[virt-tools-list] [libvirt] [PATCH 16/22] apibuild: Fix indentation not multiple of 4
Radostin Stoyanov
rstoyanov1 at gmail.com
Sat Mar 17 13:48:52 UTC 2018
PEP8 recommends that the number of spaces used for indentation of
Python code to be a multiple of four [1] [2].
1: https://lintlyci.github.io/Flake8Rules/rules/E111.html
2: https://lintlyci.github.io/Flake8Rules/rules/E114.html
Signed-off-by: Radostin Stoyanov <rstoyanov1 at gmail.com>
---
docs/apibuild.py | 298 ++++++++++++++++++++++++++++---------------------------
1 file changed, 151 insertions(+), 147 deletions(-)
diff --git a/docs/apibuild.py b/docs/apibuild.py
index bcf7e5c24..9e73b4d27 100755
--- a/docs/apibuild.py
+++ b/docs/apibuild.py
@@ -248,11 +248,12 @@ class index:
return None
d = None
try:
- d = self.identifiers[name]
- d.update(header, module, type, lineno, info, extra, conditionals)
+ d = self.identifiers[name]
+ d.update(header, module, type, lineno, info, extra, conditionals)
except:
- d = identifier(name, header, module, type, lineno, info, extra, conditionals)
- self.identifiers[name] = d
+ d = identifier(name, header, module, type, lineno, info, extra,
+ conditionals)
+ self.identifiers[name] = d
if d is not None and static == 1:
d.set_static(1)
@@ -265,16 +266,18 @@ class index:
return d
- def add(self, name, header, module, static, type, lineno, info=None, extra=None, conditionals=None):
+ def add(self, name, header, module, static, type, lineno, info=None,
+ extra=None, conditionals=None):
if name[0:2] == '__':
return None
d = None
try:
- d = self.identifiers[name]
- d.update(header, module, type, lineno, info, extra, conditionals)
+ d = self.identifiers[name]
+ d.update(header, module, type, lineno, info, extra, conditionals)
except:
- d = identifier(name, header, module, type, lineno, info, extra, conditionals)
- self.identifiers[name] = d
+ d = identifier(name, header, module, type, lineno, info, extra,
+ conditionals)
+ self.identifiers[name] = d
if d is not None and static == 1:
d.set_static(1)
@@ -308,93 +311,94 @@ class index:
def merge(self, idx):
for id in idx.functions.keys():
- #
- # macro might be used to override functions or variables
- # definitions
- #
- if id in self.macros:
- del self.macros[id]
- if id in self.functions:
- self.warning("function %s from %s redeclared in %s" % (
+ #
+ # macro might be used to override functions or variables
+ # definitions
+ #
+ if id in self.macros:
+ del self.macros[id]
+ if id in self.functions:
+ self.warning("function %s from %s redeclared in %s" % (
id, self.functions[id].header, idx.functions[id].header))
- else:
- self.functions[id] = idx.functions[id]
- self.identifiers[id] = idx.functions[id]
+ else:
+ self.functions[id] = idx.functions[id]
+ self.identifiers[id] = idx.functions[id]
for id in idx.variables.keys():
- #
- # macro might be used to override functions or variables
- # definitions
- #
- if id in self.macros:
- del self.macros[id]
- if id in self.variables:
- self.warning("variable %s from %s redeclared in %s" % (
+ #
+ # macro might be used to override functions or variables
+ # definitions
+ #
+ if id in self.macros:
+ del self.macros[id]
+ if id in self.variables:
+ self.warning("variable %s from %s redeclared in %s" % (
id, self.variables[id].header, idx.variables[id].header))
- else:
- self.variables[id] = idx.variables[id]
- self.identifiers[id] = idx.variables[id]
+ else:
+ self.variables[id] = idx.variables[id]
+ self.identifiers[id] = idx.variables[id]
for id in idx.structs.keys():
- if id in self.structs:
- self.warning("struct %s from %s redeclared in %s" % (
