bitcoin/contrib/devtools/circular-dependencies.py
Russell Yanofsky 318f41fb2c circular-dependencies: Avoid treating some .h/.cpp files as a unit
This avoids a bogus circular dependency error in the next commit:

interfaces/chain -> interfaces/wallet -> wallet/wallet -> interfaces/chain

Which is incorrect, because interfaces/chain.cpp depends only on the
interfaces/wallet.h file, not the interfaces/wallet.cpp file, and it is
wrong to treat these as a unit. Inside the interfaces directory, .h files
contain abstract class definitions and .cpp files contain implementations of
those classes, so you don't need to link against .cpp files if you're only
using the abstract class definition in the .h file.

An alternative fix might be to rename all the cpp files in the interfaces
directory like: chain.cpp->chain_impl.cpp, node.cpp->node_impl.cpp. But just
getting the linter to treat these files as independent dependencies seemed
like it would allow keeping code organization straightforward and avoiding
the need to rename things.
2019-02-22 15:43:02 -04:00

89 lines
2.9 KiB
Python
Executable File

#!/usr/bin/env python3
import sys
import re
MAPPING = {
'core_read.cpp': 'core_io.cpp',
'core_write.cpp': 'core_io.cpp',
}
# Directories with header-based modules, where the assumption that .cpp files
# define functions and variables declared in corresponding .h files is
# incorrect.
HEADER_MODULE_PATHS = [
'interfaces/'
]
def module_name(path):
if path in MAPPING:
path = MAPPING[path]
if any(path.startswith(dirpath) for dirpath in HEADER_MODULE_PATHS):
return path
if path.endswith(".h"):
return path[:-2]
if path.endswith(".c"):
return path[:-2]
if path.endswith(".cpp"):
return path[:-4]
return None
files = dict()
deps = dict()
RE = re.compile("^#include <(.*)>")
# Iterate over files, and create list of modules
for arg in sys.argv[1:]:
module = module_name(arg)
if module is None:
print("Ignoring file %s (does not constitute module)\n" % arg)
else:
files[arg] = module
deps[module] = set()
# Iterate again, and build list of direct dependencies for each module
# TODO: implement support for multiple include directories
for arg in sorted(files.keys()):
module = files[arg]
with open(arg, 'r', encoding="utf8") as f:
for line in f:
match = RE.match(line)
if match:
include = match.group(1)
included_module = module_name(include)
if included_module is not None and included_module in deps and included_module != module:
deps[module].add(included_module)
# Loop to find the shortest (remaining) circular dependency
have_cycle = False
while True:
shortest_cycle = None
for module in sorted(deps.keys()):
# Build the transitive closure of dependencies of module
closure = dict()
for dep in deps[module]:
closure[dep] = []
while True:
old_size = len(closure)
old_closure_keys = sorted(closure.keys())
for src in old_closure_keys:
for dep in deps[src]:
if dep not in closure:
closure[dep] = closure[src] + [src]
if len(closure) == old_size:
break
# If module is in its own transitive closure, it's a circular dependency; check if it is the shortest
if module in closure and (shortest_cycle is None or len(closure[module]) + 1 < len(shortest_cycle)):
shortest_cycle = [module] + closure[module]
if shortest_cycle is None:
break
# We have the shortest circular dependency; report it
module = shortest_cycle[0]
print("Circular dependency: %s" % (" -> ".join(shortest_cycle + [module])))
# And then break the dependency to avoid repeating in other cycles
deps[shortest_cycle[-1]] = deps[shortest_cycle[-1]] - set([module])
have_cycle = True
sys.exit(1 if have_cycle else 0)