summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorZac Medico <zmedico@gentoo.org>2009-08-04 23:19:45 +0000
committerZac Medico <zmedico@gentoo.org>2009-08-04 23:19:45 +0000
commit53bf20541ef338c2cfe8f818c9ae5e1b75d5146e (patch)
tree8bb5286d85cc2c17f04d9a5ac0d3e7e8e1cdf82d
parentFix bindbapi.aux_get and aux_update to work with py3k/unicode. (diff)
downloadportage-multirepo-53bf20541ef338c2cfe8f818c9ae5e1b75d5146e.tar.gz
portage-multirepo-53bf20541ef338c2cfe8f818c9ae5e1b75d5146e.tar.bz2
portage-multirepo-53bf20541ef338c2cfe8f818c9ae5e1b75d5146e.zip
Open all files in binary mode for py3k compatibility.
svn path=/main/trunk/; revision=13912
-rw-r--r--pym/portage/xpak.py30
1 files changed, 15 insertions, 15 deletions
diff --git a/pym/portage/xpak.py b/pym/portage/xpak.py
index e24da301..a1516e03 100644
--- a/pym/portage/xpak.py
+++ b/pym/portage/xpak.py
@@ -68,14 +68,14 @@ def xpak(rootdir,outfile=None):
mylist.sort()
mydata = {}
for x in mylist:
- a = open(x, "r")
+ a = open(x, 'rb')
mydata[x] = a.read()
a.close()
os.chdir(origdir)
xpak_segment = xpak_mem(mydata)
if outfile:
- outf = open(outfile, "w")
+ outf = open(outfile, 'wb')
outf.write(xpak_segment)
outf.close()
else:
@@ -104,7 +104,7 @@ def xsplit(infile):
"""(infile) -- Splits the infile into two files.
'infile.index' contains the index segment.
'infile.dat' contails the data segment."""
- myfile=open(infile,"r")
+ myfile = open(infile, 'rb')
mydat=myfile.read()
myfile.close()
@@ -112,10 +112,10 @@ def xsplit(infile):
if not splits:
return False
- myfile=open(infile+".index","w")
+ myfile = open(infile + '.index', 'wb')
myfile.write(splits[0])
myfile.close()
- myfile=open(infile+".dat","w")
+ myfile = open(infile + '.dat', 'wb')
myfile.write(splits[1])
myfile.close()
return True
@@ -130,7 +130,7 @@ def xsplit_mem(mydat):
def getindex(infile):
"""(infile) -- grabs the index segment from the infile and returns it."""
- myfile=open(infile,"r")
+ myfile = open(infile, 'rb')
myheader=myfile.read(16)
if myheader[0:8]!="XPAKPACK":
myfile.close()
@@ -143,7 +143,7 @@ def getindex(infile):
def getboth(infile):
"""(infile) -- grabs the index and data segments from the infile.
Returns an array [indexSegment,dataSegment]"""
- myfile=open(infile,"r")
+ myfile = open(infile, 'rb')
myheader=myfile.read(16)
if myheader[0:8]!="XPAKPACK":
myfile.close()
@@ -217,7 +217,7 @@ def xpand(myid,mydest):
if dirname:
if not os.path.exists(dirname):
os.makedirs(dirname)
- mydat=open(myname,"w")
+ mydat = open(myname, 'wb')
mydat.write(mydata[datapos:datapos+datalen])
mydat.close()
startpos=startpos+namelen+12
@@ -262,7 +262,7 @@ class tbz2(object):
def recompose_mem(self, xpdata):
self.scan() # Don't care about condition... We'll rewrite the data anyway.
- myfile=open(self.file,"a+")
+ myfile = open(self.file, 'ab+')
if not myfile:
raise IOError
myfile.seek(-self.xpaksize,2) # 0,2 or -0,2 just mean EOF.
@@ -298,7 +298,7 @@ class tbz2(object):
if not changed:
return 1
self.filestat=mystat
- a=open(self.file,"r")
+ a = open(self.file, 'rb')
a.seek(-16,2)
trailer=a.read()
self.infosize=0
@@ -341,7 +341,7 @@ class tbz2(object):
myresult=searchindex(self.index,myfile)
if not myresult:
return mydefault
- a=open(self.file,"r")
+ a = open(self.file, 'rb')
a.seek(self.datapos+myresult[0],0)
myreturn=a.read(myresult[1])
a.close()
@@ -365,7 +365,7 @@ class tbz2(object):
except:
os.chdir("/")
origdir="/"
- a=open(self.file,"r")
+ a = open(self.file, 'rb')
if not os.path.exists(mydest):
os.makedirs(mydest)
os.chdir(mydest)
@@ -379,7 +379,7 @@ class tbz2(object):
if dirname:
if not os.path.exists(dirname):
os.makedirs(dirname)
- mydat=open(myname,"w")
+ mydat = open(myname, 'wb')
a.seek(self.datapos+datapos)
mydat.write(a.read(datalen))
mydat.close()
@@ -392,7 +392,7 @@ class tbz2(object):
"""Returns all the files from the dataSegment as a map object."""
if not self.scan():
return 0
- a = open(self.file, "r")
+ a = open(self.file, 'rb')
mydata = {}
startpos=0
while ((startpos+8)<self.indexsize):
@@ -411,7 +411,7 @@ class tbz2(object):
if not self.scan():
return None
- a = open(self.file,"r")
+ a = open(self.file, 'rb')
a.seek(self.datapos)
mydata =a.read(self.datasize)
a.close()