summaryrefslogtreecommitdiffstats
path: root/src/tools/vppapigen/vppapigen.py
diff options
context:
space:
mode:
Diffstat (limited to 'src/tools/vppapigen/vppapigen.py')
-rwxr-xr-xsrc/tools/vppapigen/vppapigen.py813
1 files changed, 414 insertions, 399 deletions
diff --git a/src/tools/vppapigen/vppapigen.py b/src/tools/vppapigen/vppapigen.py
index 8415c28fb7b..c7bc68180a8 100755
--- a/src/tools/vppapigen/vppapigen.py
+++ b/src/tools/vppapigen/vppapigen.py
@@ -10,9 +10,10 @@ from subprocess import Popen, PIPE
import ply.lex as lex
import ply.yacc as yacc
-assert sys.version_info >= (3, 5), \
- "Not supported Python version: {}".format(sys.version)
-log = logging.getLogger('vppapigen')
+assert sys.version_info >= (3, 5), "Not supported Python version: {}".format(
+ sys.version
+)
+log = logging.getLogger("vppapigen")
# Ensure we don't leave temporary files around
sys.dont_write_bytecode = True
@@ -28,11 +29,10 @@ seen_imports = {}
def global_type_add(name, obj):
- '''Add new type to the dictionary of types '''
- type_name = 'vl_api_' + name + '_t'
+ """Add new type to the dictionary of types"""
+ type_name = "vl_api_" + name + "_t"
if type_name in global_types:
- raise KeyError("Attempted redefinition of {!r} with {!r}.".format(
- name, obj))
+ raise KeyError("Attempted redefinition of {!r} with {!r}.".format(name, obj))
global_types[type_name] = obj
@@ -49,104 +49,104 @@ class VPPAPILexer:
self.filename = filename
reserved = {
- 'service': 'SERVICE',
- 'rpc': 'RPC',
- 'returns': 'RETURNS',
- 'null': 'NULL',
- 'stream': 'STREAM',
- 'events': 'EVENTS',
- 'define': 'DEFINE',
- 'typedef': 'TYPEDEF',
- 'enum': 'ENUM',
- 'enumflag': 'ENUMFLAG',
- 'typeonly': 'TYPEONLY',
- 'manual_print': 'MANUAL_PRINT',
- 'manual_endian': 'MANUAL_ENDIAN',
- 'dont_trace': 'DONT_TRACE',
- 'autoreply': 'AUTOREPLY',
- 'autoendian': 'AUTOENDIAN',
- 'option': 'OPTION',
- 'u8': 'U8',
- 'u16': 'U16',
- 'u32': 'U32',
- 'u64': 'U64',
- 'i8': 'I8',
- 'i16': 'I16',
- 'i32': 'I32',
- 'i64': 'I64',
- 'f64': 'F64',
- 'bool': 'BOOL',
- 'string': 'STRING',
- 'import': 'IMPORT',
- 'true': 'TRUE',
- 'false': 'FALSE',
- 'union': 'UNION',
- 'counters': 'COUNTERS',
- 'paths': 'PATHS',
- 'units': 'UNITS',
- 'severity': 'SEVERITY',
- 'type': 'TYPE',
- 'description': 'DESCRIPTION',
+ "service": "SERVICE",
+ "rpc": "RPC",
+ "returns": "RETURNS",
+ "null": "NULL",
+ "stream": "STREAM",
+ "events": "EVENTS",
+ "define": "DEFINE",
+ "typedef": "TYPEDEF",
+ "enum": "ENUM",
+ "enumflag": "ENUMFLAG",
+ "typeonly": "TYPEONLY",
+ "manual_print": "MANUAL_PRINT",
+ "manual_endian": "MANUAL_ENDIAN",
+ "dont_trace": "DONT_TRACE",
+ "autoreply": "AUTOREPLY",
+ "autoendian": "AUTOENDIAN",
+ "option": "OPTION",
+ "u8": "U8",
+ "u16": "U16",
+ "u32": "U32",
+ "u64": "U64",
+ "i8": "I8",
+ "i16": "I16",
+ "i32": "I32",
+ "i64": "I64",
+ "f64": "F64",
+ "bool": "BOOL",
+ "string": "STRING",
+ "import": "IMPORT",
+ "true": "TRUE",
+ "false": "FALSE",
+ "union": "UNION",
+ "counters": "COUNTERS",
+ "paths": "PATHS",
+ "units": "UNITS",
+ "severity": "SEVERITY",
+ "type": "TYPE",
+ "description": "DESCRIPTION",
}
- tokens = ['STRING_LITERAL',
- 'ID', 'NUM'] + list(reserved.values())
+ tokens = ["STRING_LITERAL", "ID", "NUM"] + list(reserved.values())
- t_ignore_LINE_COMMENT = '//.*'
+ t_ignore_LINE_COMMENT = "//.*"
def t_FALSE(self, t):
- r'false'
+ r"false"
t.value = False
return t
def t_TRUE(self, t):
- r'false'
+ r"false"
t.value = True
return t
def t_NUM(self, t):
- r'0[xX][0-9a-fA-F]+|-?\d+\.?\d*'
- base = 16 if t.value.startswith('0x') else 10
- if '.' in t.value:
+ r"0[xX][0-9a-fA-F]+|-?\d+\.?\d*"
+ base = 16 if t.value.startswith("0x") else 10
+ if "." in t.value:
t.value = float(t.value)
else:
t.value = int(t.value, base)
return t
def t_ID(self, t):
- r'[a-zA-Z_][a-zA-Z_0-9]*'
+ r"[a-zA-Z_][a-zA-Z_0-9]*"
# Check for reserved words
- t.type = VPPAPILexer.reserved.get(t.value, 'ID')
+ t.type = VPPAPILexer.reserved.get(t.value, "ID")
return t
# C string
def t_STRING_LITERAL(self, t):
- r'\"([^\\\n]|(\\.))*?\"'
- t.value = str(t.value).replace("\"", "")
+ r"\"([^\\\n]|(\\.))*?\" "
+ t.value = str(t.value).replace('"', "")
return t
# C or C++ comment (ignore)
def t_comment(self, t):
- r'(/\*(.|\n)*?\*/)|(//.*)'
- t.lexer.lineno += t.value.count('\n')
+ r"(/\*(.|\n)*?\*/)|(//.*)"
+ t.lexer.lineno += t.value.count("\n")
# Error handling rule
def t_error(self, t):
- raise ParseError("Illegal character '{}' ({})"
- "in {}: line {}".format(t.value[0],
- hex(ord(t.value[0])),
- self.filename,
- t.lexer.lineno))
+ raise ParseError(
+ "Illegal character '{}' ({})"
+ "in {}: line {}".format(
+ t.value[0], hex(ord(t.value[0])), self.filename, t.lexer.lineno
+ )
+ )
# Define a rule so we can track line numbers
def t_newline(self, t):
- r'\n+'
+ r"\n+"
t.lexer.lineno += len(t.value)
