Change (raw_)hexdump to yield lines instead of printing directly

This commit is contained in:
dgelessus 2020-07-21 14:27:43 +02:00
parent 664e992fa3
commit 5c96baea29
1 changed files with 22 additions and 15 deletions

View File

@ -183,7 +183,7 @@ def filter_resources(rf: api.ResourceFile, filters: typing.Sequence[str]) -> typ
yield res
def hexdump(data: bytes) -> None:
def hexdump(data: bytes) -> typing.Iterable[str]:
last_line = None
asterisk_shown = False
for i in range(0, len(data), 16):
@ -192,23 +192,23 @@ def hexdump(data: bytes) -> None:
# This is unambiguous - to find out how many lines were collapsed this way, the user can compare the addresses of the lines before and after the asterisk.
if line == last_line:
if not asterisk_shown:
print("*")
yield "*"
asterisk_shown = True
else:
line_hex_left = " ".join(f"{byte:02x}" for byte in line[:8])
line_hex_right = " ".join(f"{byte:02x}" for byte in line[8:])
line_char = line.decode(_TEXT_ENCODING).translate(_TRANSLATE_NONPRINTABLES)
print(f"{i:08x} {line_hex_left:<{8*2+7}} {line_hex_right:<{8*2+7}} |{line_char}|")
yield f"{i:08x} {line_hex_left:<{8*2+7}} {line_hex_right:<{8*2+7}} |{line_char}|"
asterisk_shown = False
last_line = line
if data:
print(f"{len(data):08x}")
yield f"{len(data):08x}"
def raw_hexdump(data: bytes) -> None:
def raw_hexdump(data: bytes) -> typing.Iterable[str]:
for i in range(0, len(data), 16):
print(" ".join(f"{byte:02x}" for byte in data[i:i + 16]))
yield " ".join(f"{byte:02x}" for byte in data[i:i + 16])
def translate_text(data: bytes) -> str:
@ -276,7 +276,8 @@ def show_filtered_resources(resources: typing.Sequence[api.Resource], format: st
desc = describe_resource(res, include_type=True, decompress=decompress)
print(f"Resource {desc}:")
if format == "dump":
hexdump(data)
for line in hexdump(data):
print(line)
elif format == "dump-text":
print(translate_text(data))
else:
@ -285,7 +286,8 @@ def show_filtered_resources(resources: typing.Sequence[api.Resource], format: st
elif format == "hex":
# Data only as hex
raw_hexdump(data)
for line in raw_hexdump(data):
print(line)
elif format == "raw":
# Data only as raw bytes
@ -465,18 +467,20 @@ def do_read_header(ns: argparse.Namespace) -> typing.NoReturn:
if ns.format == "dump":
dump_func = hexdump
elif ns.format == "dump-text":
def dump_func(data: bytes) -> None:
print(translate_text(data))
def dump_func(data: bytes) -> typing.Iterable[str]:
yield translate_text(data)
else:
raise AssertionError(f"Unhandled --format: {ns.format!r}")
if ns.part in {"system", "all"}:
print("System-reserved header data:")
dump_func(rf.header_system_data)
for line in dump_func(rf.header_system_data):
print(line)
if ns.part in {"application", "all"}:
print("Application-specific header data:")
dump_func(rf.header_application_data)
for line in dump_func(rf.header_application_data):
print(line)
elif ns.format in {"hex", "raw"}:
if ns.part == "system":
data = rf.header_system_data
@ -488,7 +492,8 @@ def do_read_header(ns: argparse.Namespace) -> typing.NoReturn:
raise AssertionError(f"Unhandled --part: {ns.part!r}")
if ns.format == "hex":
raw_hexdump(data)
for line in raw_hexdump(data):
print(line)
elif ns.format == "raw":
sys.stdout.buffer.write(data)
else:
@ -502,10 +507,12 @@ def do_read_header(ns: argparse.Namespace) -> typing.NoReturn:
def do_info(ns: argparse.Namespace) -> typing.NoReturn:
with open_resource_file(ns.file, fork=ns.fork) as rf:
print("System-reserved header data:")
hexdump(rf.header_system_data)
for line in hexdump(rf.header_system_data):
print(line)
print()
print("Application-specific header data:")
hexdump(rf.header_application_data)
for line in hexdump(rf.header_application_data):
print(line)
print()
print(f"Resource data starts at {rf.data_offset:#x} and is {rf.data_length:#x} bytes long")