),
- used to generate the SOURCES and SOURCE
- construction variables
- """
- dict = {}
- if target:
- tnl = NLWrapper(target, lambda x: x.get_subst_proxy())
- dict['TARGETS'] = Targets_or_Sources(tnl)
- dict['TARGET'] = Target_or_Source(tnl)
- else:
- dict['TARGETS'] = None
- dict['TARGET'] = None
- if source:
- def get_src_subst_proxy(node):
- try:
- rfile = node.rfile
- except AttributeError:
- pass
- else:
- node = rfile()
- return node.get_subst_proxy()
- snl = NLWrapper(source, get_src_subst_proxy)
- dict['SOURCES'] = Targets_or_Sources(snl)
- dict['SOURCE'] = Target_or_Source(snl)
- else:
- dict['SOURCES'] = None
- dict['SOURCE'] = None
- return dict
- # Constants for the "mode" parameter to scons_subst_list() and
- # scons_subst(). SUBST_RAW gives the raw command line. SUBST_CMD
- # gives a command line suitable for passing to a shell. SUBST_SIG
- # gives a command line appropriate for calculating the signature
- # of a command line...if this changes, we should rebuild.
- SUBST_CMD = 0
- SUBST_RAW = 1
- SUBST_SIG = 2
- _rm = re.compile(r'$[()]')
- _remove = re.compile(r'$([^$]*($[^)][^$]*)*$)')
- # Indexed by the SUBST_* constants above.
- _regex_remove = [ _rm, None, _remove ]
- def _rm_list(list):
- #return [ l for l in list if not l in ('$(', '$)') ]
- return filter(lambda l: not l in ('$(', '$)'), list)
- def _remove_list(list):
- result = []
- do_append = result.append
- for l in list:
- if l == '$(':
- do_append = lambda x: None
- elif l == '$)':
- do_append = result.append
- else:
- do_append(l)
- return result
- # Indexed by the SUBST_* constants above.
- _list_remove = [ _rm_list, None, _remove_list ]
- # Regular expressions for splitting strings and handling substitutions,
- # for use by the scons_subst() and scons_subst_list() functions:
- #
- # The first expression compiled matches all of the $-introduced tokens
- # that we need to process in some way, and is used for substitutions.
- # The expressions it matches are:
- #
- # "$$"
- # "$("
- # "$)"
- # "$variable" [must begin with alphabetic or underscore]
- # "${any stuff}"
- #
- # The second expression compiled is used for splitting strings into tokens
- # to be processed, and it matches all of the tokens listed above, plus
- # the following that affect how arguments do or don't get joined together:
- #
- # " " [white space]
- # "non-white-space" [without any dollar signs]
- # "$" [single dollar sign]
- #
- _dollar_exps_str = r'$[$()]|$[_a-zA-Z][.w]*|${[^}]*}'
- _dollar_exps = re.compile(r'(%s)' % _dollar_exps_str)
- _separate_args = re.compile(r'(%s|s+|[^s$]+|$)' % _dollar_exps_str)
- # This regular expression is used to replace strings of multiple white
- # space characters in the string result from the scons_subst() function.
- _space_sep = re.compile(r'[t ]+(?![^{]*})')
- def scons_subst(strSubst, env, mode=SUBST_RAW, target=None, source=None, gvars={}, lvars={}, conv=None):
- """Expand a string or list containing construction variable
- substitutions.
- This is the work-horse function for substitutions in file names
- and the like. The companion scons_subst_list() function (below)
- handles separating command lines into lists of arguments, so see
- that function if that's what you're looking for.
- """
- if type(strSubst) == types.StringType and string.find(strSubst, '$') < 0:
- return strSubst
- class StringSubber:
- """A class to construct the results of a scons_subst() call.
- This binds a specific construction environment, mode, target and
- source with two methods (substitute() and expand()) that handle
- the expansion.
- """
- def __init__(self, env, mode, target, source, conv, gvars):
- self.env = env
- self.mode = mode
- self.target = target
- self.source = source
- self.conv = conv
- self.gvars = gvars
- def expand(self, s, lvars):
- """Expand a single "token" as necessary, returning an
- appropriate string containing the expansion.
- This handles expanding different types of things (strings,
- lists, callables) appropriately. It calls the wrapper
- substitute() method to re-expand things as necessary, so that
- the results of expansions of side-by-side strings still get
- re-evaluated separately, not smushed together.