literals = ":{}[];=.,"
# A string containing ignored characters (spaces and tabs)
- t_ignore = ' \t'
+ t_ignore = " \t"
def vla_mark_length_field(block):
@@ -164,23 +164,25 @@ def vla_is_last_check(name, block):
vla = True
if i + 1 < len(block):
raise ValueError(
- 'VLA field "{}" must be the last field in message "{}"'
- .format(b.fieldname, name))
- elif b.fieldtype.startswith('vl_api_'):
+ 'VLA field "{}" must be the last field in message "{}"'.format(
+ b.fieldname, name
+ )
+ )
+ elif b.fieldtype.startswith("vl_api_"):
if global_types[b.fieldtype].vla:
vla = True
if i + 1 < len(block):
raise ValueError(
'VLA field "{}" must be the last '
- 'field in message "{}"'
- .format(b.fieldname, name))
- elif b.fieldtype == 'string' and b.length == 0:
+ 'field in message "{}"'.format(b.fieldname, name)
+ )
+ elif b.fieldtype == "string" and b.length == 0:
vla = True
if i + 1 < len(block):
raise ValueError(
'VLA field "{}" must be the last '
- 'field in message "{}"'
- .format(b.fieldname, name))
+ 'field in message "{}"'.format(b.fieldname, name)
+ )
return vla
@@ -192,10 +194,9 @@ class Processable:
class Service(Processable):
- type = 'Service'
+ type = "Service"
- def __init__(self, caller, reply, events=None, stream_message=None,
- stream=False):
+ def __init__(self, caller, reply, events=None, stream_message=None, stream=False):
self.caller = caller
self.reply = reply
self.stream = stream
@@ -204,7 +205,7 @@ class Service(Processable):
class Typedef(Processable):
- type = 'Typedef'
+ type = "Typedef"
def __init__(self, name, flags, block):
self.name = name
@@ -214,9 +215,9 @@ class Typedef(Processable):
self.manual_print = False
self.manual_endian = False
for f in flags:
- if f == 'manual_print':
+ if f == "manual_print":
self.manual_print = True
- elif f == 'manual_endian':
+ elif f == "manual_endian":
self.manual_endian = True
global_type_add(name, self)
@@ -224,14 +225,14 @@ class Typedef(Processable):
vla_mark_length_field(self.block)
def process(self, result):
- result['types'].append(self)
+ result["types"].append(self)
def __repr__(self):
return self.name + str(self.flags) + str(self.block)
class Using(Processable):
- type = 'Using'
+ type = "Using"
def __init__(self, name, flags, alias):
self.name = name
@@ -243,16 +244,15 @@ class Using(Processable):
self.manual_print = False
self.manual_endian = False
for f in flags:
- if f == 'manual_print':
+ if f == "manual_print":
self.manual_print = True
- elif f == 'manual_endian':
+ elif f == "manual_endian":
self.manual_endian = True
if isinstance(alias, Array):
- a = {'type': alias.fieldtype,
- 'length': alias.length}
+ a = {"type": alias.fieldtype, "length": alias.length}
else:
- a = {'type': alias.fieldtype}
+ a = {"type": alias.fieldtype}
self.alias = a
self.using = alias
@@ -265,14 +265,14 @@ class Using(Processable):
global_type_add(name, self)
def process(self, result): # -> Dict
- result['types'].append(self)
+ result["types"].append(self)
def __repr__(self):
return self.name + str(self.alias)
class Union(Processable):
- type = 'Union'
+ type = "Union"
def __init__(self, name, flags, block):
self.manual_print = False
@@ -280,9 +280,9 @@ class Union(Processable):
self.name = name
for f in flags:
- if f == 'manual_print':
+ if f == "manual_print":
self.manual_print = True
- elif f == 'manual_endian':
+ elif f == "manual_endian":
self.manual_endian = True
self.block = block
@@ -292,14 +292,14 @@ class Union(Processable):
global_type_add(name, self)
def process(self, result):
- result['types'].append(self)
+ result["types"].append(self)
def __repr__(self):
return str(self.block)
class Define(Processable):
- type = 'Define'
+ type = "Define"
def __init__(self, name, flags, block):
self.name = name
@@ -312,15 +312,15 @@ class Define(Processable):
self.autoendian = 0
self.options = {}
for f in flags:
- if f == 'dont_trace':
+ if f == "dont_trace":
self.dont_trace = True
- elif f == 'manual_print':
+ elif f == "manual_print":
self.manual_print = True
- elif f == 'manual_endian':
+ elif f == "manual_endian":
self.manual_endian = True
- elif f == 'autoreply':
+ elif f == "autoreply":
self.autoreply = True
- elif f == 'autoendian':
+ elif f == "autoendian":
self.autoendian = 1
remove = []
@@ -337,12 +337,11 @@ class Define(Processable):
self.crc = str(block).encode()
def autoreply_block(self, name, parent):
- block = [Field('u32', 'context'),
- Field('i32', 'retval')]
+ block = [Field("u32", "context"), Field("i32", "retval")]
# inherit the parent's options
for k, v in parent.options.items():
block.append(Option(k, v))
- return Define(name + '_reply', [], block)
+ return Define(name + "_reply", [], block)
def process(self, result): # -> Dict
tname = self.__class__.__name__
@@ -355,9 +354,9 @@ class Define(Processable):
class Enum(Processable):
- type = 'Enum'