+ if id in self.structs:
+ self.warning("struct %s from %s redeclared in %s" % (
id, self.structs[id].header, idx.structs[id].header))
- else:
- self.structs[id] = idx.structs[id]
- self.identifiers[id] = idx.structs[id]
+ else:
+ self.structs[id] = idx.structs[id]
+ self.identifiers[id] = idx.structs[id]
for id in idx.unions.keys():
- if id in self.unions:
- print("union %s from %s redeclared in %s" % (
+ if id in self.unions:
+ print("union %s from %s redeclared in %s" % (
id, self.unions[id].header, idx.unions[id].header))
- else:
- self.unions[id] = idx.unions[id]
- self.identifiers[id] = idx.unions[id]
+ else:
+ self.unions[id] = idx.unions[id]
+ self.identifiers[id] = idx.unions[id]
for id in idx.typedefs.keys():
- if id in self.typedefs:
- self.warning("typedef %s from %s redeclared in %s" % (
+ if id in self.typedefs:
+ self.warning("typedef %s from %s redeclared in %s" % (
id, self.typedefs[id].header, idx.typedefs[id].header))
- else:
- self.typedefs[id] = idx.typedefs[id]
- self.identifiers[id] = idx.typedefs[id]
+ else:
+ self.typedefs[id] = idx.typedefs[id]
+ self.identifiers[id] = idx.typedefs[id]
for id in idx.macros.keys():
- #
- # macro might be used to override functions or variables
- # definitions
- #
- if id in self.variables:
- continue
- if id in self.functions:
- continue
- if id in self.enums:
- continue
- if id in self.macros:
- self.warning("macro %s from %s redeclared in %s" % (
+ #
+ # macro might be used to override functions or variables
+ # definitions
+ #
+ if id in self.variables:
+ continue
+ if id in self.functions:
+ continue
+ if id in self.enums:
+ continue
+ if id in self.macros:
+ self.warning("macro %s from %s redeclared in %s" % (
id, self.macros[id].header, idx.macros[id].header))
- else:
- self.macros[id] = idx.macros[id]
- self.identifiers[id] = idx.macros[id]
+ else:
+ self.macros[id] = idx.macros[id]
+ self.identifiers[id] = idx.macros[id]
for id in idx.enums.keys():
- if id in self.enums:
- self.warning("enum %s from %s redeclared in %s" % (
+ if id in self.enums:
+ self.warning("enum %s from %s redeclared in %s" % (
id, self.enums[id].header, idx.enums[id].header))
- else:
- self.enums[id] = idx.enums[id]
- self.identifiers[id] = idx.enums[id]
+ else:
+ self.enums[id] = idx.enums[id]
+ self.identifiers[id] = idx.enums[id]
def merge_public(self, idx):
for id in idx.functions.keys():
- if id in self.functions:
- # check that function condition agrees with header
- if idx.functions[id].conditionals != \
- self.functions[id].conditionals:
- self.warning("Header condition differs from Function"
- " for %s:" % id)
- self.warning(" H: %s" % self.functions[id].conditionals)
- self.warning(" C: %s" % idx.functions[id].conditionals)
- up = idx.functions[id]
- self.functions[id].update(None, up.module, up.type, up.info, up.extra)
- # else:
- # print("Function %s from %s is not declared in headers" % (
- # id, idx.functions[id].module))
- # TODO: do the same for variables.
+ if id in self.functions:
+ # check that function condition agrees with header
+ if idx.functions[id].conditionals != \
+ self.functions[id].conditionals:
+ self.warning("Header condition differs from Function"
+ " for %s:" % id)
+ self.warning(" H: %s" % self.functions[id].conditionals)
+ self.warning(" C: %s" % idx.functions[id].conditionals)
+ up = idx.functions[id]
+ self.functions[id].update(None, up.module, up.type, up.info,
+ up.extra)
+ # else:
+ # print("Function %s from %s is not declared in headers" % (
+ # id, idx.functions[id].module))
+ # TODO: do the same for variables.
