def manifest_load(self):
"""Parse the manifest file and set self.config_hash and
self.layer_hashes."""
def bad_key(key):
ch.FATAL("manifest: %s: no key: %s" % (self.manifest_path, key))
# read and parse the JSON
fp = ch.open_(self.manifest_path, "rt", encoding="UTF-8")
text = ch.ossafe(fp.read, "can't read: %s" % self.manifest_path)
ch.ossafe(fp.close, "can't close: %s" % self.manifest_path)
ch.DEBUG("manifest:\n%s" % text)
try:
manifest = json.loads(text)
except json.JSONDecodeError as x:
ch.FATAL("can't parse manifest file: %s:%d: %s"
% (self.manifest_path, x.lineno, x.msg))
# validate schema version
try:
version = manifest['schemaVersion']
except KeyError:
bad_key("schemaVersion")
if (version not in {1,2}):
ch.FATAL("unsupported manifest schema version: %s" % repr(version))
# load config hash
#
# FIXME: Manifest version 1 does not list a config blob. It does have
# things (plural) that look like a config at history/v1Compatibility as
# an embedded JSON string :P but I haven't dug into it.
if (version == 1):
ch.WARNING("no config; manifest schema version 1")
self.config_hash = None
else: # version == 2
try:
self.config_hash = ch.digest_trim(manifest["config"]["digest"])
except KeyError:
bad_key("config/digest")
# load layer hashes
if (version == 1):
key1 = "fsLayers"
key2 = "blobSum"
else: # version == 2
key1 = "layers"
key2 = "digest"
if (key1 not in manifest):
bad_key(key1)
self.layer_hashes = list()
for i in manifest[key1]:
if (key2 not in i):
bad_key("%s/%s" % (key1, key2))
self.layer_hashes.append(ch.digest_trim(i[key2]))
if (version == 1):
self.layer_hashes.reverse()
def main(cli_):
# CLI namespace. :P
global cli
cli = cli_
# Infer input file if needed.
if (cli.file is None):
cli.file = cli.context + "/Dockerfile"
# Infer image name if needed.
if (cli.tag is None):
m = re.search(r"(([^/]+)/)?Dockerfile(\.(.+))?$",
os.path.abspath(cli.file))
if (m is not None):
if m.group(4): # extension
cli.tag = m.group(4)
elif m.group(2): # containing directory
cli.tag = m.group(2)
# Deal with build arguments.
def build_arg_get(arg):
kv = arg.split("=")
if (len(kv) == 2):
return kv
else:
v = os.getenv(kv[0])
if (v is None):
ch.FATAL("--build-arg: %s: no value and not in environment" % kv[0])
return (kv[0], v)
if (cli.build_arg is None):
cli.build_arg = list()
cli.build_arg = dict( build_arg_get(i) for i in cli.build_arg )
# Finish CLI initialization.
ch.DEBUG(cli)
ch.dependencies_check()
# Guess whether the context is a URL, and error out if so. This can be a
# typical looking URL e.g. "https://..." or also something like
# "[email protected]:...". The line noise in the second line of the regex is
# to match this second form. Username and host characters from
# https://tools.ietf.org/html/rfc3986.
if (re.search(r""" ^((git|git+ssh|http|https|ssh)://
| ^[\w.~%!$&'\(\)\*\+,;=-]+@[\w.~%!$&'\(\)\*\+,;=-]+:)""",
cli.context, re.VERBOSE) is not None):
ch.FATAL("not yet supported: issue #773: URL context: %s" % cli.context)
if (os.path.exists(cli.context + "/.dockerignore")):
ch.WARNING("not yet supported, ignored: issue #777: .dockerignore file")
# Set up build environment.
global env
env = Environment()
# Read input file.
if (cli.file == "-"):
text = ch.ossafe(sys.stdin.read, "can't read stdin")
else:
fp = ch.open_(cli.file, "rt")
text = ch.ossafe(fp.read, "can't read: %s" % cli.file)
fp.close()
# Parse it.
parser = lark.Lark("?start: dockerfile\n" + ch.GRAMMAR,
parser="earley", propagate_positions=True)
# Avoid Lark issue #237: lark.exceptions.UnexpectedEOF if the file does not
# end in newline.
text += "\n"
try:
tree = parser.parse(text)
except lark.exceptions.UnexpectedInput as x:
ch.DEBUG(x) # noise about what was expected in the grammar
ch.FATAL("can't parse: %s:%d,%d\n\n%s" % (cli.file, x.line, x.column, x.get_context(text, 39)))
ch.DEBUG(tree.pretty())
# Sometimes we exit after parsing.
if (cli.parse_only):
sys.exit(0)
# Count the number of stages (i.e., FROM instructions)
global image_ct
image_ct = sum(1 for i in ch.tree_children(tree, "from_"))
# Traverse the tree and do what it says.
#
# We don't actually care whether the tree is traversed breadth-first or
# depth-first, but we *do* care that instruction nodes are visited in
# order. Neither visit() nor visit_topdown() are documented as of
# 2020-06-11 [1], but examining source code [2] shows that visit_topdown()
# uses Tree.iter_trees_topdown(), which *is* documented to be in-order [3].
#
# This change seems to have been made in 0.8.6 (see PR #761); before then,
# visit() was in order. Therefore, we call that instead, if visit_topdown()
# is not present, to improve compatibility (see issue #792).
#
# [1]: https://lark-parser.readthedocs.io/en/latest/visitors/#visitors
# [2]: https://github.com/lark-parser/lark/blob/445c8d4/lark/visitors.py#L211
# [3]: https://lark-parser.readthedocs.io/en/latest/classes/#tree
ml = Main_Loop()
if (hasattr(ml, 'visit_topdown')):
ml.visit_topdown(tree)
else:
ml.visit(tree)
# Check that all build arguments were consumed.
if (len(cli.build_arg) != 0):
ch.FATAL("--build-arg: not consumed: " + " ".join(cli.build_arg.keys()))
# Print summary & we're done.
if (ml.instruction_ct == 0):
ch.FATAL("no instructions found: %s" % cli.file)
assert (image_i + 1 == image_ct) # should have errored already if not
ch.INFO("grown in %d instructions: %s"
% (ml.instruction_ct, images[image_i]))
def execute_(self):
env.env[self.key] = self.value
with ch.open_(images[image_i].unpack_path + "/ch/environment", "wt") \
as fp:
for (k, v) in env.env.items():
print("%s=%s" % (k, v), file=fp)