summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorPaul Varner <fuzzyray@gentoo.org>2009-12-08 21:53:45 +0000
committerZac Medico <zmedico@gentoo.org>2013-01-18 20:56:46 -0800
commit8b61e83f8f88f035eaea3d0305d565b507cbdc3d (patch)
treedf31fc0774d738403854240c773ef5dd686b7f95
parent6dc9c3b3c9ff814498663af38227476f59f99033 (diff)
downloadportage-8b61e83f8f88f035eaea3d0305d565b507cbdc3d.tar.gz
portage-8b61e83f8f88f035eaea3d0305d565b507cbdc3d.tar.bz2
portage-8b61e83f8f88f035eaea3d0305d565b507cbdc3d.zip
Merge rev 113 from djanderson's genscripts repo
svn path=/trunk/gentoolkit/; revision=703 http://git.overlays.gentoo.org/gitweb/?p=proj/gentoolkit.git;a=commit;h=acdf616efa73b77936963eaa8b5c715db97646d2
-rw-r--r--pym/portage/glsa.py142
1 files changed, 68 insertions, 74 deletions
diff --git a/pym/portage/glsa.py b/pym/portage/glsa.py
index af6e71437..155d3e543 100644
--- a/pym/portage/glsa.py
+++ b/pym/portage/glsa.py
@@ -28,7 +28,7 @@ from portage.dep import _slot_separator
# Note: the space for rgt and rlt is important !!
# FIXME: use slot deps instead, requires GLSA format versioning
-opMapping = {"le": "<=", "lt": "<", "eq": "=", "gt": ">", "ge": ">=",
+opMapping = {"le": "<=", "lt": "<", "eq": "=", "gt": ">", "ge": ">=",
"rge": ">=~", "rle": "<=~", "rgt": " >~", "rlt": " <~"}
NEWLINE_ESCAPE = "!;\\n" # some random string to mark newlines that should be preserved
SPACE_ESCAPE = "!;_" # some random string to mark spaces that should be preserved
@@ -49,15 +49,15 @@ def get_applied_glsas(settings):
def wrap(text, width, caption=""):
"""
Wraps the given text at column I{width}, optionally indenting
- it so that no text is under I{caption}. It's possible to encode
+ it so that no text is under I{caption}. It's possible to encode
hard linebreaks in I{text} with L{NEWLINE_ESCAPE}.
-
+
@type text: String
@param text: the text to be wrapped
@type width: Integer
@param width: the column at which the text should be wrapped
@type caption: String
- @param caption: this string is inserted at the beginning of the
+ @param caption: this string is inserted at the beginning of the
return value and the paragraph is indented up to
C{len(caption)}.
@rtype: String
@@ -68,7 +68,7 @@ def wrap(text, width, caption=""):
text = text.replace(2*NEWLINE_ESCAPE, NEWLINE_ESCAPE+" "+NEWLINE_ESCAPE)
words = text.split()
indentLevel = len(caption)+1
-
+
for w in words:
if line != "" and line[-1] == "\n":
rValue += line
@@ -97,10 +97,10 @@ def get_glsa_list(myconfig):
Returns a list of all available GLSAs in the given repository
by comparing the filelist there with the pattern described in
the config.
-
+
@type myconfig: portage.config
@param myconfig: Portage settings instance
-
+
@rtype: List of Strings
@return: a list of GLSA IDs in this repository
"""
@@ -116,7 +116,7 @@ def get_glsa_list(myconfig):
dirlist = os.listdir(repository)
prefix = "glsa-"
suffix = ".xml"
-
+
for f in dirlist:
try:
if f[:len(prefix)] == prefix and f[-1*len(suffix):] == suffix:
@@ -128,7 +128,7 @@ def get_glsa_list(myconfig):
def getListElements(listnode):
"""
Get all <li> elements for a given <ol> or <ul> node.
