From 0898a007bac173f0a61564914d54d0496a851235 Mon Sep 17 00:00:00 2001 From: Neels Hofmeyr Date: Wed, 16 Nov 2016 14:36:29 +0100 Subject: contrib: add fsm-to-dot.py to draw osmo_fsm dotty graphs Add a first version of a python script that tries to analyze .c source files to draw graphs of osmo_fsm implementations. So far it uses quick-and-dirty regexes. Change-Id: I155f57a608d600f59aedfd27ef66eb9772c124e7 --- contrib/fsm-to-dot.py | 710 ++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 710 insertions(+) create mode 100755 contrib/fsm-to-dot.py (limited to 'contrib') diff --git a/contrib/fsm-to-dot.py b/contrib/fsm-to-dot.py new file mode 100755 index 00000000..3549d1e5 --- /dev/null +++ b/contrib/fsm-to-dot.py @@ -0,0 +1,710 @@ +#!/usr/bin/env python + +__doc__ = ''' +fsm-to-dot: convert FSM definitons to graph images + +Usage: + ./fsm-to-dot.py ~/openbsc/openbsc/src/libvlr/*.c + for f in *.dot ; do dot -Tpng $f > $f.png; done + # dot comes from 'apt-get install graphviz' + +Looks for osmo_fsm finite state machine definitions and madly parses .c files +to draw graphs of them. This uses wild regexes that rely on coding style etc.. +No proper C parsing is done here (pycparser sucked, unfortunately). +''' + +import sys, re + +def err(msg): + sys.stderr.write(msg + '\n') + +class listdict(object): + def __getattr__(ld, name): + if name == 'add': + return ld.__getattribute__(name) + return ld.__dict__.__getattribute__(name) + + def _have(ld, name): + l = ld.__dict__.get(name) + if not l: + l = [] + ld.__dict__[name] = l + return l + + def add(ld, name, item): + l = ld._have(name) + l.append(item) + return ld + + def add_dict(ld, d): + for k,v in d.items(): + ld.add(k, v) + + def __setitem__(ld, name, val): + return ld.__dict__.__setitem__(name, val) + + def __getitem__(ld, name): + return ld.__dict__.__getitem__(name) + + def __str__(ld): + return ld.__dict__.__str__() + + def __repr__(ld): + return ld.__dict__.__repr__() + + def update(ld, other_ld): + for name, items in other_ld.items(): + ld.extend(name, items) + return ld + + def extend(ld, name, vals): + l = ld._have(name) + l.extend(vals) + return ld + +re_state_start = re.compile(r'\[([A-Z_][A-Z_0-9]*)\]') +re_event = re.compile(r'S\(([A-Z_][A-Z_0-9]*)\)') +re_action = re.compile(r'.action *= *([a-z_][a-z_0-9]*)') + +def state_starts(line): + m = re_state_start.search(line) + if m: + return m.group(1) + return None + +def in_event_starts(line): + return line.find('in_event_mask') >= 0 + +def out_state_starts(line): + return line.find('out_state_mask') >= 0 + +def states_or_events(line): + return re_event.findall(line) + +def parse_action(line): + a = re_action.findall(line) + if a: + return a[0] + return None + +def _common_prefix(a, b): + for l in reversed(range(1,len(a))): + aa = a[:l+1] + if b.startswith(aa): + return aa + return '' + +def common_prefix(strs): + if not strs: + return '' + p = None + for s in strs: + if p is None: + p = s + continue + p = _common_prefix(p, s) + if not p: + return '' + return p + +KIND_STATE = 'KIND_STATE' +KIND_FUNC = 'KIND_FUNC' +KIND_FSM = 'KIND_FSM' +BOX_SHAPES = { + KIND_STATE : None, + KIND_FUNC : 'box', + KIND_FSM : 'box3d', +} + +class Event: + def __init__(event, name): + event.name = name + event.short_name = name + + def __cmp__(event, other): + return cmp(event.name, other.name) + +class Edge: + def __init__(edge, to_state, event_name=None, style=None, action=None): + edge.to_state = to_state + edge.style = style + edge.events = [] + edge.actions = [] + edge.add_event_name(event_name) + edge.add_action(action) + + def add_event_name(edge, event_name): + if not event_name: + return + edge.add_event(Event(event_name)) + + def add_event(edge, event): + if not event: + return + if event in edge.events: + return + edge.events.append(event) + + def