- """
- if is_String(s):
- try:
- s0, s1 = s[:2]
- except (IndexError, ValueError):
- return s
- if s0 != '$':
- return s
- if s1 == '$':
- return '$'
- elif s1 in '()':
- return s
- else:
- key = s[1:]
- if key[0] == '{' or string.find(key, '.') >= 0:
- if key[0] == '{':
- key = key[1:-1]
- try:
- s = eval(key, self.gvars, lvars)
- except KeyboardInterrupt:
- raise
- except Exception, e:
- if e.__class__ in AllowableExceptions:
- return ''
- raise_exception(e, self.target, s)
- else:
- if lvars.has_key(key):
- s = lvars[key]
- elif self.gvars.has_key(key):
- s = self.gvars[key]
- elif not NameError in AllowableExceptions:
- raise_exception(NameError(key), self.target, s)
- else:
- return ''
-
- # Before re-expanding the result, handle
- # recursive expansion by copying the local
- # variable dictionary and overwriting a null
- # string for the value of the variable name
- # we just expanded.
- #
- # This could potentially be optimized by only
- # copying lvars when s contains more expansions,
- # but lvars is usually supposed to be pretty
- # small, and deeply nested variable expansions
- # are probably more the exception than the norm,
- # so it should be tolerable for now.
- lv = lvars.copy()
- var = string.split(key, '.')[0]
- lv[var] = ''
- return self.substitute(s, lv)
- elif is_Sequence(s):
- def func(l, conv=self.conv, substitute=self.substitute, lvars=lvars):
- return conv(substitute(l, lvars))
- return map(func, s)
- elif callable(s):
- try:
- s = s(target=self.target,
- source=self.source,
- env=self.env,
- for_signature=(self.mode != SUBST_CMD))
- except TypeError:
- # This probably indicates that it's a callable
- # object that doesn't match our calling arguments
- # (like an Action).
- if self.mode == SUBST_RAW:
- return s
- s = self.conv(s)
- return self.substitute(s, lvars)
- elif s is None:
- return ''
- else:
- return s
- def substitute(self, args, lvars):
- """Substitute expansions in an argument or list of arguments.
- This serves as a wrapper for splitting up a string into
- separate tokens.
- """
- if is_String(args) and not isinstance(args, CmdStringHolder):
- args = str(args) # In case it's a UserString.
- try:
- def sub_match(match, conv=self.conv, expand=self.expand, lvars=lvars):
- return conv(expand(match.group(1), lvars))
- result = _dollar_exps.sub(sub_match, args)
- except TypeError:
- # If the internal conversion routine doesn't return
- # strings (it could be overridden to return Nodes, for
- # example), then the 1.5.2 re module will throw this
- # exception. Back off to a slower, general-purpose
- # algorithm that works for all data types.
- args = _separate_args.findall(args)
- result = []
- for a in args:
- result.append(self.conv(self.expand(a, lvars)))
- if len(result) == 1:
- result = result[0]
- else:
- result = string.join(map(str, result), '')
- return result
- else:
- return self.expand(args, lvars)
- if conv is None:
- conv = _strconv[mode]
- # Doing this every time is a bit of a waste, since the Executor
- # has typically already populated the OverrideEnvironment with
- # $TARGET/$SOURCE variables. We're keeping this (for now), though,
- # because it supports existing behavior that allows us to call
- # an Action directly with an arbitrary target+source pair, which
- # we use in Tool/tex.py to handle calling $BIBTEX when necessary.
- # If we dropped that behavior (or found another way to cover it),
- # we could get rid of this call completely and just rely on the
- # Executor setting the variables.
- d = subst_dict(target, source)
- if d:
- lvars = lvars.copy()
- lvars.update(d)
- # We're (most likely) going to eval() things. If Python doesn't
- # find a __builtins__ value in the global dictionary used for eval(),
- # it copies the current global values for you. Avoid this by
- # setting it explicitly and then deleting, so we don't pollute the
- # construction environment Dictionary(ies) that are typically used
- # for expansion.
- gvars['__builtins__'] = __builtins__
- ss = StringSubber(env, mode, target, source, conv, gvars)
- result = ss.substitute(strSubst, lvars)
- try:
- del gvars['__builtins__']
- except KeyError:
- pass
- if is_String(result):
- # Remove $(-$) pairs and any stuff in between,
- # if that's appropriate.
- remove = _regex_remove[mode]
- if remove:
- result = remove.sub('', result)
- if mode != SUBST_RAW:
- # Compress strings of white space characters into
- # a single space.
- result = string.strip(_space_sep.sub(' ', result))
- elif is_Sequence(result):
- remove = _list_remove[mode]
- if remove:
- result = remove(result)
- return result
- #Subst_List_Strings = {}
- def scons_subst_list(strSubst, env, mode=SUBST_RAW, target=None, source=None, gvars={}, lvars={}, conv=None):
- """Substitute construction variables in a string (or list or other
- object) and separate the arguments into a command list.
- The companion scons_subst() function (above) handles basic
- substitutions within strings, so see that function instead
- if that's what you're looking for.