+ type = "Enum"
- def __init__(self, name, block, enumtype='u32'):
+ def __init__(self, name, block, enumtype="u32"):
self.name = name
self.enumtype = enumtype
self.vla = False
@@ -369,47 +368,50 @@ class Enum(Processable):
bc_set = False
for b in block:
- if 'value' in b:
- count = b['value']
+ if "value" in b:
+ count = b["value"]
else:
count += 1
- block2.append([b['id'], count])
+ block2.append([b["id"], count])
try:
- if b['option']['backwards_compatible']:
+ if b["option"]["backwards_compatible"]:
pass
bc_set = True
except KeyError:
- block3.append([b['id'], count])
+ block3.append([b["id"], count])
if bc_set:
- raise ValueError("Backward compatible enum must "
- "be last {!r} {!r}"
- .format(name, b['id']))
+ raise ValueError(
+ "Backward compatible enum must "
+ "be last {!r} {!r}".format(name, b["id"])
+ )
self.block = block2
self.crc = str(block3).encode()
global_type_add(name, self)
def process(self, result):
- result['types'].append(self)
+ result["types"].append(self)
def __repr__(self):
return self.name + str(self.block)
class EnumFlag(Enum):
- type = 'EnumFlag'
+ type = "EnumFlag"
- def __init__(self, name, block, enumtype='u32'):
+ def __init__(self, name, block, enumtype="u32"):
super(EnumFlag, self).__init__(name, block, enumtype)
for b in self.block:
if bin(b[1])[2:].count("1") > 1:
- raise TypeError("%s is not a flag enum. No element in a "
- "flag enum may have more than a "
- "single bit set." % self.name)
+ raise TypeError(
+ "%s is not a flag enum. No element in a "
+ "flag enum may have more than a "
+ "single bit set." % self.name
+ )
class Import(Processable):
- type = 'Import'
+ type = "Import"
_initialized = False
def __new__(cls, *args, **kwargs):
@@ -440,7 +442,7 @@ class Import(Processable):
class Option(Processable):
- type = 'Option'
+ type = "Option"
def __init__(self, option, value=None):
self.option = option
@@ -458,7 +460,7 @@ class Option(Processable):
class Array(Processable):
- type = 'Array'
+ type = "Array"
def __init__(self, fieldtype, name, length, modern_vla=False):
self.fieldtype = fieldtype
@@ -474,12 +476,11 @@ class Array(Processable):
self.vla = False
def __repr__(self):
- return str([self.fieldtype, self.fieldname, self.length,
- self.lengthfield])
+ return str([self.fieldtype, self.fieldname, self.length, self.lengthfield])
class Field(Processable):
- type = 'Field'
+ type = "Field"
def __init__(self, fieldtype, name, limit=None):
# limit field has been expanded to an options dict.
@@ -487,13 +488,14 @@ class Field(Processable):
self.fieldtype = fieldtype
self.is_lengthfield = False
- if self.fieldtype == 'string':
- raise ValueError("The string type {!r} is an "
- "array type ".format(name))
+ if self.fieldtype == "string":
+ raise ValueError("The string type {!r} is an " "array type ".format(name))
if name in keyword.kwlist:
- raise ValueError("Fieldname {!r} is a python keyword and is not "
- "accessible via the python API. ".format(name))
+ raise ValueError(
+ "Fieldname {!r} is a python keyword and is not "
+ "accessible via the python API. ".format(name)
+ )
self.fieldname = name
self.limit = limit
@@ -502,35 +504,34 @@ class Field(Processable):
class Counter(Processable):
- type = 'Counter'
+ type = "Counter"
def __init__(self, path, counter):
self.name = path
self.block = counter
def process(self, result): # -> Dict
- result['Counters'].append(self)
+ result["Counters"].append(self)
class Paths(Processable):
- type = 'Paths'
+ type = "Paths"
def __init__(self, pathset):
self.paths = pathset
def __repr__(self):
- return "%s(paths=%s)" % (
- self.__class__.__name__, self.paths
- )
+ return "%s(paths=%s)" % (self.__class__.__name__, self.paths)
class Coord:
- """ Coordinates of a syntactic element. Consists of:
- - File name
- - Line number
- - (optional) column number, for the Lexer
+ """Coordinates of a syntactic element. Consists of:
+ - File name
+ - Line number
+ - (optional) column number, for the Lexer
"""
- __slots__ = ('file', 'line', 'column', '__weakref__')
+
+ __slots__ = ("file", "line", "column", "__weakref__")
def __init__(self, file, line, column=None):
self.file = file
@@ -568,49 +569,47 @@ class VPPAPIParser:
self.logger.warning("%s: %s" % (coord, msg))
def _coord(self, lineno, column=None):
- return Coord(
- file=self.filename,
- line=lineno, column=column)
+ return Coord(file=self.filename, line=lineno, column=column)
def _token_coord(self, p, token_idx):
- """ Returns the coordinates for the YaccProduction object 'p' indexed
- with 'token_idx'. The coordinate includes the 'lineno' and
- 'column'. Both follow the lex semantic, starting from 1.
+ """Returns the coordinates for the YaccProduction object 'p' indexed
+ with 'token_idx'. The coordinate includes the 'lineno' and
+ 'column'. Both follow the lex semantic, starting from 1.