def analyze_dict(self, type, dict):
count = 0
@@ -726,7 +730,7 @@ class CParser:
elif line[i] == '*':
return line[:i] + line[i + 1:]
else:
- return line
+ return line
return line
def cleanupComment(self):
@@ -877,11 +881,11 @@ class CParser:
return((args, desc))
- #
- # Parse a comment block and merge the information found in the
- # parameters descriptions, finally returns a block as complete
- # as possible
- #
+ #
+ # Parse a comment block and merge the information found in the
+ # parameters descriptions, finally returns a block as complete
+ # as possible
+ #
def mergeFunctionComment(self, name, description, quiet=0):
global ignored_functions
@@ -989,9 +993,9 @@ class CParser:
desc = desc.strip()
if quiet == 0:
- #
- # report missing comments
- #
+ #
+ # report missing comments
+ #
i = 0
while i < nbargs:
if args[i][2] is None and args[i][0] != "void" and args[i][1] is not None:
@@ -1023,7 +1027,7 @@ class CParser:
if token is None:
return None
if token[0] == 'preproc':
- # TODO macros with arguments
+ # TODO macros with arguments
name = token[1]
lst = []
token = self.lexer.token()
@@ -1108,11 +1112,11 @@ class CParser:
token = self.lexer.token()
return token
- #
- # token acquisition on top of the lexer, it handle internally
- # preprocessor and comments since they are logically not part of
- # the program structure.
- #
+ #
+ # token acquisition on top of the lexer, it handle internally
+ # preprocessor and comments since they are logically not part of
+ # the program structure.
+ #
def push(self, tok):
self.lexer.push(tok)
@@ -1149,9 +1153,9 @@ class CParser:
return token
return None
- #
- # Parse a typedef, it records the type and its name.
- #
+ #
+ # Parse a typedef, it records the type and its name.
+ #
def parseTypedef(self, token):
if token is None:
return None
@@ -1161,7 +1165,7 @@ class CParser:
return None
base_type = self.type
type = base_type
- #self.debug("end typedef type", token)
+ # self.debug("end typedef type", token)
while token is not None:
if token[0] == "name":
name = token[1]
@@ -1186,7 +1190,7 @@ class CParser:
else:
self.error("parsing typedef: expecting a name")
return token
- #self.debug("end typedef", token)
+ # self.debug("end typedef", token)
if token is not None and token[0] == 'sep' and token[1] == ',':
type = base_type
token = self.token()
@@ -1204,10 +1208,10 @@ class CParser:
token = self.token()
return token
- #
- # Parse a C code block, used for functions it parse till
- # the balancing } included
- #
+ #
+ # Parse a C code block, used for functions it parse till
+ # the balancing } included
+ #
def parseBlock(self, token):
while token is not None:
if token[0] == "sep" and token[1] == "{":
@@ -1243,27 +1247,27 @@ class CParser:
token = self.token()
return token
- #
- # Parse a C struct definition till the balancing }
- #
+ #
+ # Parse a C struct definition till the balancing }
+ #
def parseStruct(self, token):
fields = []
- #self.debug("start parseStruct", token)
+ # self.debug("start parseStruct", token)
while token is not None:
if token[0] == "sep" and token[1] == "{":
token = self.token()
token = self.parseTypeBlock(token)
elif token[0] == "sep" and token[1] == "}":
self.struct_fields = fields
- #self.debug("end parseStruct", token)
- #print(fields)
+ # self.debug("end parseStruct", token)
+ # print(fields)
token = self.token()
return token
else:
base_type = self.type
- #self.debug("before parseType", token)
+ # self.debug("before parseType", token)
token = self.parseType(token)
- #self.debug("after parseType", token)
+ # self.debug("after parseType", token)
if token is not None and token[0] == "name":
fname = token[1]
token = self.token()
@@ -1294,13 +1298,13 @@ class CParser:
token = self.token()
self.type = base_type
self.struct_fields = fields
- #self.debug("end parseStruct", token)
- #print(fields)