- """
- # try:
- # Subst_List_Strings[strSubst] = Subst_List_Strings[strSubst] + 1
- # except KeyError:
- # Subst_List_Strings[strSubst] = 1
- # import SCons.Debug
- # SCons.Debug.caller_trace(1)
- class ListSubber(UserList.UserList):
- """A class to construct the results of a scons_subst_list() call.
- Like StringSubber, this class binds a specific construction
- environment, mode, target and source with two methods
- (substitute() and expand()) that handle the expansion.
- In addition, however, this class is used to track the state of
- the result(s) we're gathering so we can do the appropriate thing
- whenever we have to append another word to the result--start a new
- line, start a new word, append to the current word, etc. We do
- this by setting the "append" attribute to the right method so
- that our wrapper methods only need ever call ListSubber.append(),
- and the rest of the object takes care of doing the right thing
- internally.
- """
- def __init__(self, env, mode, target, source, conv, gvars):
- UserList.UserList.__init__(self, [])
- self.env = env
- self.mode = mode
- self.target = target
- self.source = source
- self.conv = conv
- self.gvars = gvars
- if self.mode == SUBST_RAW:
- self.add_strip = lambda x, s=self: s.append(x)
- else:
- self.add_strip = lambda x, s=self: None
- self.in_strip = None
- self.next_line()
- def expand(self, s, lvars, within_list):
- """Expand a single "token" as necessary, appending the
- expansion to the current result.
- This handles expanding different types of things (strings,
- lists, callables) appropriately. It calls the wrapper
- substitute() method to re-expand things as necessary, so that
- the results of expansions of side-by-side strings still get
- re-evaluated separately, not smushed together.
- """
- if is_String(s):
- try:
- s0, s1 = s[:2]
- except (IndexError, ValueError):
- self.append(s)
- return
- if s0 != '$':
- self.append(s)
- return
- if s1 == '$':
- self.append('$')
- elif s1 == '(':
- self.open_strip('$(')
- elif s1 == ')':
- self.close_strip('$)')
- else:
- key = s[1:]
- if key[0] == '{' or string.find(key, '.') >= 0:
- if key[0] == '{':
- key = key[1:-1]
- try:
- s = eval(key, self.gvars, lvars)
- except KeyboardInterrupt:
- raise
- except Exception, e:
- if e.__class__ in AllowableExceptions:
- return
- raise_exception(e, self.target, s)
- else:
- if lvars.has_key(key):
- s = lvars[key]
- elif self.gvars.has_key(key):
- s = self.gvars[key]
- elif not NameError in AllowableExceptions:
- raise_exception(NameError(), self.target, s)
- else:
- return
- # Before re-expanding the result, handle
- # recursive expansion by copying the local
- # variable dictionary and overwriting a null
- # string for the value of the variable name
- # we just expanded.
- lv = lvars.copy()
- var = string.split(key, '.')[0]
- lv[var] = ''
- self.substitute(s, lv, 0)
- self.this_word()
- elif is_Sequence(s):
- for a in s:
- self.substitute(a, lvars, 1)
- self.next_word()
- elif callable(s):
- try:
- s = s(target=self.target,
- source=self.source,
- env=self.env,
- for_signature=(self.mode != SUBST_CMD))
- except TypeError:
- # This probably indicates that it's a callable
- # object that doesn't match our calling arguments
- # (like an Action).
- if self.mode == SUBST_RAW:
- self.append(s)
- return
- s = self.conv(s)
- self.substitute(s, lvars, within_list)
- elif s is None:
- self.this_word()
- else:
- self.append(s)
- def substitute(self, args, lvars, within_list):
- """Substitute expansions in an argument or list of arguments.
- This serves as a wrapper for splitting up a string into
- separate tokens.
- """
- if is_String(args) and not isinstance(args, CmdStringHolder):
- args = str(args) # In case it's a UserString.
- args = _separate_args.findall(args)
- for a in args:
- if a[0] in ' tnrfv':
- if 'n' in a:
- self.next_line()
- elif within_list:
- self.append(a)
- else:
- self.next_word()
- else:
- self.expand(a, lvars, within_list)
- else:
- self.expand(args, lvars, within_list)
- def next_line(self):
- """Arrange for the next word to start a new line. This
- is like starting a new word, except that we have to append
- another line to the result."""
- UserList.UserList.append(self, [])
- self.next_word()
- def this_word(self):
- """Arrange for the next word to append to the end of the
- current last word in the result."""
- self.append = self.add_to_current_word
- def next_word(self):
- """Arrange for the next word to start a new word."""