-
+
@type listnode: xml.dom.Node
@param listnode: <ul> or <ol> list to get the elements for
@rtype: List of Strings
@@ -149,7 +149,7 @@ def getText(node, format, textfd = None):
parameter the text might be formatted by adding/removing newlines,
tabs and spaces. This function is only useful for the GLSA DTD,
it's not applicable for other DTDs.
-
+
@type node: xml.dom.Node
@param node: the root node to start with the parsing
@type format: String
@@ -216,7 +216,7 @@ def getMultiTagsText(rootnode, tagname, format):
"""
Returns a list with the text of all subnodes of type I{tagname}
under I{rootnode} (which itself is not parsed) using the given I{format}.
-
+
@type rootnode: xml.dom.Node
@param rootnode: the node to search for I{tagname}
@type tagname: String
@@ -232,9 +232,9 @@ def getMultiTagsText(rootnode, tagname, format):
def makeAtom(pkgname, versionNode):
"""
- creates from the given package name and information in the
+ creates from the given package name and information in the
I{versionNode} a (syntactical) valid portage atom.
-
+
@type pkgname: String
@param pkgname: the name of the package for this atom
@type versionNode: xml.dom.Node
@@ -257,9 +257,9 @@ def makeAtom(pkgname, versionNode):
def makeVersion(versionNode):
"""
- creates from the information in the I{versionNode} a
+ creates from the information in the I{versionNode} a
version string (format <op><version>).
-
+
@type versionNode: xml.dom.Node
@param versionNode: a <vulnerable> or <unaffected> Node that
contains the version information for this atom
@@ -279,17 +279,17 @@ def makeVersion(versionNode):
def match(atom, dbapi, match_type="default"):
"""
- wrapper that calls revisionMatch() or portage.dbapi.dbapi.match() depending on
+ wrapper that calls revisionMatch() or portage.dbapi.dbapi.match() depending on
the given atom.
-
+
@type atom: string
@param atom: a <~ or >~ atom or a normal portage atom that contains the atom to match against
@type dbapi: portage.dbapi.dbapi
@param dbapi: one of the portage databases to use as information source
@type match_type: string
- @param match_type: if != "default" passed as first argument to dbapi.xmatch
+ @param match_type: if != "default" passed as first argument to dbapi.xmatch
to apply the wanted visibility filters
-
+
@rtype: list of strings
@return: a list with the matching versions
"""
@@ -305,15 +305,15 @@ def revisionMatch(revisionAtom, dbapi, match_type="default"):
handler for the special >~, >=~, <=~ and <~ atoms that are supposed to behave
as > and < except that they are limited to the same version, the range only
applies to the revision part.
-
+
@type revisionAtom: string
@param revisionAtom: a <~ or >~ atom that contains the atom to match against
@type dbapi: portage.dbapi.dbapi
@param dbapi: one of the portage databases to use as information source
@type match_type: string
- @param match_type: if != "default" passed as first argument to portdb.xmatch
+ @param match_type: if != "default" passed as first argument to portdb.xmatch
to apply the wanted visibility filters
-
+
@rtype: list of strings
@return: a list with the matching versions
"""
@@ -334,22 +334,19 @@ def revisionMatch(revisionAtom, dbapi, match_type="default"):
if eval(r1+" "+revisionAtom[0:2]+" "+r2):
rValue.append(v)
return rValue
-
+
def getMinUpgrade(vulnerableList, unaffectedList, portdbapi, vardbapi, minimize=True):
"""
- Checks if the state of installed packages matches an atom in
- I{vulnerableList} and returns an update path.
-
- Return value is:
- * None if the system is not affected
- * a list of tuples (a,b) where
- a is a cpv describing an installed vulnerable atom
- b is a cpv describing an uninstalled unaffected atom
- in the same slot as a
- OR the empty string ("") which means no upgrade
- is possible
-
+ Checks if the systemstate is matching an atom in
+ I{vulnerableList} and returns string describing
+ the lowest version for the package that matches an atom in
+ I{unaffectedList} and is greater than the currently installed
+ version. It will return an empty list if the system is affected,
+ and no upgrade is possible or None if the system is not affected.