"""
- last_cr = p.lexer.lexdata.rfind('\n', 0, p.lexpos(token_idx))
+ last_cr = p.lexer.lexdata.rfind("\n", 0, p.lexpos(token_idx))
if last_cr < 0:
last_cr = -1
- column = (p.lexpos(token_idx) - (last_cr))
+ column = p.lexpos(token_idx) - (last_cr)
return self._coord(p.lineno(token_idx), column)
def p_slist(self, p):
- '''slist : stmt
- | slist stmt'''
+ """slist : stmt
+ | slist stmt"""
if len(p) == 2:
p[0] = [p[1]]
else:
p[0] = p[1] + [p[2]]
def p_stmt(self, p):
- '''stmt : define
- | typedef
- | option
- | import
- | enum
- | enumflag
- | union
- | service
- | paths
- | counters'''
+ """stmt : define
+ | typedef
+ | option
+ | import
+ | enum
+ | enumflag
+ | union
+ | service
+ | paths
+ | counters"""
p[0] = p[1]
def p_import(self, p):
- '''import : IMPORT STRING_LITERAL ';' '''
+ """import : IMPORT STRING_LITERAL ';'"""
p[0] = Import(p[2], revision=self.revision)
def p_path_elements(self, p):
- '''path_elements : path_element
- | path_elements path_element'''
+ """path_elements : path_element
+ | path_elements path_element"""
if len(p) == 2:
p[0] = p[1]
else:
@@ -620,20 +619,20 @@ class VPPAPIParser:
p[0] = p[1] + [p[2]]
def p_path_element(self, p):
- '''path_element : STRING_LITERAL STRING_LITERAL ';' '''
- p[0] = {'path': p[1], 'counter': p[2]}
+ """path_element : STRING_LITERAL STRING_LITERAL ';'"""
+ p[0] = {"path": p[1], "counter": p[2]}
def p_paths(self, p):
- '''paths : PATHS '{' path_elements '}' ';' '''
+ """paths : PATHS '{' path_elements '}' ';'"""
p[0] = Paths(p[3])
def p_counters(self, p):
- '''counters : COUNTERS ID '{' counter_elements '}' ';' '''
+ """counters : COUNTERS ID '{' counter_elements '}' ';'"""
p[0] = Counter(p[2], p[4])
def p_counter_elements(self, p):
- '''counter_elements : counter_element
- | counter_elements counter_element'''
+ """counter_elements : counter_element
+ | counter_elements counter_element"""
if len(p) == 2:
p[0] = [p[1]]
else:
@@ -643,46 +642,47 @@ class VPPAPIParser:
p[0] = p[1] + [p[2]]
def p_counter_element(self, p):
- '''counter_element : ID '{' counter_statements '}' ';' '''
- p[0] = {**{'name': p[1]}, **p[3]}
+ """counter_element : ID '{' counter_statements '}' ';'"""
+ p[0] = {**{"name": p[1]}, **p[3]}
def p_counter_statements(self, p):
- '''counter_statements : counter_statement
- | counter_statements counter_statement'''
+ """counter_statements : counter_statement
+ | counter_statements counter_statement"""
if len(p) == 2:
p[0] = p[1]
else:
p[0] = {**p[1], **p[2]}
def p_counter_statement(self, p):
- '''counter_statement : SEVERITY ID ';'
- | UNITS STRING_LITERAL ';'
- | DESCRIPTION STRING_LITERAL ';'
- | TYPE ID ';' '''
+ """counter_statement : SEVERITY ID ';'
+ | UNITS STRING_LITERAL ';'
+ | DESCRIPTION STRING_LITERAL ';'
+ | TYPE ID ';'"""
p[0] = {p[1]: p[2]}
def p_service(self, p):
- '''service : SERVICE '{' service_statements '}' ';' '''
+ """service : SERVICE '{' service_statements '}' ';'"""
p[0] = p[3]
def p_service_statements(self, p):
- '''service_statements : service_statement
- | service_statements service_statement'''
+ """service_statements : service_statement
+ | service_statements service_statement"""
if len(p) == 2:
p[0] = [p[1]]
else:
p[0] = p[1] + [p[2]]
def p_service_statement(self, p):
- '''service_statement : RPC ID RETURNS NULL ';'
- | RPC ID RETURNS ID ';'
- | RPC ID RETURNS STREAM ID ';'
- | RPC ID RETURNS ID EVENTS event_list ';' '''
+ """service_statement : RPC ID RETURNS NULL ';'
+ | RPC ID RETURNS ID ';'
+ | RPC ID RETURNS STREAM ID ';'
+ | RPC ID RETURNS ID EVENTS event_list ';'"""
if p[2] == p[4]:
# Verify that caller and reply differ
self._parse_error(
- 'Reply ID ({}) should not be equal to Caller ID'.format(p[2]),
- self._token_coord(p, 1))
+ "Reply ID ({}) should not be equal to Caller ID".format(p[2]),
+ self._token_coord(p, 1),
+ )
if len(p) == 8:
p[0] = Service(p[2], p[4], p[6])
elif len(p) == 7:
@@ -691,280 +691,283 @@ class VPPAPIParser:
p[0] = Service(p[2], p[4])
def p_service_statement2(self, p):
- '''service_statement : RPC ID RETURNS ID STREAM ID ';' '''
+ """service_statement : RPC ID RETURNS ID STREAM ID ';'"""
p[0] = Service(p[2], p[4], stream_message=p[6], stream=True)
def p_event_list(self, p):
- '''event_list : events
- | event_list events '''
+ """event_list : events
+ | event_list events"""
if len(p) == 2:
p[0] = [p[1]]
else:
p[0] = p[1] + [p[2]]
def p_event(self, p):
- '''events : ID
- | ID ',' '''
+ """events : ID
+ | ID ','"""
p[0] = p[1]
def p_enum(self, p):
- '''enum : ENUM ID '{' enum_statements '}' ';' '''
+ """enum : ENUM ID '{' enum_statements '}' ';'"""
p[0] = Enum(p[2], p[4])
def p_enum_type(self, p):
- ''' enum : ENUM ID ':' enum_size '{' enum_statements '}' ';' '''