add_events(edge, events): + for event in events: + edge.add_event(event) + + def add_action(edge, action): + if not action or action in edge.actions: + return + edge.actions.append(action) + + def add_actions(edge, actions): + for action in actions: + edge.add_action(action) + + def event_names(edge): + return sorted([event.name for event in edge.events]) + + def event_labels(edge): + return sorted([event.short_name for event in edge.events]) + + def action_labels(edge): + return sorted([action + '()' for action in edge.actions]) + + def has_event_name(edge, event_name): + return event_name in edge.event_names() + +class State: + name = None + short_name = None + action = None + label = None + in_event_names = None + out_state_names = None + out_edges = None + kind = None + + def __init__(state): + state.in_event_names = [] + state.out_state_names = [] + state.out_edges = [] + state.kind = KIND_STATE + + def add_out_edge(state, edge): + for out_edge in state.out_edges: + if out_edge.to_state is edge.to_state: + if out_edge.style == edge.style: + out_edge.add_events(edge.events) + out_edge.add_actions(edge.actions) + return + # sanity + if out_edge.to_state.get_label() == edge.to_state.get_label(): + raise Exception('Two distinct states exist with identical labels.') + state.out_edges.append(edge) + + def get_label(state): + if state.label: + return state.label + l = [state.short_name] + if state.action: + if state.short_name == state.action: + l = [] + l.append(state.action + '()') + return r'\n'.join(l) + + def event_names(state): + event_names = [] + for out_edge in state.out_edges: + event_names.extend(out_edge.event_names()) + return event_names + + def shape_str(state): + shape = BOX_SHAPES.get(state.kind, None) + if not shape: + return '' + return ',shape=%s' % shape + + def __repr__(state): + return 'State(name=%r,short_name=%r,out=%d)' % (state.name, state.short_name, len(state.out_edges)) + +class Fsm: + def __init__(fsm, struct_name, states_struct_name, from_file=None): + fsm.states = [] + fsm.struct_name = struct_name + fsm.states_struct_name = states_struct_name + fsm.from_file = from_file + fsm.action_funcs = set() + fsm.event_names = set() + + def parse_states(fsm, src): + state = None + started = None + + IN_EVENTS = 'events' + OUT_STATES = 'states' + + lines = src.splitlines() + + for line in lines: + state_name = state_starts(line) + if state_name: + state = State() + fsm.states.append(state) + started = None + state.name = state_name + + if in_event_starts(line): + started = IN_EVENTS + if out_state_starts(line): + started = OUT_STATES + + if not state or not started: + continue + + tokens = states_or_events(line) + if started == IN_EVENTS: + state.in_event_names.extend(tokens) + elif started == OUT_STATES: + state.out_state_names.extend(tokens) + else: + err('ignoring: %r' % tokens) + + a = parse_action(line) + if a: + state.action = a + + + for state in fsm.states: + if state.action: + fsm.action_funcs.add(state.action) + if state.in_event_names: + fsm.event_names.update(state.in_event_names) + + fsm.make_states_short_names() + fsm.ref_out_states() + + def make_states_short_names(fsm): + p = common_prefix([s.name for s in fsm.states]) + for s in fsm.states: + s.short_name = s.name[len(p):] + return p + + def make_events_short_names(fsm): + p = common_prefix(fsm.event_names) + for state in fsm.states: + for edge in state.out_edges: + for event in edge.events: + event.short_name = event.name[len(p):] + + def ref_out_states(fsm): + for state in fsm.states: + for e in [Edge(fsm.find_state_by_name(n, True)) for n in state.out_state_names]: + state.add_out_edge(e) + + def find_state_by_name(fsm, name, strict=False): + for state in fsm.states: + if state.name == name: + return state + if strict: + raise Exception("State