+ Both I{vulnerableList} and I{unaffectedList} should have the
+ same base package.
+
@type vulnerableList: List of Strings
@param vulnerableList: atoms matching vulnerable package versions
@type unaffectedList: List of Strings
@@ -360,10 +357,12 @@ def getMinUpgrade(vulnerableList, unaffectedList, portdbapi, vardbapi, minimize=
@param vardbapi: Installed package repository
@type minimize: Boolean
@param minimize: True for a least-change upgrade, False for emerge-like algorithm
-
- @rtype: List | None
- @return: None if unaffected or a list of (vuln, upgrade) atoms.
- """
+
+ @rtype: String | None
+ @return: the lowest unaffected version that is greater than
+ the installed version.
+ """
+ rValue = ""
v_installed = reduce(operator.add, [match(v, vardbapi) for v in vulnerableList], [])
u_installed = reduce(operator.add, [match(u, vardbapi) for u in unaffectedList], [])
@@ -385,17 +384,12 @@ def getMinUpgrade(vulnerableList, unaffectedList, portdbapi, vardbapi, minimize=
for vuln in v_installed:
update = ""
- # find the best update path for the vuln atom
for c in avail_updates:
c_pv = portage.catpkgsplit(c)
- if vercmp(c.version, vuln.version) <= 0:
- # c is less or equal than vuln
- continue
- if portdbapi._pkg_str(c, None).slot != \
- vardbapi._pkg_str(vuln, None).slot:
- # upgrade to a different slot
- continue
- if update == "" or (minimize ^ (vercmp(c.version, update.version) > 0)):
+ if vercmp(c.version, vuln.version) > 0 \
+ and (update == "" \
+ or (minimize ^ (vercmp(c.version, update.version) > 0))) \
+ and portdbapi._pkg_str(c, None).slot == vardbapi._pkg_str(vuln, None).slot:
update = c_pv[0]+"/"+c_pv[1]+"-"+c_pv[2]
if c_pv[3] != "r0": # we don't like -r0 for display
update += "-"+c_pv[3]
@@ -407,7 +401,7 @@ def format_date(datestr):
"""
Takes a date (announced, revised) date from a GLSA and formats
it as readable text (i.e. "January 1, 2008").
-
+
@type date: String
@param date: the date string to reformat
@rtype: String
@@ -417,16 +411,16 @@ def format_date(datestr):
splitdate = datestr.split("-", 2)
if len(splitdate) != 3:
return datestr
-
+
# This cannot raise an error as we use () instead of []
splitdate = (int(x) for x in splitdate)
-
+
from datetime import date
try:
d = date(*splitdate)
except ValueError:
return datestr
-
+
# TODO We could format to local date format '%x' here?
return _unicode_decode(d.strftime("%B %d, %Y"),
encoding=_encodings['content'], errors='replace')
@@ -438,7 +432,7 @@ class GlsaTypeException(Exception):
class GlsaFormatException(Exception):
pass
-
+
class GlsaArgumentException(Exception):
pass
@@ -450,9 +444,9 @@ class Glsa:
"""
def __init__(self, myid, myconfig, vardbapi, portdbapi):
"""
- Simple constructor to set the ID, store the config and gets the
+ Simple constructor to set the ID, store the config and gets the
XML data by calling C{self.read()}.
-
+
@type myid: String
@param myid: String describing the id for the GLSA object (standard
GLSAs have an ID of the form YYYYMM-nn) or an existing
@@ -482,7 +476,7 @@ class Glsa:
"""
Here we build the filename from the config and the ID and pass
it to urllib to fetch it from the filesystem or a remote server.
-
+
@rtype: None
@return: None
"""
@@ -505,10 +499,10 @@ class Glsa:
def parse(self, myfile):
"""
- This method parses the XML file and sets up the internal data
+ This method parses the XML file and sets up the internal data
structures by calling the different helper functions in this
module.