- self.append = self.add_new_word
- def add_to_current_word(self, x):
- """Append the string x to the end of the current last word
- in the result. If that is not possible, then just add
- it as a new word. Make sure the entire concatenated string
- inherits the object attributes of x (in particular, the
- escape function) by wrapping it as CmdStringHolder."""
- if not self.in_strip or self.mode != SUBST_SIG:
- try:
- current_word = self[-1][-1]
- except IndexError:
- self.add_new_word(x)
- else:
- # All right, this is a hack and it should probably
- # be refactored out of existence in the future.
- # The issue is that we want to smoosh words together
- # and make one file name that gets escaped if
- # we're expanding something like foo$EXTENSION,
- # but we don't want to smoosh them together if
- # it's something like >$TARGET, because then we'll
- # treat the '>' like it's part of the file name.
- # So for now, just hard-code looking for the special
- # command-line redirection characters...
- try:
- last_char = str(current_word)[-1]
- except IndexError:
- last_char = ' '
- if last_char in '<>|':
- self.add_new_word(x)
- else:
- y = current_word + x
- # We used to treat a word appended to a literal
- # as a literal itself, but this caused problems
- # with interpreting quotes around space-separated
- # targets on command lines. Removing this makes
- # none of the "substantive" end-to-end tests fail,
- # so we'll take this out but leave it commented
- # for now in case there's a problem not covered
- # by the test cases and we need to resurrect this.
- #literal1 = self.literal(self[-1][-1])
- #literal2 = self.literal(x)
- y = self.conv(y)
- if is_String(y):
- #y = CmdStringHolder(y, literal1 or literal2)
- y = CmdStringHolder(y, None)
- self[-1][-1] = y
- def add_new_word(self, x):
- if not self.in_strip or self.mode != SUBST_SIG:
- literal = self.literal(x)
- x = self.conv(x)
- if is_String(x):
- x = CmdStringHolder(x, literal)
- self[-1].append(x)
- self.append = self.add_to_current_word
- def literal(self, x):
- try:
- l = x.is_literal
- except AttributeError:
- return None
- else:
- return l()
- def open_strip(self, x):
- """Handle the "open strip" $( token."""
- self.add_strip(x)
- self.in_strip = 1
- def close_strip(self, x):
- """Handle the "close strip" $) token."""
- self.add_strip(x)
- self.in_strip = None
- if conv is None:
- conv = _strconv[mode]
- # Doing this every time is a bit of a waste, since the Executor
- # has typically already populated the OverrideEnvironment with
- # $TARGET/$SOURCE variables. We're keeping this (for now), though,
- # because it supports existing behavior that allows us to call
- # an Action directly with an arbitrary target+source pair, which
- # we use in Tool/tex.py to handle calling $BIBTEX when necessary.
- # If we dropped that behavior (or found another way to cover it),
- # we could get rid of this call completely and just rely on the
- # Executor setting the variables.
- d = subst_dict(target, source)
- if d:
- lvars = lvars.copy()
- lvars.update(d)
- # We're (most likely) going to eval() things. If Python doesn't
- # find a __builtins__ value in the global dictionary used for eval(),
- # it copies the current global values for you. Avoid this by
- # setting it explicitly and then deleting, so we don't pollute the
- # construction environment Dictionary(ies) that are typically used
- # for expansion.
- gvars['__builtins__'] = __builtins__
- ls = ListSubber(env, mode, target, source, conv, gvars)
- ls.substitute(strSubst, lvars, 0)
- try:
- del gvars['__builtins__']
- except KeyError:
- pass
- return ls.data
- def scons_subst_once(strSubst, env, key):
- """Perform single (non-recursive) substitution of a single
- construction variable keyword.
- This is used when setting a variable when copying or overriding values
- in an Environment. We want to capture (expand) the old value before
- we override it, so people can do things like:
- env2 = env.Clone(CCFLAGS = '$CCFLAGS -g')
- We do this with some straightforward, brute-force code here...
- """
- if type(strSubst) == types.StringType and string.find(strSubst, '$') < 0:
- return strSubst
- matchlist = ['$' + key, '${' + key + '}']
- val = env.get(key, '')
- def sub_match(match, val=val, matchlist=matchlist):
- a = match.group(1)
- if a in matchlist:
- a = val
- if is_Sequence(a):
- return string.join(map(str, a))
- else:
- return str(a)
- if is_Sequence(strSubst):
- result = []
- for arg in strSubst:
- if is_String(arg):
- if arg in matchlist:
- arg = val
- if is_Sequence(arg):
- result.extend(arg)
- else:
- result.append(arg)
- else:
- result.append(_dollar_exps.sub(sub_match, arg))
- else:
- result.append(arg)
- return result
- elif is_String(strSubst):
- return _dollar_exps.sub(sub_match, strSubst)
- else:
- return strSubst