not found: %r" % name); + return None + + def find_state_by_action(fsm, action): + for state in fsm.states: + if state.action == action: + return state + return None + + def add_special_state(fsm, additional_states, name, in_state=None, + out_state=None, event_name=None, kind=KIND_FUNC, + state_action=None, label=None, edge_action=None): + additional_state = None + for s in additional_states: + if s.short_name == name: + additional_state = s + break; + + if not additional_state: + for s in fsm.states: + if s.short_name == name: + additional_state = s + break; + + if kind == KIND_FUNC and not state_action: + state_action = name + + if not additional_state: + additional_state = State() + additional_state.short_name = name + additional_state.action = state_action + additional_state.kind = kind + additional_state.label = label + additional_states.append(additional_state) + + if out_state: + additional_state.out_state_names.append(out_state.name) + additional_state.add_out_edge(Edge(out_state, event_name, 'dotted', + action=edge_action)) + + if in_state: + in_state.out_state_names.append(additional_state.name) + in_state.add_out_edge(Edge(additional_state, event_name, 'dotted', + action=edge_action)) + + + def find_event_edges(fsm, c_files): + # enrich state transitions between the states with event labels + func_to_state_transitions = listdict() + for c_file in c_files: + func_to_state_transitions.update( c_file.find_state_transitions(fsm.event_names) ) + + # edges between explicit states + for state in fsm.states: + transitions = func_to_state_transitions.get(state.action) + if not transitions: + continue + + for to_state_name, event_name in transitions: + if not event_name: + continue + for out_edge in state.out_edges: + if out_edge.to_state.name == to_state_name: + out_edge.add_event_name(event_name) + + additional_states = [] + + + # functions that aren't state actions but still effect state transitions + for func_name, transitions in func_to_state_transitions.items(): + if func_name in fsm.action_funcs: + continue + for to_state_name, event_name in transitions: + to_state = fsm.find_state_by_name(to_state_name) + if not to_state: + continue + fsm.add_special_state(additional_states, func_name, None, to_state, event_name) + + + event_sources = c_files.find_event_sources(fsm.event_names) + + for state in fsm.states: + + for in_event_name in state.in_event_names: + funcs_for_in_event = event_sources.get(in_event_name) + if not funcs_for_in_event: + continue + + found = False + for out_edge in state.out_edges: + if out_edge.has_event_name(in_event_name): + out_edge.action = r'\n'.join([(f + '()') for f in funcs_for_in_event + if f != state.action]) + + # if any functions that don't belong to a state trigger events, add + # them to the graph as well + additional_funcs = [f for f in funcs_for_in_event if f not in fsm.action_funcs] + for af in additional_funcs: + fsm.add_special_state(additional_states, af, None, state, in_event_name) + + fsm.states.extend(additional_states) + + # do any existing action functions by chance call other action functions? + for state in fsm.states: + if not state.action: + continue + callers = c_files.find_callers(state.action) + if not callers: + continue + for other_state in fsm.states: + if other_state.action in callers: + other_state.add_out_edge(Edge(state, None, 'dotted')) + + def add_fsm_alloc(fsm, c_files): + + allocating_funcs = [] + for c_file in c_files: + allocating_funcs.extend(c_file.fsm_allocators.get(fsm.struct_name, [])) + + starting_state = None + if fsm.states: + # assume the first state starts + starting_state = fsm.states[0] + + additional_states = [] + for func_name in allocating_funcs: + fsm.add_special_state(additional_states, func_name, None, starting_state) + + fsm.states.extend(additional_states) + + def