+ """enum : ENUM ID ':' enum_size '{' enum_statements '}' ';'"""
if len(p) == 9:
p[0] = Enum(p[2], p[6], enumtype=p[4])
else:
p[0] = Enum(p[2], p[4])
def p_enumflag(self, p):
- '''enumflag : ENUMFLAG ID '{' enum_statements '}' ';' '''
+ """enumflag : ENUMFLAG ID '{' enum_statements '}' ';'"""
p[0] = EnumFlag(p[2], p[4])
def p_enumflag_type(self, p):
- ''' enumflag : ENUMFLAG ID ':' enumflag_size '{' enum_statements '}' ';' ''' # noqa : E502
+ """enumflag : ENUMFLAG ID ':' enumflag_size '{' enum_statements '}' ';'""" # noqa : E502
if len(p) == 9:
p[0] = EnumFlag(p[2], p[6], enumtype=p[4])
else:
p[0] = EnumFlag(p[2], p[4])
def p_enum_size(self, p):
- ''' enum_size : U8
- | U16
- | U32
- | I8
- | I16
- | I32 '''
+ """enum_size : U8
+ | U16
+ | U32
+ | I8
+ | I16
+ | I32"""
p[0] = p[1]
def p_enumflag_size(self, p):
- ''' enumflag_size : U8
- | U16
- | U32 '''
+ """enumflag_size : U8
+ | U16
+ | U32"""
p[0] = p[1]
def p_define(self, p):
- '''define : DEFINE ID '{' block_statements_opt '}' ';' '''
+ """define : DEFINE ID '{' block_statements_opt '}' ';'"""
self.fields = []
p[0] = Define(p[2], [], p[4])
def p_define_flist(self, p):
- '''define : flist DEFINE ID '{' block_statements_opt '}' ';' '''
+ """define : flist DEFINE ID '{' block_statements_opt '}' ';'"""
# Legacy typedef
- if 'typeonly' in p[1]:
- self._parse_error('legacy typedef. use typedef: {} {}[{}];'
- .format(p[1], p[2], p[4]),
- self._token_coord(p, 1))
+ if "typeonly" in p[1]:
+ self._parse_error(
+ "legacy typedef. use typedef: {} {}[{}];".format(p[1], p[2], p[4]),
+ self._token_coord(p, 1),
+ )
else:
p[0] = Define(p[3], p[1], p[5])
def p_flist(self, p):
- '''flist : flag
- | flist flag'''
+ """flist : flag
+ | flist flag"""
if len(p) == 2:
p[0] = [p[1]]
else:
p[0] = p[1] + [p[2]]
def p_flag(self, p):
- '''flag : MANUAL_PRINT
- | MANUAL_ENDIAN
- | DONT_TRACE
- | TYPEONLY
- | AUTOENDIAN
- | AUTOREPLY'''
+ """flag : MANUAL_PRINT
+ | MANUAL_ENDIAN
+ | DONT_TRACE
+ | TYPEONLY
+ | AUTOENDIAN
+ | AUTOREPLY"""
if len(p) == 1:
return
p[0] = p[1]
def p_typedef(self, p):
- '''typedef : TYPEDEF ID '{' block_statements_opt '}' ';' '''
+ """typedef : TYPEDEF ID '{' block_statements_opt '}' ';'"""
p[0] = Typedef(p[2], [], p[4])
def p_typedef_flist(self, p):
- '''typedef : flist TYPEDEF ID '{' block_statements_opt '}' ';' '''
+ """typedef : flist TYPEDEF ID '{' block_statements_opt '}' ';'"""
p[0] = Typedef(p[3], p[1], p[5])
def p_typedef_alias(self, p):
- '''typedef : TYPEDEF declaration '''
+ """typedef : TYPEDEF declaration"""
p[0] = Using(p[2].fieldname, [], p[2])
def p_typedef_alias_flist(self, p):
- '''typedef : flist TYPEDEF declaration '''
+ """typedef : flist TYPEDEF declaration"""
p[0] = Using(p[3].fieldname, p[1], p[3])
def p_block_statements_opt(self, p):
- '''block_statements_opt : block_statements '''
+ """block_statements_opt : block_statements"""
p[0] = p[1]
def p_block_statements(self, p):
- '''block_statements : block_statement
- | block_statements block_statement'''
+ """block_statements : block_statement
+ | block_statements block_statement"""
if len(p) == 2:
p[0] = [p[1]]
else:
p[0] = p[1] + [p[2]]
def p_block_statement(self, p):
- '''block_statement : declaration
- | option '''
+ """block_statement : declaration
+ | option"""
p[0] = p[1]
def p_enum_statements(self, p):
- '''enum_statements : enum_statement
- | enum_statements enum_statement'''
+ """enum_statements : enum_statement
+ | enum_statements enum_statement"""
if len(p) == 2:
p[0] = [p[1]]
else:
p[0] = p[1] + [p[2]]
def p_enum_statement(self, p):
- '''enum_statement : ID '=' NUM ','
- | ID ','
- | ID '[' field_options ']' ','
- | ID '=' NUM '[' field_options ']' ',' '''
+ """enum_statement : ID '=' NUM ','
+ | ID ','
+ | ID '[' field_options ']' ','
+ | ID '=' NUM '[' field_options ']' ','"""
if len(p) == 3:
- p[0] = {'id': p[1]}
+ p[0] = {"id": p[1]}
elif len(p) == 5:
- p[0] = {'id': p[1], 'value': p[3]}
+ p[0] = {"id": p[1], "value": p[3]}
elif len(p) == 6:
- p[0] = {'id': p[1], 'option': p[3]}
+ p[0] = {"id": p[1], "option": p[3]}
elif len(p) == 8:
- p[0] = {'id': p[1], 'value': p[3], 'option': p[5]}
+ p[0] = {"id": p[1], "value": p[3], "option": p[5]}
else:
- self._parse_error('ERROR', self._token_coord(p, 1))
+ self._parse_error("ERROR", self._token_coord(p, 1))
def p_field_options(self, p):
- '''field_options : field_option
- | field_options field_option'''
+ """field_options : field_option
+ | field_options field_option"""
if len(p) == 2:
p[0] = p[1]
else:
p[0] = {**p[1], **p[2]}
def p_field_option(self, p):
- '''field_option : ID
- | ID '=' assignee ','
- | ID '=' assignee
+ """field_option : ID
+ | ID '=' assignee ','
+ | ID '=' assignee
- '''
+ """
