mirror of
https://github.com/xemu-project/xemu.git
synced 2024-11-27 13:30:52 +00:00
8d60f37740
Numbering files according to rules causes confusion, because CUSTOM_COMMAND3.stamp from a previous build might represent completely different targets after Makefile.ninja is regenerated. As a result, the new targets are not rebuilt and compilation fails. Use the targets to build a SHA1 hash; the chances for collision are one in 2^24 even with a 12-character prefix of the hash. Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
1006 lines
33 KiB
Python
Executable File
1006 lines
33 KiB
Python
Executable File
#! /bin/sh
|
|
|
|
# Python module for parsing and processing .ninja files.
|
|
#
|
|
# Author: Paolo Bonzini
|
|
#
|
|
# Copyright (C) 2019 Red Hat, Inc.
|
|
|
|
|
|
# We don't want to put "#! @PYTHON@" as the shebang and
|
|
# make the file executable, so instead we make this a
|
|
# Python/shell polyglot. The first line below starts a
|
|
# multiline string literal for Python, while it is just
|
|
# ":" for bash. The closing of the multiline string literal
|
|
# is never parsed by bash since it exits before.
|
|
|
|
'''':
|
|
case "$0" in
|
|
/*) me=$0 ;;
|
|
*) me=$(command -v "$0") ;;
|
|
esac
|
|
python="@PYTHON@"
|
|
case $python in
|
|
@*) python=python3 ;;
|
|
esac
|
|
exec $python "$me" "$@"
|
|
exit 1
|
|
'''
|
|
|
|
|
|
from collections import namedtuple, defaultdict
|
|
import sys
|
|
import os
|
|
import re
|
|
import json
|
|
import argparse
|
|
import hashlib
|
|
import shutil
|
|
|
|
|
|
class InvalidArgumentError(Exception):
|
|
pass
|
|
|
|
# faster version of os.path.normpath: do nothing unless there is a double
|
|
# slash or a "." or ".." component. The filter does not have to be super
|
|
# precise, but it has to be fast. os.path.normpath is the hottest function
|
|
# for ninja2make without this optimization!
|
|
if os.path.sep == '/':
|
|
def normpath(path, _slow_re=re.compile('/[./]')):
|
|
return os.path.normpath(path) if _slow_re.search(path) or path[0] == '.' else path
|
|
else:
|
|
normpath = os.path.normpath
|
|
|
|
|
|
def sha1_text(text):
|
|
return hashlib.sha1(text.encode()).hexdigest()
|
|
|
|
# ---- lexer and parser ----
|
|
|
|
PATH_RE = r"[^$\s:|]+|\$[$ :]|\$[a-zA-Z0-9_-]+|\$\{[a-zA-Z0-9_.-]+\}"
|
|
|
|
SIMPLE_PATH_RE = re.compile(r"^[^$\s:|]+$")
|
|
IDENT_RE = re.compile(r"[a-zA-Z0-9_.-]+$")
|
|
STRING_RE = re.compile(r"(" + PATH_RE + r"|[\s:|])(?:\r?\n)?|.")
|
|
TOPLEVEL_RE = re.compile(r"([=:#]|\|\|?|^ +|(?:" + PATH_RE + r")+)\s*|.")
|
|
VAR_RE=re.compile(r'\$\$|\$\{([^}]*)\}')
|
|
|
|
BUILD = 1
|
|
POOL = 2
|
|
RULE = 3
|
|
DEFAULT = 4
|
|
EQUALS = 5
|
|
COLON = 6
|
|
PIPE = 7
|
|
PIPE2 = 8
|
|
IDENT = 9
|
|
INCLUDE = 10
|
|
INDENT = 11
|
|
EOL = 12
|
|
|
|
|
|
class LexerError(Exception):
|
|
pass
|
|
|
|
|
|
class ParseError(Exception):
|
|
pass
|
|
|
|
|
|
class NinjaParserEvents(object):
|
|
def __init__(self, parser):
|
|
self.parser = parser
|
|
|
|
def dollar_token(self, word, in_path=False):
|
|
return '$$' if word == '$' else word
|
|
|
|
def variable_expansion_token(self, varname):
|
|
return '${%s}' % varname
|
|
|
|
def variable(self, name, arg):
|
|
pass
|
|
|
|
def begin_file(self):
|
|
pass
|
|
|
|
def end_file(self):
|
|
pass
|
|
|
|
def end_scope(self):
|
|
pass
|
|
|
|
def begin_pool(self, name):
|
|
pass
|
|
|
|
def begin_rule(self, name):
|
|
pass
|
|
|
|
def begin_build(self, out, iout, rule, in_, iin, orderdep):
|
|
pass
|
|
|
|
def default(self, targets):
|
|
pass
|
|
|
|
|
|
class NinjaParser(object):
|
|
|
|
InputFile = namedtuple('InputFile', 'filename iter lineno')
|
|
|
|
def __init__(self, filename, input):
|
|
self.stack = []
|
|
self.top = None
|
|
self.iter = None
|
|
self.lineno = None
|
|
self.match_keyword = False
|
|
self.push(filename, input)
|
|
|
|
def file_changed(self):
|
|
self.iter = self.top.iter
|
|
self.lineno = self.top.lineno
|
|
if self.top.filename is not None:
|
|
os.chdir(os.path.dirname(self.top.filename) or '.')
