summaryrefslogtreecommitdiff
path: root/engine/SCons/Node/FS.py
diff options
context:
space:
mode:
authorJörg Frings-Fürst <debian@jff-webhosting.net>2015-06-21 07:06:09 +0200
committerJörg Frings-Fürst <debian@jff-webhosting.net>2015-06-21 07:06:09 +0200
commit71c19b5c2d38022afb2a5767a3f8da11d0f06444 (patch)
tree1f2a79dfe0ebe98c6b95d5a0a4f5615d9e4afaab /engine/SCons/Node/FS.py
parent3f6052c28746cf9f29443a610ab60f5b02d3b658 (diff)
parentb0f56d74df2a6904ddceb963833d4ea357251853 (diff)
Merge tag 'upstream/2.3.5'
Upstream version 2.3.5
Diffstat (limited to 'engine/SCons/Node/FS.py')
-rw-r--r--engine/SCons/Node/FS.py86
1 files changed, 57 insertions, 29 deletions
diff --git a/engine/SCons/Node/FS.py b/engine/SCons/Node/FS.py
index 4b95613..f261e9e 100644
--- a/engine/SCons/Node/FS.py
+++ b/engine/SCons/Node/FS.py
@@ -11,7 +11,7 @@ that can be used by scripts or modules looking for the canonical default.
"""
#
-# Copyright (c) 2001 - 2014 The SCons Foundation
+# Copyright (c) 2001 - 2015 The SCons Foundation
#
# Permission is hereby granted, free of charge, to any person obtaining
# a copy of this software and associated documentation files (the
@@ -32,7 +32,7 @@ that can be used by scripts or modules looking for the canonical default.
# 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/Node/FS.py 2014/09/27 12:51:43 garyo"
+__revision__ = "src/engine/SCons/Node/FS.py pchdll:3325:cd517fae59a4 2015/06/18 06:53:27 bdbaddog"
import fnmatch
import os
@@ -1401,7 +1401,7 @@ class FS(LocalFS):
message = fmt % ' '.join(map(str, targets))
return targets, message
- def Glob(self, pathname, ondisk=True, source=True, strings=False, cwd=None):
+ def Glob(self, pathname, ondisk=True, source=True, strings=False, exclude=None, cwd=None):
"""
Globs
@@ -1409,7 +1409,7 @@ class FS(LocalFS):
"""
if cwd is None:
cwd = self.getcwd()
- return cwd.glob(pathname, ondisk, source, strings)
+ return cwd.glob(pathname, ondisk, source, strings, exclude)
class DirNodeInfo(SCons.Node.NodeInfoBase):
# This should get reset by the FS initialization.
@@ -1830,6 +1830,12 @@ class Dir(Base):
return self.tpath + OS_SEP + name
def entry_exists_on_disk(self, name):
+ """ Searches through the file/dir entries of the current
+ directory, and returns True if a physical entry with the given
+ name could be found.
+
+ @see rentry_exists_on_disk
+ """
try:
d = self.on_disk_entries
except AttributeError:
@@ -1854,6 +1860,33 @@ class Dir(Base):
else:
return name in d
+ def rentry_exists_on_disk(self, name):
+ """ Searches through the file/dir entries of the current
+ *and* all its remote directories (repos), and returns
+ True if a physical entry with the given name could be found.
+ The local directory (self) gets searched first, so
+ repositories take a lower precedence regarding the
+ searching order.
+
+ @see entry_exists_on_disk
+ """
+
+ rentry_exists = self.entry_exists_on_disk(name)
+ if not rentry_exists:
+ # Search through the repository folders
+ norm_name = _my_normcase(name)
+ for rdir in self.get_all_rdirs():
+ try:
+ node = rdir.entries[norm_name]
+ if node:
+ rentry_exists = True
+ break
+ except KeyError:
+ if rdir.entry_exists_on_disk(name):
+ rentry_exists = True
+ break
+ return rentry_exists
+
memoizer_counters.append(SCons.Memoize.CountValue('srcdir_list'))
def srcdir_list(self):
@@ -1987,7 +2020,7 @@ class Dir(Base):
for dirname in [n for n in names if isinstance(entries[n], Dir)]:
entries[dirname].walk(func, arg)
- def glob(self, pathname, ondisk=True, source=False, strings=False):
+ def glob(self, pathname, ondisk=True, source=False, strings=False, exclude=None):
"""
Returns a list of Nodes (or strings) matching a specified
pathname pattern.
@@ -2015,24 +2048,31 @@ class Dir(Base):
The "strings" argument, when true, returns the matches as strings,
not Nodes. The strings are path names relative to this directory.
+ The "exclude" argument, if not None, must be a pattern or a list
+ of patterns following the same UNIX shell semantics.
+ Elements matching a least one pattern of this list will be excluded
+ from the result.
+
The underlying algorithm is adapted from the glob.glob() function
in the Python library (but heavily modified), and uses fnmatch()
under the covers.
"""
dirname, basename = os.path.split(pathname)
if not dirname:
- return sorted(self._glob1(basename, ondisk, source, strings),
- key=lambda t: str(t))
- if has_glob_magic(dirname):
- list = self.glob(dirname, ondisk, source, strings=False)
+ result = self._glob1(basename, ondisk, source, strings)
else:
- list = [self.Dir(dirname, create=True)]
- result = []
- for dir in list:
- r = dir._glob1(basename, ondisk, source, strings)
- if strings:
- r = [os.path.join(str(dir), x) for x in r]
- result.extend(r)
+ if has_glob_magic(dirname):
+ list = self.glob(dirname, ondisk, source, False, exclude)
+ else:
+ list = [self.Dir(dirname, create=True)]
+ result = []
+ for dir in list:
+ r = dir._glob1(basename, ondisk, source, strings)
+ if strings:
+ r = [os.path.join(str(dir), x) for x in r]
+ result.extend(r)
+ if exclude:
+ result = filter(lambda x: not any(fnmatch.fnmatch(str(x), e) for e in SCons.Util.flatten(exclude)), result)
return sorted(result, key=lambda a: str(a))
def _glob1(self, pattern, ondisk=True, source=False, strings=False):
@@ -2095,14 +2135,12 @@ class Dir(Base):
names = set(names)
if pattern[0] != '.':
- #names = [ n for n in names if n[0] != '.' ]
names = [x for x in names if x[0] != '.']
names = fnmatch.filter(names, pattern)
if strings:
return names
- #return [ self.entries[_my_normcase(n)] for n in names ]
return [self.entries[_my_normcase(n)] for n in names]
class RootDir(Dir):
@@ -2197,17 +2235,7 @@ class RootDir(Dir):
raise SCons.Errors.UserError(msg)
# There is no Node for this path name, and we're allowed
# to create it.
- # (note: would like to use p.rsplit('/',1) here but
- # that's not in python 2.3)
- # e.g.: dir_name, file_name = p.rsplit('/',1)
- last_slash = p.rindex('/')
- if (last_slash >= 0):
- dir_name = p[:last_slash]
- file_name = p[last_slash+1:]
- else:
- dir_name = p # shouldn't happen, just in case
- file_name = ''
-
+ dir_name, file_name = p.rsplit('/',1)
dir_node = self._lookup_abs(dir_name, Dir)
result = klass(file_name, dir_node, self.fs)