if len(p) == 2:
p[0] = {p[1]: None}
else:
p[0] = {p[1]: p[3]}
def p_variable_name(self, p):
- '''variable_name : ID
- | TYPE
- | SEVERITY
- | DESCRIPTION
- | COUNTERS
- | PATHS
- '''
+ """variable_name : ID
+ | TYPE
+ | SEVERITY
+ | DESCRIPTION
+ | COUNTERS
+ | PATHS
+ """
p[0] = p[1]
def p_declaration(self, p):
- '''declaration : type_specifier variable_name ';'
- | type_specifier variable_name '[' field_options ']' ';'
- '''
+ """declaration : type_specifier variable_name ';'
+ | type_specifier variable_name '[' field_options ']' ';'
+ """
if len(p) == 7:
p[0] = Field(p[1], p[2], p[4])
elif len(p) == 4:
p[0] = Field(p[1], p[2])
else:
- self._parse_error('ERROR', self._token_coord(p, 1))
+ self._parse_error("ERROR", self._token_coord(p, 1))
self.fields.append(p[2])
def p_declaration_array_vla(self, p):
- '''declaration : type_specifier variable_name '[' ']' ';' '''
+ """declaration : type_specifier variable_name '[' ']' ';'"""
p[0] = Array(p[1], p[2], 0, modern_vla=True)
def p_declaration_array(self, p):
- '''declaration : type_specifier variable_name '[' NUM ']' ';'
- | type_specifier variable_name '[' ID ']' ';' '''
+ """declaration : type_specifier variable_name '[' NUM ']' ';'
+ | type_specifier variable_name '[' ID ']' ';'"""
if len(p) != 7:
return self._parse_error(
- 'array: %s' % p.value,
- self._coord(lineno=p.lineno))
+ "array: %s" % p.value, self._coord(lineno=p.lineno)
+ )
# Make this error later
if type(p[4]) is int and p[4] == 0:
# XXX: Line number is wrong
- self._parse_warning('Old Style VLA: {} {}[{}];'
- .format(p[1], p[2], p[4]),
- self._token_coord(p, 1))
+ self._parse_warning(
+ "Old Style VLA: {} {}[{}];".format(p[1], p[2], p[4]),
+ self._token_coord(p, 1),
+ )
if type(p[4]) is str and p[4] not in self.fields:
# Verify that length field exists
- self._parse_error('Missing length field: {} {}[{}];'
- .format(p[1], p[2], p[4]),
- self._token_coord(p, 1))
+ self._parse_error(
+ "Missing length field: {} {}[{}];".format(p[1], p[2], p[4]),
+ self._token_coord(p, 1),
+ )
p[0] = Array(p[1], p[2], p[4])
def p_option(self, p):
- '''option : OPTION ID '=' assignee ';'
- | OPTION ID ';' '''
+ """option : OPTION ID '=' assignee ';'
+ | OPTION ID ';'"""
if len(p) == 4:
p[0] = Option(p[2])
else:
p[0] = Option(p[2], p[4])
def p_assignee(self, p):
- '''assignee : NUM
- | TRUE
- | FALSE
- | STRING_LITERAL '''
+ """assignee : NUM
+ | TRUE
+ | FALSE
+ | STRING_LITERAL"""
p[0] = p[1]
def p_type_specifier(self, p):
- '''type_specifier : U8
- | U16
- | U32
- | U64
- | I8
- | I16
- | I32
- | I64
- | F64
- | BOOL
- | STRING'''
+ """type_specifier : U8
+ | U16
+ | U32
+ | U64
+ | I8
+ | I16
+ | I32
+ | I64
+ | F64
+ | BOOL
+ | STRING"""
p[0] = p[1]
# Do a second pass later to verify that user defined types are defined
def p_typedef_specifier(self, p):
- '''type_specifier : ID '''
+ """type_specifier : ID"""
if p[1] not in global_types:
- self._parse_error('Undefined type: {}'.format(p[1]),
- self._token_coord(p, 1))
+ self._parse_error(
+ "Undefined type: {}".format(p[1]), self._token_coord(p, 1)
+ )
p[0] = p[1]
def p_union(self, p):
- '''union : UNION ID '{' block_statements_opt '}' ';' '''
+ """union : UNION ID '{' block_statements_opt '}' ';'"""
p[0] = Union(p[2], [], p[4])
def p_union_flist(self, p):
- '''union : flist UNION ID '{' block_statements_opt '}' ';' '''
+ """union : flist UNION ID '{' block_statements_opt '}' ';'"""
p[0] = Union(p[3], p[1], p[5])
# Error rule for syntax errors
def p_error(self, p):
if p:
- self._parse_error(
- 'before: %s' % p.value,
- self._coord(lineno=p.lineno))
+ self._parse_error("before: %s" % p.value, self._coord(lineno=p.lineno))
else:
- self._parse_error('At end of input', self.filename)
-
+ self._parse_error("At end of input", self.filename)
-class VPPAPI():
- def __init__(self, debug=False, filename='', logger=None, revision=None):
+class VPPAPI:
+ def __init__(self, debug=False, filename="", logger=None, revision=None):
self.lexer = lex.lex(module=VPPAPILexer(filename), debug=debug)
- self.parser = yacc.yacc(module=VPPAPIParser(filename, logger,
- revision=revision),
- write_tables=False, debug=debug)
+ self.parser = yacc.yacc(
+ module=VPPAPIParser(filename, logger, revision=revision),
+ write_tables=False,
+ debug=debug,
+ )
self.logger = logger
self.revision = revision
self.filename = filename
@@ -979,38 +982,40 @@ class VPPAPI():
def parse_filename(self, filename, debug=0):
if self.revision:
- git_show = 'git show {}:{}'.format(self.revision, filename)
- proc = Popen(git_show.split(), stdout=PIPE, encoding='utf-8')
+ git_show = "git show {}:{}".format(self.revision, filename)
+ proc = Popen(git_show.split(), stdout=PIPE, encoding="utf-8")
try:
data, errs = proc.communicate()
if proc.returncode != 0:
- print('File not found: {}:{}'
- .format(self.revision, filename), file=sys.stderr)
+ print(
+ "File not found: {}:{}".format(self.revision, filename),
+ file=sys.stderr,
+ )
sys.exit(2)
return self.parse_string(data, debug=debug)
except Exception:
sys.exit(3)
else:
try:
- with open(filename, encoding='utf-8') as fd:
+ with open(filename, encoding="utf-8") as fd:
return self.parse_fd(fd, None)
except FileNotFoundError:
- print('File not found: {}'.format(filename), file=sys.stderr)
+ print("File not found: {}".format(filename), file=sys.stderr)
sys.exit(2)
def process(self, objs):
s = {}
- s['Option'] = {}
- s['Define'] = []
- s['Service'] = []
- s['types'] = []
- s['Import'] = []
- s['Counters'] = []
- s['Paths'] = []
+ s["Option"] = {}
+ s["Define"] = []
+ s["Service"] = []
+ s["types"] = []
+ s["Import"] = []
+ s["Counters"] = []
+ s["Paths"] = []
crc = 0
for o in objs:
try:
- crc = binascii.crc32(o.crc, crc) & 0xffffffff
+ crc = binascii.crc32(o.crc, crc) & 0xFFFFFFFF
except AttributeError:
pass
@@ -1021,82 +1026,84 @@ class VPPAPI():
else:
o.process(s)
- msgs = {d.name: d for d in s['Define']}
- svcs = {s.caller: s for s in s['Service']}
- replies = {s.reply: s for s in s['Service']}
+ msgs = {d.name: d for d in s["Define"]}
+ svcs = {s.caller: s for s in s["Service"]}
+ replies = {s.reply: s for s in s["Service"]}
seen_services = {}
- s['file_crc'] = crc
+ s["file_crc"] = crc
for service in svcs:
if service not in msgs:
raise ValueError(
- 'Service definition refers to unknown message'
- ' definition: {}'.format(service))
- if svcs[service].reply != 'null' and \
- svcs[service].reply not in msgs:
- raise ValueError('Service definition refers to unknown message'
- ' definition in reply: {}'
- .format(svcs[service].reply))
+ "Service definition refers to unknown message"
+ " definition: {}".format(service)
+ )
+ if svcs[service].reply != "null" and svcs[service].reply not in msgs:
+ raise ValueError(
+ "Service definition refers to unknown message"
+ " definition in reply: {}".format(svcs[service].reply)
+ )
if service in replies:
- raise ValueError('Service definition refers to message'
- ' marked as reply: {}'.format(service))
+ raise ValueError(
+ "Service definition refers to message"
+ " marked as reply: {}".format(service)
+ )
for event in svcs[service].events:
if event not in msgs:
- raise ValueError('Service definition refers to unknown '
- 'event: {} in message: {}'
- .format(event, service))
+ raise ValueError(
+ "Service definition refers to unknown "
+ "event: {} in message: {}".format(event, service)
+ )
seen_services[event] = True
# Create services implicitly
for d in msgs:
if d in seen_services:
continue
- if d.endswith('_reply'):
+ if d.endswith("_reply"):
if d[:-6] in svcs:
continue
if d[:-6] not in msgs:
- raise ValueError('{} missing calling message'
- .format(d))
+ raise ValueError("{} missing calling message".format(d))
continue
- if d.endswith('_dump'):
+ if d.endswith("_dump"):
if d in svcs:
continue
- if d[:-5]+'_details' in msgs:
- s['Service'].append(Service(d, d[:-5]+'_details',
- stream=True))
+ if d[:-5] + "_details" in msgs:
+ s["Service"].append(Service(d, d[:-5] + "_details", stream=True))
else:
- raise ValueError('{} missing details message'
- .format(d))
+ raise ValueError("{} missing details message".format(d))
continue
- if d.endswith('_details'):
- if d[:-8]+'_get' in msgs:
- if d[:-8]+'_get' in svcs:
+ if d.endswith("_details"):
+ if d[:-8] + "_get" in msgs:
+ if d[:-8] + "_get" in svcs:
continue
- raise ValueError('{} should be in a stream service'
- .format(d[:-8]+'_get'))
- if d[:-8]+'_dump' in msgs:
+ raise ValueError(
+ "{} should be in a stream service".format(d[:-8] + "_get")
+ )
+ if d[:-8] + "_dump" in msgs:
continue
- raise ValueError('{} missing dump or get message'
- .format(d))
+ raise ValueError("{} missing dump or get message".format(d))
if d in svcs:
continue
- if d+'_reply' in msgs:
- s['Service'].append(Service(d, d+'_reply'))
+ if d + "_reply" in msgs:
+ s["Service"].append(Service(d, d + "_reply"))
else:
raise ValueError(
- '{} missing reply message ({}) or service definition'
- .format(d, d+'_reply'))
+ "{} missing reply message ({}) or service definition".format(
+ d, d + "_reply"
+ )
+ )
return s
def process_imports(self, objs, in_import, result): # -> List
for o in objs:
# Only allow the following object types from imported file
- if in_import and not isinstance(o, (Enum, Import, Typedef,
- Union, Using)):
+ if in_import and not isinstance(o, (Enum, Import, Typedef, Union, Using)):
continue
if isinstance(o, Import):
result.append(o)
@@ -1109,7 +1116,7 @@ class VPPAPI():