|
|
|
|
def push(self, filename, input):
|
|
if self.top:
|
|
self.top.lineno = self.lineno
|
|
self.top.iter = self.iter
|
|
self.stack.append(self.top)
|
|
self.top = self.InputFile(filename=filename or 'stdin',
|
|
iter=self._tokens(input), lineno=0)
|
|
self.file_changed()
|
|
|
|
def pop(self):
|
|
if len(self.stack):
|
|
self.top = self.stack[-1]
|
|
self.stack.pop()
|
|
self.file_changed()
|
|
else:
|
|
self.top = self.iter = None
|
|
|
|
def next_line(self, input):
|
|
line = next(input).rstrip()
|
|
self.lineno += 1
|
|
while len(line) and line[-1] == '$':
|
|
line = line[0:-1] + next(input).strip()
|
|
self.lineno += 1
|
|
return line
|
|
|
|
def print_token(self, tok):
|
|
if tok == EOL:
|
|
return "end of line"
|
|
if tok == BUILD:
|
|
return '"build"'
|
|
if tok == POOL:
|
|
return '"pool"'
|
|
if tok == RULE:
|
|
return '"rule"'
|
|
if tok == DEFAULT:
|
|
return '"default"'
|
|
if tok == EQUALS:
|
|
return '"="'
|
|
if tok == COLON:
|
|
return '":"'
|
|
if tok == PIPE:
|
|
return '"|"'
|
|
if tok == PIPE2:
|
|
return '"||"'
|
|
if tok == INCLUDE:
|
|
return '"include"'
|
|
if tok == IDENT:
|
|
return 'identifier'
|
|
return '"%s"' % tok
|
|
|
|
def error(self, msg):
|
|
raise LexerError("%s:%d: %s" % (self.stack[-1].filename, self.lineno, msg))
|
|
|
|
def parse_error(self, msg):
|
|
raise ParseError("%s:%d: %s" % (self.stack[-1].filename, self.lineno, msg))
|
|
|
|
def expected(self, expected, tok):
|
|
msg = "found %s, expected " % (self.print_token(tok), )
|
|
for i, exp_tok in enumerate(expected):
|
|
if i > 0:
|
|
msg = msg + (' or ' if i == len(expected) - 1 else ', ')
|
|
msg = msg + self.print_token(exp_tok)
|
|
self.parse_error(msg)
|
|
|
|
def _variable_tokens(self, value):
|
|
for m in STRING_RE.finditer(value):
|
|
match = m.group(1)
|
|
if not match:
|
|
self.error("unexpected '%s'" % (m.group(0), ))
|
|
yield match
|
|
|
|
def _tokens(self, input):
|
|
while True:
|
|
try:
|
|
line = self.next_line(input)
|
|
except StopIteration:
|
|
return
|
|
for m in TOPLEVEL_RE.finditer(line):
|
|
match = m.group(1)
|
|
if not match:
|
|
self.error("unexpected '%s'" % (m.group(0), ))
|
|
if match == ':':
|
|
yield COLON
|
|
continue
|
|
if match == '|':
|
|
yield PIPE
|
|
continue
|
|
if match == '||':
|
|
yield PIPE2
|
|
continue
|
|
if match[0] == ' ':
|
|
yield INDENT
|
|
continue
|
|
if match[0] == '=':
|
|
yield EQUALS
|
|
value = line[m.start() + 1:].lstrip()
|
|
yield from self._variable_tokens(value)
|
|
break
|
|
if match[0] == '#':
|
|
break
|
|
|
|
# identifier
|
|
if self.match_keyword:
|
|
if match == 'build':
|
|
yield BUILD
|
|
continue
|
|
if match == 'pool':
|
|
yield POOL
|
|
continue
|
|
if match == 'rule':
|
|
yield RULE
|
|
continue
|
|
if match == 'default':
|
|
yield DEFAULT
|
|
continue
|
|
if match == 'include':
|
|
filename = line[m.start() + 8:].strip()
|
|
self.push(filename, open(filename, 'r'))
|
|
break
|
|
if match == 'subninja':
|
|
self.error('subninja is not supported')
|
|
yield match
|
|
yield EOL
|
|
|
|
def parse(self, events):
|
|
global_var = True
|
|
|
|
def look_for(*expected):
