summaryrefslogtreecommitdiffstats
path: root/Lib/plat-mac
diff options
context:
space:
mode:
Diffstat (limited to 'Lib/plat-mac')
-rw-r--r--Lib/plat-mac/buildtools.py3
-rw-r--r--Lib/plat-mac/cfmfile.py183
2 files changed, 3 insertions, 183 deletions
diff --git a/Lib/plat-mac/buildtools.py b/Lib/plat-mac/buildtools.py
index 62c456c..5678a9f 100644
--- a/Lib/plat-mac/buildtools.py
+++ b/Lib/plat-mac/buildtools.py
@@ -13,6 +13,9 @@ import macresource
import EasyDialogs
import shutil
+import warnings
+warnings.warn("the buildtools module is deprecated", DeprecationWarning, 2)
+
BuildError = "BuildError"
diff --git a/Lib/plat-mac/cfmfile.py b/Lib/plat-mac/cfmfile.py
deleted file mode 100644
index df157fd..0000000
--- a/Lib/plat-mac/cfmfile.py
+++ /dev/null
@@ -1,183 +0,0 @@
-"""codefragments.py -- wrapper to modify code fragments."""
-
-# (c) 1998, Just van Rossum, Letterror
-
-__version__ = "0.8b3"
-__author__ = "jvr"
-
-import Carbon.File
-import struct
-from Carbon import Res
-import os
-import sys
-
-DEBUG = 0
-
-error = "cfm.error"
-
-BUFSIZE = 0x80000
-
-def mergecfmfiles(srclist, dst, architecture = 'fat'):
- """Merge all files in srclist into a new file dst.
-
- If architecture is given, only code fragments of that type will be used:
- "pwpc" for PPC, "m68k" for cfm68k. This does not work for "classic"
- 68k code, since it does not use code fragments to begin with.
- If architecture is None, all fragments will be used, enabling FAT binaries.
- """
-
- srclist = list(srclist)
- for i in range(len(srclist)):
- srclist[i] = Carbon.File.pathname(srclist[i])
- dst = Carbon.File.pathname(dst)
-
- dstfile = open(dst, "wb")
- rf = Res.FSpOpenResFile(dst, 3)
- try:
- dstcfrg = CfrgResource()
- for src in srclist:
- srccfrg = CfrgResource(src)
- for frag in srccfrg.fragments:
- if frag.architecture == 'pwpc' and architecture == 'm68k':
- continue
- if frag.architecture == 'm68k' and architecture == 'pwpc':
- continue
- dstcfrg.append(frag)
-
- frag.copydata(dstfile)
-
- cfrgres = Res.Resource(dstcfrg.build())
- Res.UseResFile(rf)
- cfrgres.AddResource('cfrg', 0, "")
- finally:
- dstfile.close()
- rf = Res.CloseResFile(rf)
-
-
-class CfrgResource:
-
- def __init__(self, path = None):
- self.version = 1
- self.fragments = []
- self.path = path
- if path is not None and os.path.exists(path):
- currentresref = Res.CurResFile()
- resref = Res.FSpOpenResFile(path, 1)
- Res.UseResFile(resref)
- try:
- try:
- data = Res.Get1Resource('cfrg', 0).data
- except Res.Error:
- raise Res.Error, "no 'cfrg' resource found", sys.exc_traceback
- finally:
- Res.CloseResFile(resref)
- Res.UseResFile(currentresref)
- self.parse(data)
- if self.version != 1:
- raise error, "unknown 'cfrg' resource format"
-
- def parse(self, data):
- (res1, res2, self.version,
- res3, res4, res5, res6,
- self.memberCount) = struct.unpack("8l", data[:32])
- data = data[32:]
- while data:
- frag = FragmentDescriptor(self.path, data)
- data = data[frag.memberSize:]
- self.fragments.append(frag)
-
- def build(self):
- self.memberCount = len(self.fragments)
- data = struct.pack("8l", 0, 0, self.version, 0, 0, 0, 0, self.memberCount)
- for frag in self.fragments:
- data = data + frag.build()
- return data
-
- def append(self, frag):
- self.fragments.append(frag)
-
-
-class FragmentDescriptor:
-
- def __init__(self, path, data = None):
- self.path = path
- if data is not None:
- self.parse(data)
-
- def parse(self, data):
- self.architecture = data[:4]
- ( self.updatelevel,
- self.currentVersion,
- self.oldDefVersion,
- self.stacksize,
- self.applibdir,
- self.fragtype,
- self.where,
- self.offset,
- self.length,
- self.res1, self.res2,
- self.memberSize,) = struct.unpack("4lhBB4lh", data[4:42])
- pname = data[42:self.memberSize]
- self.name = pname[1:1+ord(pname[0])]
-
- def build(self):
- data = self.architecture
- data = data + struct.pack("4lhBB4l",
- self.updatelevel,
- self.currentVersion,
- self.oldDefVersion,
- self.stacksize,
- self.applibdir,
- self.fragtype,
- self.where,
- self.offset,
- self.length,
- self.res1, self.res2)
- self.memberSize = len(data) + 2 + 1 + len(self.name)
- # pad to 4 byte boundaries
- if self.memberSize % 4:
- self.memberSize = self.memberSize + 4 - (self.memberSize % 4)
- data = data + struct.pack("hb", self.memberSize, len(self.name))
- data = data + self.name
- data = data + '\000' * (self.memberSize - len(data))
- return data
-
- def getfragment(self):
- if self.where != 1:
- raise error, "can't read fragment, unsupported location"
- f = open(self.path, "rb")
- f.seek(self.offset)
- if self.length:
- frag = f.read(self.length)
- else:
- frag = f.read()
- f.close()
- return frag
-
- def copydata(self, outfile):
- if self.where != 1:
- raise error, "can't read fragment, unsupported location"
- infile = open(self.path, "rb")
- if self.length == 0:
- infile.seek(0, 2)
- self.length = infile.tell()
-
- # Position input file and record new offset from output file
- infile.seek(self.offset)
-
- # pad to 16 byte boundaries
- offset = outfile.tell()
- if offset % 16:
- offset = offset + 16 - (offset % 16)
- outfile.seek(offset)
- self.offset = offset
-
- l = self.length
- while l:
- if l > BUFSIZE:
- outfile.write(infile.read(BUFSIZE))
- l = l - BUFSIZE
- else:
- outfile.write(infile.read(l))
- l = 0
- infile.close()