# Add message ids to each message.
def add_msg_id(s):
for o in s:
- o.block.insert(0, Field('u16', '_vl_msg_id'))
+ o.block.insert(0, Field("u16", "_vl_msg_id"))
return s
@@ -1129,11 +1136,11 @@ def dirlist_get():
def foldup_blocks(block, crc):
for b in block:
# Look up CRC in user defined types
- if b.fieldtype.startswith('vl_api_'):
+ if b.fieldtype.startswith("vl_api_"):
# Recursively
t = global_types[b.fieldtype]
try:
- crc = binascii.crc32(t.crc, crc) & 0xffffffff
+ crc = binascii.crc32(t.crc, crc) & 0xFFFFFFFF
crc = foldup_blocks(t.block, crc)
except AttributeError:
pass
@@ -1142,34 +1149,43 @@ def foldup_blocks(block, crc):
def foldup_crcs(s):
for f in s:
- f.crc = foldup_blocks(f.block,
- binascii.crc32(f.crc) & 0xffffffff)
+ f.crc = foldup_blocks(f.block, binascii.crc32(f.crc) & 0xFFFFFFFF)
#
# Main
#
def main():
- if sys.version_info < (3, 5,):
- log.exception('vppapigen requires a supported version of python. '
- 'Please use version 3.5 or greater. '
- 'Using %s', sys.version)
+ if sys.version_info < (
+ 3,
+ 5,
+ ):
+ log.exception(
+ "vppapigen requires a supported version of python. "
+ "Please use version 3.5 or greater. "
+ "Using %s",
+ sys.version,
+ )
return 1
- cliparser = argparse.ArgumentParser(description='VPP API generator')
- cliparser.add_argument('--pluginpath', default="")
- cliparser.add_argument('--includedir', action='append')
- cliparser.add_argument('--outputdir', action='store')
- cliparser.add_argument('--input')
- cliparser.add_argument('--output', nargs='?',
- type=argparse.FileType('w', encoding='UTF-8'),
- default=sys.stdout)
-
- cliparser.add_argument('output_module', nargs='?', default='C')
- cliparser.add_argument('--debug', action='store_true')
- cliparser.add_argument('--show-name', nargs=1)
- cliparser.add_argument('--git-revision',
- help="Git revision to use for opening files")
+ cliparser = argparse.ArgumentParser(description="VPP API generator")
+ cliparser.add_argument("--pluginpath", default="")
+ cliparser.add_argument("--includedir", action="append")
+ cliparser.add_argument("--outputdir", action="store")
+ cliparser.add_argument("--input")
+ cliparser.add_argument(
+ "--output",
+ nargs="?",
+ type=argparse.FileType("w", encoding="UTF-8"),
+ default=sys.stdout,
+ )
+
+ cliparser.add_argument("output_module", nargs="?", default="C")
+ cliparser.add_argument("--debug", action="store_true")
+ cliparser.add_argument("--show-name", nargs=1)
+ cliparser.add_argument(
+ "--git-revision", help="Git revision to use for opening files"
+ )
args = cliparser.parse_args()
dirlist_add(args.includedir)
@@ -1182,7 +1198,7 @@ def main():
elif args.input:
filename = args.input
else:
- filename = ''
+ filename = ""
if args.debug:
logging.basicConfig(stream=sys.stdout, level=logging.WARNING)
@@ -1195,36 +1211,34 @@ def main():
from importlib.machinery import SourceFileLoader
# Default path
- pluginpath = ''
+ pluginpath = ""
if not args.pluginpath:
cand = []
cand.append(os.path.dirname(os.path.realpath(__file__)))
- cand.append(os.path.dirname(os.path.realpath(__file__)) +
- '/../share/vpp/')
+ cand.append(os.path.dirname(os.path.realpath(__file__)) + "/../share/vpp/")
for c in cand:
- c += '/'
- if os.path.isfile('{}vppapigen_{}.py'
- .format(c, args.output_module.lower())):
+ c += "/"
+ if os.path.isfile(
+ "{}vppapigen_{}.py".format(c, args.output_module.lower())
+ ):
pluginpath = c
break
else:
- pluginpath = args.pluginpath + '/'
- if pluginpath == '':
- log.exception('Output plugin not found')
+ pluginpath = args.pluginpath + "/"
+ if pluginpath == "":
+ log.exception("Output plugin not found")
return 1
- module_path = '{}vppapigen_{}.py'.format(pluginpath,
- args.output_module.lower())
+ module_path = "{}vppapigen_{}.py".format(pluginpath, args.output_module.lower())
try:
- plugin = SourceFileLoader(args.output_module,
- module_path).load_module()
+ plugin = SourceFileLoader(args.output_module, module_path).load_module()
except Exception as err:
- log.exception('Error importing output plugin: %s, %s',
- module_path, err)
+ log.exception("Error importing output plugin: %s, %s", module_path, err)
return 1
- parser = VPPAPI(debug=args.debug, filename=filename, logger=log,
- revision=args.git_revision)
+ parser = VPPAPI(
+ debug=args.debug, filename=filename, logger=log, revision=args.git_revision
+ )
try:
if not args.input:
@@ -1232,7 +1246,7 @@ def main():
else:
parsed_objects = parser.parse_filename(args.input, log)
except ParseError as e:
- print('Parse error: ', e, file=sys.stderr)
+ print("Parse error: ", e, file=sys.stderr)
sys.exit(1)
# Build a list of objects. Hash of lists.
@@ -1250,32 +1264,33 @@ def main():
else:
s = parser.process(parsed_objects)
imports = parser.process_imports(parsed_objects, False, result)
- s['imported'] = parser.process(imports)
+ s["imported"] = parser.process(imports)
# Add msg_id field
- s['Define'] = add_msg_id(s['Define'])
+ s["Define"] = add_msg_id(s["Define"])
# Fold up CRCs
- foldup_crcs(s['Define'])
+ foldup_crcs(s["Define"])
#
# Debug
if args.debug:
import pprint
+
pp = pprint.PrettyPrinter(indent=4, stream=sys.stderr)
- for t in s['Define']:
+ for t in s["Define"]:
pp.pprint([t.name, t.flags, t.block])
- for t in s['types']:
+ for t in s["types"]:
pp.pprint([t.name, t.block])
result = plugin.run(args, filename, s)
if result:
print(result, file=args.output)
else:
- log.exception('Running plugin failed: %s %s', filename, result)
+ log.exception("Running plugin failed: %s %s", filename, result)
return 1
return 0
-if __name__ == '__main__':
+if __name__ == "__main__":
sys.exit(main())