|
|
# The last token in the token stream is always EOL. This
|
|
# is exploited to avoid catching StopIteration everywhere.
|
|
tok = next(self.iter)
|
|
if tok not in expected:
|
|
self.expected(expected, tok)
|
|
return tok
|
|
|
|
def look_for_ident(*expected):
|
|
tok = next(self.iter)
|
|
if isinstance(tok, str):
|
|
if not IDENT_RE.match(tok):
|
|
self.parse_error('variable expansion not allowed')
|
|
elif tok not in expected:
|
|
self.expected(expected + (IDENT,), tok)
|
|
return tok
|
|
|
|
def parse_assignment_rhs(gen, expected, in_path):
|
|
tokens = []
|
|
for tok in gen:
|
|
if not isinstance(tok, str):
|
|
if tok in expected:
|
|
break
|
|
self.expected(expected + (IDENT,), tok)
|
|
if tok[0] != '$':
|
|
tokens.append(tok)
|
|
elif tok == '$ ' or tok == '$$' or tok == '$:':
|
|
tokens.append(events.dollar_token(tok[1], in_path))
|
|
else:
|
|
var = tok[2:-1] if tok[1] == '{' else tok[1:]
|
|
tokens.append(events.variable_expansion_token(var))
|
|
else:
|
|
# gen must have raised StopIteration
|
|
tok = None
|
|
|
|
if tokens:
|
|
# Fast path avoiding str.join()
|
|
value = tokens[0] if len(tokens) == 1 else ''.join(tokens)
|
|
else:
|
|
value = None
|
|
return value, tok
|
|
|
|
def look_for_path(*expected):
|
|
# paths in build rules are parsed one space-separated token
|
|
# at a time and expanded
|
|
token = next(self.iter)
|
|
if not isinstance(token, str):
|
|
return None, token
|
|
# Fast path if there are no dollar and variable expansion
|
|
if SIMPLE_PATH_RE.match(token):
|
|
return token, None
|
|
gen = self._variable_tokens(token)
|
|
return parse_assignment_rhs(gen, expected, True)
|
|
|
|
def parse_assignment(tok):
|
|
name = tok
|
|
assert isinstance(name, str)
|
|
look_for(EQUALS)
|
|
value, tok = parse_assignment_rhs(self.iter, (EOL,), False)
|
|
assert tok == EOL
|
|
events.variable(name, value)
|
|
|
|
def parse_build():
|
|
# parse outputs
|
|
out = []
|
|
iout = []
|
|
while True:
|
|
value, tok = look_for_path(COLON, PIPE)
|
|
if value is None:
|
|
break
|
|
out.append(value)
|
|
if tok == PIPE:
|
|
while True:
|
|
value, tok = look_for_path(COLON)
|
|
if value is None:
|
|
break
|
|
iout.append(value)
|
|
|
|
# parse rule
|
|
assert tok == COLON
|
|
rule = look_for_ident()
|
|
|
|
# parse inputs and dependencies
|
|
in_ = []
|
|
iin = []
|
|
orderdep = []
|
|
while True:
|
|
value, tok = look_for_path(PIPE, PIPE2, EOL)
|
|
if value is None:
|
|
break
|
|
in_.append(value)
|
|
if tok == PIPE:
|
|
while True:
|
|
value, tok = look_for_path(PIPE2, EOL)
|
|
if value is None:
|
|
break
|
|
iin.append(value)
|
|
if tok == PIPE2:
|
|
while True:
|
|
value, tok = look_for_path(EOL)
|
|
if value is None:
|
|
break
|
|
orderdep.append(value)
|
|
assert tok == EOL
|
|
events.begin_build(out, iout, rule, in_, iin, orderdep)
|
|
nonlocal global_var
|
|
global_var = False
|
|
|
|
def parse_pool():
|
|
# pool declarations are ignored. Just gobble all the variables
|
|
ident = look_for_ident()
|
|
look_for(EOL)
|
|
events.begin_pool(ident)
|
|
nonlocal global_var
|
|
global_var = False
|
|
|
|
def parse_rule():
|
|
ident = look_for_ident()
|
|
look_for(EOL)
|
|
events.begin_rule(ident)
|
|
nonlocal global_var
|
|
global_var = False
|
|
|
|
def parse_default():
|
|
idents = []
|
|
while True:
|
|
ident = look_for_ident(EOL)
|
|
if ident == EOL:
|
|
break
|
|
idents.append(ident)
|
|
events.default(idents)
|
|
|
|
def parse_declaration(tok):
|
|
if tok == EOL:
|
|
return
|
|
|
|
nonlocal global_var
|
|
if tok == INDENT:
|
|
if global_var:
|
|
self.parse_error('indented line outside rule or edge')
|
|
tok = look_for_ident(EOL)
|
|
if tok == EOL:
|
|
return
|
|
parse_assignment(tok)
|
|
return
|
|
|
|
if not global_var:
|
|
events.end_scope()
|
|
global_var = True
|
|
if tok == POOL:
|
|
parse_pool()
|
|
elif tok == BUILD:
|
|
parse_build()
|
|
elif tok == RULE:
|
|
parse_rule()
|
|
elif tok == DEFAULT:
|
|
parse_default()
|
|
elif isinstance(tok, str):
|
|
parse_assignment(tok)
|
|
else:
|
|
self.expected((POOL, BUILD, RULE, INCLUDE, DEFAULT, IDENT), tok)
|
|
|
|
events.begin_file()
|
|
while self.iter:
|
|
try:
|
|
self.match_keyword = True
|
|
token = next(self.iter)
|
|
self.match_keyword = False
|
|
parse_declaration(token)
|
|
except StopIteration:
|
|
self.pop()
|
|
events.end_file()
|
|
|
|
|
|
# ---- variable handling ----
|
|
|
|
def expand(x, rule_vars=None, build_vars=None, global_vars=None):
|
|
if x is None:
|
|
return None
|
|
changed = True
|
|
have_dollar_replacement = False
|
|
while changed:
|
|
changed = False
|
|
matches = list(VAR_RE.finditer(x))
|
|
if not matches:
|
|
break
|
|
|
|
# Reverse the match so that expanding later matches does not
|
|
# invalidate m.start()/m.end() for earlier ones. Do not reduce $$ to $
|
|
# until all variables are dealt with.
|
|
for m in reversed(matches):
|
|
name = m.group(1)
|
|
if not name:
|
|
have_dollar_replacement = True
|
|
continue
|
|
changed = True
|
|
if build_vars and name in build_vars:
|
|
value = build_vars[name]
|
|
elif rule_vars and name in rule_vars:
|
|
value = rule_vars[name]
|
|
elif name in global_vars:
|
|
value = global_vars[name]
|
|
else:
|
|
value = ''
|
|
x = x[:m.start()] + value + x[m.end():]
|
|
return x.replace('$$', '$') if have_dollar_replacement else x
|
|
|
|
|
|
class Scope(object):
|
|
def __init__(self, events):
|
|
self.events = events
|
|
|
|
def on_left_scope(self):
|
|
pass
|
|
|
|
def on_variable(self, key, value):
|
|
pass
|
|
|
|
|
|
class BuildScope(Scope):
|
|
def __init__(self, events, out, iout, rule, in_, iin, orderdep, rule_vars):
|
|
super().__init__(events)
|
|
self.rule = rule
|
|
self.out = [events.expand_and_normalize(x) for x in out]
|
|
self.in_ = [events.expand_and_normalize(x) for x in in_]
|
|
self.iin = [events.expand_and_normalize(x) for x in iin]
|
|
self.orderdep = [events.expand_and_normalize(x) for x in orderdep]
|
|
self.iout = [events.expand_and_normalize(x) for x in iout]
|
|
self.rule_vars = rule_vars
|
|
self.build_vars = dict()
|
|
self._define_variable('out', ' '.join(self.out))
|
|
self._define_variable('in', ' '.join(self.in_))
|
|
|
|
def expand(self, x):
|
|
return self.events.expand(x, self.rule_vars, self.build_vars)
|
|
|
|
def on_left_scope(self):
|
|
self.events.variable('out', self.build_vars['out'])
|
|
self.events.variable('in', self.build_vars['in'])
|
|
self.events.end_build(self, self.out, self.iout, self.rule, self.in_,
|
|
self.iin, self.orderdep)
|
|
|
|
def _define_variable(self, key, value):
|
|
# The value has been expanded already, quote it for further
|
|
# expansion from rule variables
|
|
value = value.replace('$', '$$')
|
|
self.build_vars[key] = value
|
|
|
|
def on_variable(self, key, value):