add_cross_fsm_links(fsm, fsms, c_files, fsm_meta): + for state in fsm.states: + if not state.action: + continue + if state.kind == KIND_FSM: + continue + callers = c_files.find_callers(state.action) + + if state.kind == KIND_FUNC: + callers.append(state.action) + + if not callers: + continue + + for caller in callers: + for calling_fsm in fsms: + if calling_fsm is fsm: + continue + calling_state = calling_fsm.find_state_by_action(caller) + if not calling_state: + continue + if calling_state.kind == KIND_FSM: + continue + + label = None + if state.kind == KIND_STATE: + label=fsm.struct_name + ': ' + state.short_name + edge_action = caller + if calling_state.action == edge_action: + edge_action = None + calling_fsm.add_special_state(calling_fsm.states, fsm.struct_name, + calling_state, kind=KIND_FSM, edge_action=edge_action, label=label) + + label = None + if calling_state.kind == KIND_STATE: + label=calling_fsm.struct_name + ': ' + calling_state.short_name + edge_action = caller + if state.action == edge_action: + edge_action = None + fsm.add_special_state(fsm.states, calling_fsm.struct_name, None, + state, kind=KIND_FSM, edge_action=edge_action, + label=label) + + # meta overview + meta_called_fsm = fsm_meta.have_state(fsm.struct_name, KIND_FSM) + meta_calling_fsm = fsm_meta.have_state(calling_fsm.struct_name, KIND_FSM) + meta_calling_fsm.add_out_edge(Edge(meta_called_fsm)) + + + def have_state(fsm, name, kind=KIND_STATE): + state = fsm.find_state_by_name(name) + if not state: + state = State() + state.name = name + state.short_name = name + state.kind = kind + fsm.states.append(state) + return state + + def to_dot(fsm): + out = ['digraph G {', 'rankdir=LR;'] + + for state in fsm.states: + out.append('%s [label="%s"%s]' % (state.short_name, state.get_label(), + state.shape_str())) + + for state in fsm.states: + for out_edge in state.out_edges: + attrs = [] + labels = [] + if out_edge.events: + labels.extend(out_edge.event_labels()) + if out_edge.actions: + labels.extend(out_edge.action_labels()) + if labels: + attrs.append('label="%s"' % (r'\n'.join(labels))) + if out_edge.style: + attrs.append('style=%s'% out_edge.style) + attrs_str = '' + if attrs: + attrs_str = ' [%s]' % (','.join(attrs)) + out.append('%s->%s%s' % (state.short_name, out_edge.to_state.short_name, attrs_str)) + + out.append('}\n') + + return '\n'.join(out) + + def write_dot_file(fsm): + dot_path = '%s.dot' % fsm.struct_name + f = open(dot_path, 'w') + f.write(fsm.to_dot()) + f.close() + print(dot_path) + + +re_fsm = re.compile(r'struct osmo_fsm ([a-z_][a-z_0-9]*) =') +re_fsm_states_struct_name = re.compile(r'\bstates = ([a-z_][a-z_0-9]*)\W*,') +re_fsm_states = re.compile(r'struct osmo_fsm_state ([a-z_][a-z_0-9]*)\[\] =') +re_func = re.compile(r'(\b[a-z_][a-z_0-9]*\b)\([^)]*\)\W*^{', re.MULTILINE) +re_state_trigger = re.compile(r'osmo_fsm_inst_state_chg\([^,]+,\W*([A-Z_][A-Z_0-9]*)\W*,', re.M) +re_fsm_alloc = re.compile(r'osmo_fsm_inst_alloc[_child]*\(\W*&([a-z_][a-z_0-9]*),', re.M) +re_fsm_event_dispatch = re.compile(r'osmo_fsm_inst_dispatch\(\W*[^,]+,\W*([A-Z_][A-Z_0-9]*)\W*,', re.M) + +class CFile(): + def __init__(c_file, path): + c_file.path = path + c_file.src = open(path).read() + c_file.funcs = {} + c_file.fsm_allocators = listdict() + + def extract_block(c_file, brace_open, brace_close, start): + pos = 0 + try: + src = c_file.src + block_start = src.find(brace_open, start) + + pos = block_start + level = 1 + while level > 0: + pos += 1 + if src[pos] == brace_open: + level += 1 + elif src[pos] == brace_close: + level -= 1 + + return src[block_start+1:pos] + except: + print("Error while trying to extract a code block from %r char pos %d" % (c_file.path, pos)) + print("Block start at char pos %d" % block_start) + try: + print(src[block_start - 20 : block_start + 20]) + print('...') + print(src[pos - 20 : pos + 20]) + except: + pass + return '' + + + def find_fsms(c_file): + fsms = [] + for m in re_fsm.finditer(c_file.src): + struct_name = m.group(1) + struct_def = c_file.extract_block('{', '}', m.start()) + states_struct_name = re_fsm_states_struct_name.findall(struct_def)[0] + fsm = Fsm(struct_name, states_struct_name, c_file) + fsms.append(fsm) + return fsms + + def find_fsm_states(c_file, fsms): + for m in re_fsm_states.finditer(c_file.src): + states_struct_name = m.group(1) + for fsm in fsms: + if states_struct_name == fsm.states_struct_name: + fsm.parse_states(c_file.extract_block('{', '}', m.start())) + + def parse_functions(c_file): + funcs = {} + for m in re_func.finditer(c_file.src): + name = m.group(1) + func_src = c_file.extract_block('{', '}', m.start()) + funcs[name] = func_src + c_file.funcs = funcs + c_file.find_fsm_allocators() + + def find_callers(c_file, func_name): + func_call = func_name + '(' + callers = [] + for func_name, src in c_file.funcs.items(): + if src.find(func_call) >= 0: + callers.append(func_name) + return callers + + def find_fsm_allocators(c_file): + c_file.fsm_allocators = listdict() + for func_name, src in c_file.funcs.items(): + for m in re_fsm_alloc.finditer(src): + fsm_struct_name = m.group(1) + c_file.fsm_allocators.add(fsm_struct_name, func_name) + + def find_state_transitions(c_file, event_names): + TO_STATE = 'TO_STATE' + EVENT = 'EVENT' + func_to_state_transitions = listdict() + + for func_name, src in c_file.funcs.items(): + found_tokens = [] + + for m in re_state_trigger.finditer(src): + to_state = m.group(1) + found_tokens.append((m.start(), TO_STATE, to_state)) + + for event in event_names: + re_event = re.compile(r'\b(' + event + r')\b') + for m in re_event.finditer(src): + event = m.group(1) + found_tokens.append((m.start(), EVENT, event)) + + found_tokens = sorted(found_tokens) + + last_event = None + for start, kind, name in found_tokens: + if kind == EVENT: + last_event = name + else: + func_to_state_transitions.add(func_name, (name, last_event)) + + return func_to_state_transitions + + + def find_event_sources(c_file, event_names): + c_file.event_sources = listdict() + for func_name, src in c_file.funcs.items(): + for m in re_fsm_event_dispatch.finditer(src): + event_name = m.group(1) + c_file.event_sources.add(event_name, func_name) + +class CFiles(list): + + def find_callers(c_files, func_name): + callers = [] + for c_file in c_files: + callers.extend(c_file.find_callers(func_name)) + return callers + + def find_func_to_state_transitions(c_files): + func_to_state_transitions = listdict() + for c_file in c_files: + func_to_state_transitions.update( c_file.find_state_transitions(fsm.event_names) ) + return func_to_state_transitions + + def find_event_sources(c_files, event_names): + event_sources = listdict() + for c_file in c_files: + for event, sources in c_file.event_sources.items(): + if event in event_names: + event_sources.extend(event, sources) + return event_sources + +c_files = CFiles() +paths_seen = set() +for path in sys.argv[1:]: + if path in paths_seen: + continue + paths_seen.add(path) + c_file = CFile(path) + c_files.append(c_file) + +for c_file in c_files: + c_file.parse_functions() + +fsms = [] +for c_file in c_files: + fsms.extend(c_file.find_fsms()) + +for c_file in c_files: + c_file.find_fsm_states(fsms) + c_file.find_event_sources(fsms) + +for fsm in fsms: + fsm.find_event_edges(c_files) + fsm.add_fsm_alloc(c_files) + +fsm_meta = Fsm("meta", "meta") +for fsm in fsms: + fsm.add_cross_fsm_links(fsms, c_files, fsm_meta) + +for fsm in fsms: + fsm.make_events_short_names() + +for fsm in fsms: + fsm.write_dot_file() + +fsm_meta.write_dot_file() + + +# vim: tabstop=2 shiftwidth=2 expandtab -- cgit v1.2.3