+ # self.debug("end parseStruct", token)
+ # print(fields)
return token
- #
- # Parse a C union definition till the balancing }
- #
+ #
+ # Parse a C union definition till the balancing }
+ #
def parseUnion(self, token):
fields = []
# self.debug("start parseUnion", token)
@@ -1348,9 +1352,9 @@ class CParser:
# print(fields)
return token
- #
- # Parse a C enum block, parse till the balancing }
- #
+ #
+ # Parse a C enum block, parse till the balancing }
+ #
def parseEnumBlock(self, token):
self.enums = []
name = None
@@ -1513,10 +1517,10 @@ class CParser:
return token
- #
- # Parse a C definition block, used for structs or unions it parse till
- # the balancing }
- #
+ #
+ # Parse a C definition block, used for structs or unions it parse till
+ # the balancing }
+ #
def parseTypeBlock(self, token):
while token is not None:
if token[0] == "sep" and token[1] == "{":
@@ -1529,11 +1533,11 @@ class CParser:
token = self.token()
return token
- #
- # Parse a type: the fact that the type name can either occur after
- # the definition or within the definition makes it a little harder
- # if inside, the name token is pushed back before returning
- #
+ #
+ # Parse a type: the fact that the type name can either occur after
+ # the definition or within the definition makes it a little harder
+ # if inside, the name token is pushed back before returning
+ #
def parseType(self, token):
self.type = ""
self.struct_fields = []
@@ -1711,9 +1715,9 @@ class CParser:
self.type = self.type + " " + token[1]
token = self.token()
- #
- # if there is a parenthesis here, this means a function type
- #
+ #
+ # if there is a parenthesis here, this means a function type
+ #
if token is not None and token[0] == "sep" and token[1] == '(':
self.type = self.type + token[1]
token = self.token()
@@ -1744,9 +1748,9 @@ class CParser:
token = nametok
return token
- #
- # do some lookahead for arrays
- #
+ #
+ # do some lookahead for arrays
+ #
if token is not None and token[0] == "name":
nametok = token
token = self.token()
@@ -1766,7 +1770,7 @@ class CParser:
self.error("parsing array type, ']' expected", token)
return token
elif token is not None and token[0] == "sep" and token[1] == ':':
- # remove :12 in case it's a limited int size
+ # remove :12 in case it's a limited int size
token = self.token()
token = self.token()
self.lexer.push(token)
@@ -1774,9 +1778,9 @@ class CParser:
return token
- #
- # Parse a signature: '(' has been parsed and we scan the type definition
- # up to the ')' included
+ #
+ # Parse a signature: '(' has been parsed and we scan the type definition
+ # up to the ')' included
def parseSignature(self, token):
signature = []
if token is not None and token[0] == "sep" and token[1] == ')':
@@ -1792,7 +1796,7 @@ class CParser:
token = self.token()
continue
elif token is not None and token[0] == "sep" and token[1] == ')':
- # only the type was provided
+ # only the type was provided
if self.type == "...":
signature.append((self.type, "...", None))
else:
@@ -1886,10 +1890,10 @@ class CParser:
"field '%s', use long long instead")
% (name, field[1]))
- #
- # Parse a global definition, be it a type, variable or function
- # the extern "C" blocks are a bit nasty and require it to recurse.
- #
+ #
+ # Parse a global definition, be it a type, variable or function
+ # the extern "C" blocks are a bit nasty and require it to recurse.
+ #
def parseGlobal(self, token):
static = 0
if token[1] == 'extern':
@@ -1946,9 +1950,9 @@ class CParser:
token = self.token()
if token is not None and token[0] == "op" and token[1] == "=":
- #
- # Skip the initialization of the variable
- #
+ #
+ # Skip the initialization of the variable
+ #
token = self.token()
if token[0] == 'sep' and token[1] == '{':
token = self.token()
--
2.14.3
More information about the virt-tools-list
mailing list