|
|
# in and out are at the top of the lookup order and cannot
|
|
# be overridden. Also, unlike what the manual says, build
|
|
# variables only lookup global variables. They never lookup
|
|
# rule variables, earlier build variables, or in/out.
|
|
if key not in ('in', 'in_newline', 'out'):
|
|
self._define_variable(key, self.events.expand(value))
|
|
|
|
|
|
class RuleScope(Scope):
|
|
def __init__(self, events, name, vars_dict):
|
|
super().__init__(events)
|
|
self.name = name
|
|
self.vars_dict = vars_dict
|
|
self.generator = False
|
|
|
|
def on_left_scope(self):
|
|
self.events.end_rule(self, self.name)
|
|
|
|
def on_variable(self, key, value):
|
|
self.vars_dict[key] = value
|
|
if key == 'generator':
|
|
self.generator = True
|
|
|
|
|
|
class NinjaParserEventsWithVars(NinjaParserEvents):
|
|
def __init__(self, parser):
|
|
super().__init__(parser)
|
|
self.rule_vars = defaultdict(lambda: dict())
|
|
self.global_vars = dict()
|
|
self.scope = None
|
|
|
|
def variable(self, name, value):
|
|
if self.scope:
|
|
self.scope.on_variable(name, value)
|
|
else:
|
|
self.global_vars[name] = self.expand(value)
|
|
|
|
def begin_build(self, out, iout, rule, in_, iin, orderdep):
|
|
if rule != 'phony' and rule not in self.rule_vars:
|
|
self.parser.parse_error("undefined rule '%s'" % rule)
|
|
|
|
self.scope = BuildScope(self, out, iout, rule, in_, iin, orderdep, self.rule_vars[rule])
|
|
|
|
def begin_pool(self, name):
|
|
# pool declarations are ignored. Just gobble all the variables
|
|
self.scope = Scope(self)
|
|
|
|
def begin_rule(self, name):
|
|
if name in self.rule_vars:
|
|
self.parser.parse_error("duplicate rule '%s'" % name)
|
|
self.scope = RuleScope(self, name, self.rule_vars[name])
|
|
|
|
def end_scope(self):
|
|
self.scope.on_left_scope()
|
|
self.scope = None
|
|
|
|
# utility functions:
|
|
|
|
def expand(self, x, rule_vars=None, build_vars=None):
|
|
return expand(x, rule_vars, build_vars, self.global_vars)
|
|
|
|
def expand_and_normalize(self, x):
|
|
return normpath(self.expand(x))
|
|
|
|
# extra events not present in the superclass:
|
|
|
|
def end_build(self, scope, out, iout, rule, in_, iin, orderdep):
|
|
pass
|
|
|
|
def end_rule(self, scope, name):
|
|
pass
|
|
|
|
|
|
# ---- test client that just prints back whatever it parsed ----
|
|
|
|
class Writer(NinjaParserEvents):
|
|
ARGS = argparse.ArgumentParser(description='Rewrite input build.ninja to stdout.')
|
|
|
|
def __init__(self, output, parser, args):
|
|
super().__init__(parser)
|
|
self.output = output
|
|
self.indent = ''
|
|
self.had_vars = False
|
|
|
|
def dollar_token(self, word, in_path=False):
|
|
return '$' + word
|
|
|
|
def print(self, *args, **kwargs):
|
|
if len(args):
|
|
self.output.write(self.indent)
|
|
print(*args, **kwargs, file=self.output)
|
|
|
|
def variable(self, name, value):
|
|
self.print('%s = %s' % (name, value))
|
|
self.had_vars = True
|
|
|
|
def begin_scope(self):
|
|
self.indent = ' '
|
|
self.had_vars = False
|
|
|
|
def end_scope(self):
|
|
if self.had_vars:
|
|
self.print()
|
|
self.indent = ''
|
|
self.had_vars = False
|
|
|
|
def begin_pool(self, name):
|
|
self.print('pool %s' % name)
|
|
self.begin_scope()
|
|
|
|
def begin_rule(self, name):
|
|
self.print('rule %s' % name)
|
|
self.begin_scope()
|
|
|
|
def begin_build(self, outputs, implicit_outputs, rule, inputs, implicit, order_only):
|
|
all_outputs = list(outputs)
|
|
all_inputs = list(inputs)
|
|
|
|
if implicit:
|
|
all_inputs.append('|')
|
|
all_inputs.extend(implicit)
|
|
if order_only:
|
|
all_inputs.append('||')
|
|
all_inputs.extend(order_only)
|
|
if implicit_outputs:
|
|
all_outputs.append('|')
|
|
all_outputs.extend(implicit_outputs)
|
|
|
|
self.print('build %s: %s' % (' '.join(all_outputs),
|
|
' '.join([rule] + all_inputs)))
|
|
self.begin_scope()
|
|
|
|
def default(self, targets):
|
|
self.print('default %s' % ' '.join(targets))
|
|
|
|
|
|
# ---- emit compile_commands.json ----
|
|
|
|
class Compdb(NinjaParserEventsWithVars):
|
|
ARGS = argparse.ArgumentParser(description='Emit compile_commands.json.')
|
|
ARGS.add_argument('rules', nargs='*',
|
|
help='The ninja rules to emit compilation commands for.')
|
|
|
|
def __init__(self, output, parser, args):
|
|
super().__init__(parser)
|
|
self.output = output
|
|
self.rules = args.rules
|
|
self.sep = ''
|
|
|
|
def begin_file(self):
|
|
self.output.write('[')
|
|
self.directory = os.getcwd()
|
|
|
|
def print_entry(self, **entry):
|
|
entry['directory'] = self.directory
|
|
self.output.write(self.sep + json.dumps(entry))
|
|
self.sep = ',\n'
|
|
|
|
def begin_build(self, out, iout, rule, in_, iin, orderdep):
|
|
if in_ and rule in self.rules:
|
|
super().begin_build(out, iout, rule, in_, iin, orderdep)
|
|
else:
|
|
self.scope = Scope(self)
|
|
|
|
def end_build(self, scope, out, iout, rule, in_, iin, orderdep):
|
|
self.print_entry(command=scope.expand('${command}'), file=in_[0])
|
|
|
|
def end_file(self):
|
|
self.output.write(']\n')
|
|
|
|
|
|
# ---- clean output files ----
|
|
|
|
class Clean(NinjaParserEventsWithVars):
|
|
ARGS = argparse.ArgumentParser(description='Remove output build files.')