-
+
@type myfile: String
@param myfile: Filename to grab the XML data from
@rtype: None
@@ -531,7 +525,7 @@ class Glsa:
self.title = getText(myroot.getElementsByTagName("title")[0], format="strip")
self.synopsis = getText(myroot.getElementsByTagName("synopsis")[0], format="strip")
self.announced = format_date(getText(myroot.getElementsByTagName("announced")[0], format="strip"))
-
+
count = 1
# Support both formats of revised:
# <revised>December 30, 2007: 02</revised>
@@ -543,15 +537,15 @@ class Glsa:
count = revisedEl.getAttribute("count")
elif (self.revised.find(":") >= 0):
(self.revised, count) = self.revised.split(":")
-
+
self.revised = format_date(self.revised)
-
+
try:
self.count = int(count)
except ValueError:
# TODO should this raise a GlsaFormatException?
self.count = 1
-
+
# now the optional and 0-n toplevel, #PCDATA tags and references
try:
self.access = getText(myroot.getElementsByTagName("access")[0], format="strip")
@@ -559,7 +553,7 @@ class Glsa:
self.access = ""
self.bugs = getMultiTagsText(myroot, "bug", format="strip")
self.references = getMultiTagsText(myroot.getElementsByTagName("references")[0], "uri", format="keep")
-
+
# and now the formatted text elements
self.description = getText(myroot.getElementsByTagName("description")[0], format="xml")
self.workaround = getText(myroot.getElementsByTagName("workaround")[0], format="xml")
@@ -569,7 +563,7 @@ class Glsa:
try:
self.background = getText(myroot.getElementsByTagName("background")[0], format="xml")
except IndexError:
- self.background = ""
+ self.background = ""
# finally the interesting tags (product, affected, package)
self.glsatype = myroot.getElementsByTagName("product")[0].getAttribute("type")
@@ -601,10 +595,10 @@ class Glsa:
def dump(self, outstream=sys.stdout, encoding="utf-8"):
"""
- Dumps a plaintext representation of this GLSA to I{outfile} or
+ Dumps a plaintext representation of this GLSA to I{outfile} or
B{stdout} if it is ommitted. You can specify an alternate
I{encoding} if needed (default is utf-8).
-
+
@type outstream: File
@param outfile: Stream that should be used for writing
(defaults to sys.stdout)
@@ -645,13 +639,13 @@ class Glsa:
myreferences = " ".join(r.replace(" ", SPACE_ESCAPE)+NEWLINE_ESCAPE for r in self.references)
outstream.write("\n"+wrap(myreferences, width, caption=_("References: ")))
outstream.write("\n")
-
+
def isVulnerable(self):
"""
Tests if the system is affected by this GLSA by checking if any
vulnerable package versions are installed. Also checks for affected
architectures.
-
+
@rtype: Boolean
@return: True if the system is affected, False if not
"""
@@ -666,12 +660,12 @@ class Glsa:
and None != getMinUpgrade(path["vul_atoms"], path["unaff_atoms"], \
self.portdbapi, self.vardbapi))
return rValue
-
+
def isInjected(self):
"""
Looks if the GLSA ID is in the GLSA checkfile to check if this
GLSA should be marked as applied.
-
+
@rtype: Boolean
@returns: True if the GLSA is in the inject file, False if not
"""
@@ -683,7 +677,7 @@ class Glsa:
def inject(self):
"""
Puts the ID of this GLSA into the GLSA checkfile, so it won't
- show up on future checks. Should be called after a GLSA is
+ show up on future checks. Should be called after a GLSA is
applied or on explicit user request.
@rtype: None
@@ -698,13 +692,13 @@ class Glsa:
checkfile.write(_unicode_decode(self.nr + "\n"))
checkfile.close()
return None
-
+
def getMergeList(self, least_change=True):
"""
Returns the list of package-versions that have to be merged to
- apply this GLSA properly. The versions are as low as possible
+ apply this GLSA properly. The versions are as low as possible
while avoiding downgrades (see L{getMinUpgrade}).
-
+
@type least_change: Boolean
@param least_change: True if the smallest possible upgrade should be selected,
False for an emerge-like algorithm