|
|
ARGS.add_argument('-g', dest='generator', action='store_true',
|
|
help='clean generated files too')
|
|
|
|
def __init__(self, output, parser, args):
|
|
super().__init__(parser)
|
|
self.dry_run = args.dry_run
|
|
self.verbose = args.verbose or args.dry_run
|
|
self.generator = args.generator
|
|
|
|
def begin_file(self):
|
|
print('Cleaning... ', end=(None if self.verbose else ''), flush=True)
|
|
self.cnt = 0
|
|
|
|
def end_file(self):
|
|
print('%d files' % self.cnt)
|
|
|
|
def do_clean(self, *files):
|
|
for f in files:
|
|
if self.dry_run:
|
|
if os.path.exists(f):
|
|
self.cnt += 1
|
|
print('Would remove ' + f)
|
|
continue
|
|
else:
|
|
try:
|
|
if os.path.isdir(f):
|
|
shutil.rmtree(f)
|
|
else:
|
|
os.unlink(f)
|
|
self.cnt += 1
|
|
if self.verbose:
|
|
print('Removed ' + f)
|
|
except FileNotFoundError:
|
|
pass
|
|
|
|
def end_build(self, scope, out, iout, rule, in_, iin, orderdep):
|
|
if rule == 'phony':
|
|
return
|
|
if self.generator:
|
|
rspfile = scope.expand('${rspfile}')
|
|
if rspfile:
|
|
self.do_clean(rspfile)
|
|
if self.generator or not scope.expand('${generator}'):
|
|
self.do_clean(*out, *iout)
|
|
depfile = scope.expand('${depfile}')
|
|
if depfile:
|
|
self.do_clean(depfile)
|
|
|
|
|
|
# ---- convert build.ninja to makefile ----
|
|
|
|
class Ninja2Make(NinjaParserEventsWithVars):
|
|
ARGS = argparse.ArgumentParser(description='Convert build.ninja to a Makefile.')
|
|
ARGS.add_argument('--clean', dest='emit_clean', action='store_true',
|
|
help='Emit clean/distclean rules.')
|
|
ARGS.add_argument('--doublecolon', action='store_true',
|
|
help='Emit double-colon rules for phony targets.')
|
|
ARGS.add_argument('--omit', metavar='TARGET', nargs='+',
|
|
help='Targets to omit.')
|
|
|
|
def __init__(self, output, parser, args):
|
|
super().__init__(parser)
|
|
self.output = output
|
|
|
|
self.emit_clean = args.emit_clean
|
|
self.doublecolon = args.doublecolon
|
|
self.omit = set(args.omit)
|
|
|
|
if self.emit_clean:
|
|
self.omit.update(['clean', 'distclean'])
|
|
|
|
# Lists of targets are kept in memory and emitted only at the
|
|
# end because appending is really inefficient in GNU make.
|
|
# We only do it when it's O(#rules) or O(#variables), but
|
|
# never when it could be O(#targets).
|
|
self.depfiles = list()
|
|
self.rspfiles = list()
|
|
self.build_vars = defaultdict(lambda: dict())
|
|
self.rule_targets = defaultdict(lambda: list())
|
|
self.stamp_targets = defaultdict(lambda: list())
|
|
self.all_outs = set()
|
|
self.all_ins = set()
|
|
self.all_phony = set()
|
|
self.seen_default = False
|
|
|
|
def print(self, *args, **kwargs):
|
|
print(*args, **kwargs, file=self.output)
|
|
|
|
def dollar_token(self, word, in_path=False):
|
|
if in_path and word == ' ':
|
|
self.parser.parse_error('Make does not support spaces in filenames')
|
|
return '$$' if word == '$' else word
|
|
|
|
def print_phony(self, outs, ins):
|
|
targets = ' '.join(outs).replace('$', '$$')
|
|
deps = ' '.join(ins).replace('$', '$$')
|
|
deps = deps.strip()
|
|
if self.doublecolon:
|
|
self.print(targets + '::' + (' ' if deps else '') + deps + ';@:')
|
|
else:
|
|
self.print(targets + ':' + (' ' if deps else '') + deps)
|
|
self.all_phony.update(outs)
|
|
|
|
def begin_file(self):
|
|
self.print(r'# This is an automatically generated file, and it shows.')
|
|
self.print(r'ninja-default:')
|
|
self.print(r'.PHONY: ninja-default ninja-clean ninja-distclean')
|
|
if self.emit_clean:
|
|
self.print(r'ninja-clean:: ninja-clean-start; $(if $V,,@)rm -f ${ninja-depfiles}')
|
|
self.print(r'ninja-clean-start:; $(if $V,,@echo Cleaning...)')
|
|
self.print(r'ninja-distclean:: clean; $(if $V,,@)rm -f ${ninja-rspfiles}')
|
|
self.print(r'.PHONY: ninja-clean-start')
|
|
self.print_phony(['clean'], ['ninja-clean'])
|
|
self.print_phony(['distclean'], ['ninja-distclean'])
|
|
self.print(r'vpath')
|
|
self.print(r'NULL :=')
|
|
self.print(r'SPACE := ${NULL} #')
|
|
self.print(r'MAKEFLAGS += -rR')
|
|
self.print(r'define NEWLINE')
|
|
self.print(r'')
|
|
self.print(r'endef')
|
|
self.print(r'.var.in_newline = $(subst $(SPACE),$(NEWLINE),${.var.in})')
|
|
self.print(r"ninja-command = $(if $V,,$(if ${.var.description},@printf '%s\n' '$(subst ','\'',${.var.description})' && ))${.var.command}")
|
|
self.print(r"ninja-command-restat = $(if $V,,$(if ${.var.description},@printf '%s\n' '$(subst ','\'',${.var.description})' && ))${.var.command} && if test -e $(firstword ${.var.out}); then printf '%s\n' ${.var.out} > $@; fi")
|
|
|
|
def end_file(self):
|
|
def natural_sort_key(s, _nsre=re.compile('([0-9]+)')):
|
|
return [int(text) if text.isdigit() else text.lower()
|
|
for text in _nsre.split(s)]
|
|
|
|
self.print()
|
|
self.print('ninja-outputdirs :=')
|
|
for rule in self.rule_vars:
|
|
if rule == 'phony':
|
|
continue
|
|
self.print('ninja-targets-%s := %s' % (rule, ' '.join(self.rule_targets[rule])))
|
|
self.print('ninja-stamp-%s := %s' % (rule, ' '.join(self.stamp_targets[rule])))
|
|
self.print('ninja-outputdirs += $(sort $(dir ${ninja-targets-%s}))' % rule)
|
|
self.print()
|
|
self.print('dummy := $(shell mkdir -p . $(sort $(ninja-outputdirs)))')
|
|
self.print('ninja-depfiles :=' + ' '.join(self.depfiles))
|
|
self.print('ninja-rspfiles :=' + ' '.join(self.rspfiles))
|
|
self.print('-include ${ninja-depfiles}')
|
|
self.print()
|
|
for targets in self.build_vars:
|
|
for name, value in self.build_vars[targets].items():
|
|
self.print('%s: private .var.%s := %s' %
|
|
(targets, name, value.replace('$', '$$')))
|
|
self.print()
|
|
if not self.seen_default:
|
|
default_targets = sorted(self.all_outs - self.all_ins, key=natural_sort_key)
|
|
self.print('ninja-default: ' + ' '.join(default_targets))
|
|
|
|
# This is a hack... Meson declares input meson.build files as
|
|
# phony, because Ninja does not have an equivalent of Make's
|
|
# "path/to/file:" declaration that ignores "path/to/file" even
|
|
# if it is absent. However, Makefile.ninja wants to depend on
|
|
# build.ninja, which in turn depends on these phony targets which
|
|
# would cause Makefile.ninja to be rebuilt in a loop.
|
|
phony_targets = sorted(self.all_phony - self.all_ins, key=natural_sort_key)
|
|
self.print('.PHONY: ' + ' '.join(phony_targets))
|
|
|
|
def variable(self, name, value):
|
|
super().variable(name, value)
|
|
if self.scope is None:
|
|
self.global_vars[name] = self.expand(value)
|
|
self.print('.var.%s := %s' % (name, self.global_vars[name]))
|
|
|
|
def begin_build(self, out, iout, rule, in_, iin, orderdep):
|
|
if any(x in self.omit for x in out):
|
|
self.scope = Scope(self)
|
|
return
|
|
|
|
super().begin_build(out, iout, rule, in_, iin, orderdep)
|
|
self.current_targets = ' '.join(self.scope.out + self.scope.iout).replace('$', '$$')
|
|
|
|
def end_build(self, scope, out, iout, rule, in_, iin, orderdep):
|
|
self.rule_targets[rule] += self.scope.out
|
|
self.rule_targets[rule] += self.scope.iout
|
|
|
|
self.all_outs.update(self.scope.iout)
|
|
self.all_outs.update(self.scope.out)
|
|
self.all_ins.update(self.scope.in_)
|
|
self.all_ins.update(self.scope.iin)
|
|
|
|
targets = self.current_targets
|
|
self.current_targets = None
|
|
if rule == 'phony':
|
|
# Phony rules treat order-only dependencies as normal deps
|
|
self.print_phony(out + iout, in_ + iin + orderdep)
|
|
return
|
|
|
|
inputs = ' '.join(in_ + iin).replace('$', '$$')
|
|
orderonly = ' '.join(orderdep).replace('$', '$$')
|
|
|
|
rspfile = scope.expand('${rspfile}')
|
|
if rspfile:
|
|
rspfile_content = scope.expand('${rspfile_content}')
|
|
with open(rspfile, 'w') as f:
|
|
f.write(rspfile_content)
|
|
inputs += ' ' + rspfile
|
|
self.rspfiles.append(rspfile)
|
|
|
|
restat = 'restat' in self.scope.build_vars or 'restat' in self.rule_vars[rule]
|
|
depfile = scope.expand('${depfile}')
|
|
build_vars = {
|
|
'command': scope.expand('${command}'),
|
|
'description': scope.expand('${description}'),
|
|
'out': scope.expand('${out}')
|
|
}
|
|
|
|
if restat and not depfile:
|
|
if len(out) == 1:
|
|
stamp = out[0] + '.stamp'
|
|
else:
|
|
stamp = '%s@%s.stamp' % (rule, sha1_text(targets)[0:11])
|
|
self.print('%s: %s; @:' % (targets, stamp))
|
|
self.print('%s: %s | %s; ${ninja-command-restat}' % (stamp, inputs, orderonly))
|
|
self.rule_targets[rule].append(stamp)
|
|
self.stamp_targets[rule].append(stamp)
|
|
self.build_vars[stamp] = build_vars
|
|
else:
|
|
self.print('%s: %s | %s; ${ninja-command}' % (targets, inputs, orderonly))
|
|
self.build_vars[targets] = build_vars
|
|
if depfile:
|
|
self.depfiles.append(depfile)
|
|
|
|
def end_rule(self, scope, name):
|
|
# Note that the generator pseudo-variable could also be attached
|
|
# to a build block rather than a rule. This is not handled here
|
|
# in order to reduce the number of "rm" invocations. However,
|
|
# "ninjatool.py -t clean" does that correctly.
|
|
target = 'distclean' if scope.generator else 'clean'
|
|
self.print('ninja-%s:: ; $(if $V,,@)rm -f ${ninja-stamp-%s}' % (target, name))
|
|
if self.emit_clean:
|
|
self.print('ninja-%s:: ; $(if $V,,@)rm -rf ${ninja-targets-%s}' % (target, name))
|
|
|
|
def default(self, targets):
|
|
self.print("ninja-default: " + ' '.join(targets))
|
|
self.seen_default = True
|
|
|
|
|
|
# ---- command line parsing ----
|
|
|
|
# we cannot use subparsers because tools are chosen through the "-t"
|
|
# option.
|
|
|
|
class ToolAction(argparse.Action):
|
|
def __init__(self, option_strings, dest, choices, metavar='TOOL', nargs=None, **kwargs):
|
|
if nargs is not None:
|
|
raise ValueError("nargs not allowed")
|
|
super().__init__(option_strings, dest, required=True, choices=choices,
|
|
metavar=metavar, **kwargs)
|
|
|
|
def __call__(self, parser, namespace, value, option_string):
|
|
tool = self.choices[value]
|
|
setattr(namespace, self.dest, tool)
|
|
tool.ARGS.prog = '%s %s %s' % (parser.prog, option_string, value)
|
|
|
|
|
|
class ToolHelpAction(argparse.Action):
|
|
def __init__(self, option_strings, dest, nargs=None, **kwargs):
|
|
if nargs is not None:
|
|
raise ValueError("nargs not allowed")
|
|
super().__init__(option_strings, dest, nargs=0, **kwargs)
|
|
|
|
def __call__(self, parser, namespace, values, option_string=None):
|
|
if namespace.tool:
|
|
namespace.tool.ARGS.print_help()
|
|
else:
|
|
parser.print_help()
|
|
parser.exit()
|
|
|
|
|
|
tools = {
|
|
'test': Writer,
|
|
'ninja2make': Ninja2Make,
|
|
'compdb': Compdb,
|
|
'clean': Clean,
|
|
}
|
|
|
|
parser = argparse.ArgumentParser(description='Process and transform build.ninja files.',
|
|
add_help=False)
|
|
parser.add_argument('-C', metavar='DIR', dest='dir', default='.',
|
|
help='change to DIR before doing anything else')
|
|
parser.add_argument('-f', metavar='FILE', dest='file', default='build.ninja',
|
|
help='specify input build file [default=build.ninja]')
|
|
parser.add_argument('-n', dest='dry_run', action='store_true',
|
|
help='do not actually do anything')
|
|
parser.add_argument('-v', dest='verbose', action='store_true',
|
|
help='be more verbose')
|
|
|
|
parser.add_argument('-t', dest='tool', choices=tools, action=ToolAction,
|
|
help='choose the tool to run')
|
|
parser.add_argument('-h', '--help', action=ToolHelpAction,
|
|
help='show this help message and exit')
|
|
|
|
if len(sys.argv) >= 2 and sys.argv[1] == '--version':
|
|
print('1.8')
|
|
sys.exit(0)
|
|
|
|
args, tool_args = parser.parse_known_args()
|
|
args.tool.ARGS.parse_args(tool_args, args)
|
|
|
|
os.chdir(args.dir)
|
|
with open(args.file, 'r') as f:
|
|
parser = NinjaParser(args.file, f)
|
|
try:
|
|
events = args.tool(sys.stdout, parser, args)
|
|
except InvalidArgumentError as e:
|
|
parser.error(str(e))
|
|
parser.parse(events)
|