# Copyright 2013-2022 Lawrence Livermore National Security, LLC and other
# Spack Project Developers. See the top-level COPYRIGHT file for details.
#
# SPDX-License-Identifier: (Apache-2.0 OR MIT)
from __future__ import division, print_function
from collections import defaultdict
import six.moves.urllib.parse as urllib_parse
import llnl.util.tty.color as color
from llnl.util import tty
import spack.fetch_strategy as fs
import spack.repo
import spack.util.crypto as crypto
from spack.url import (
UndetectableNameError,
UndetectableVersionError,
UrlParseError,
color_url,
parse_name,
parse_name_offset,
parse_version,
parse_version_offset,
substitute_version,
substitution_offsets,
)
from spack.util.naming import simplify_name
from spack.util.web import find_versions_of_archive
description = "debugging tool for url parsing"
section = "developer"
level = "long"
[docs]def setup_parser(subparser):
sp = subparser.add_subparsers(metavar='SUBCOMMAND', dest='subcommand')
# Parse
parse_parser = sp.add_parser('parse', help='attempt to parse a url')
parse_parser.add_argument(
'url',
help='url to parse')
parse_parser.add_argument(
'-s', '--spider', action='store_true',
help='spider the source page for versions')
# List
list_parser = sp.add_parser('list', help='list urls in all packages')
list_parser.add_argument(
'-c', '--color', action='store_true',
help='color the parsed version and name in the urls shown '
'(versions will be cyan, name red)')
list_parser.add_argument(
'-e', '--extrapolation', action='store_true',
help='color the versions used for extrapolation as well '
'(additional versions will be green, names magenta)')
excl_args = list_parser.add_mutually_exclusive_group()
excl_args.add_argument(
'-n', '--incorrect-name', action='store_true',
help='only list urls for which the name was incorrectly parsed')
excl_args.add_argument(
'-N', '--correct-name', action='store_true',
help='only list urls for which the name was correctly parsed')
excl_args.add_argument(
'-v', '--incorrect-version', action='store_true',
help='only list urls for which the version was incorrectly parsed')
excl_args.add_argument(
'-V', '--correct-version', action='store_true',
help='only list urls for which the version was correctly parsed')
# Summary
sp.add_parser(
'summary',
help='print a summary of how well we are parsing package urls')
# Stats
stats_parser = sp.add_parser(
'stats',
help='print statistics on versions and checksums for all packages')
stats_parser.add_argument(
"--show-issues", action="store_true",
help="show packages with issues (md5 hashes, http urls)"
)
[docs]def url(parser, args):
action = {
'parse': url_parse,
'list': url_list,
'summary': url_summary,
'stats': url_stats,
}
action[args.subcommand](args)
[docs]def url_parse(args):
url = args.url
tty.msg('Parsing URL: {0}'.format(url))
print()
ver, vs, vl, vi, vregex = parse_version_offset(url)
tty.msg('Matched version regex {0:>2}: r{1!r}'.format(vi, vregex))
name, ns, nl, ni, nregex = parse_name_offset(url, ver)
tty.msg('Matched name regex {0:>2}: r{1!r}'.format(ni, nregex))
print()
tty.msg('Detected:')
try:
print_name_and_version(url)
except UrlParseError as e:
tty.error(str(e))
print(' name: {0}'.format(name))
print(' version: {0}'.format(ver))
print()
tty.msg('Substituting version 9.9.9b:')
newurl = substitute_version(url, '9.9.9b')
print_name_and_version(newurl)
if args.spider:
print()
tty.msg('Spidering for versions:')
versions = find_versions_of_archive(url)
if not versions:
print(' Found no versions for {0}'.format(name))
return
max_len = max(len(str(v)) for v in versions)
for v in sorted(versions):
print('{0:{1}} {2}'.format(v, max_len, versions[v]))
[docs]def url_list(args):
urls = set()
# Gather set of URLs from all packages
for pkg in spack.repo.path.all_packages():
url = getattr(pkg, 'url', None)
urls = url_list_parsing(args, urls, url, pkg)
for params in pkg.versions.values():
url = params.get('url', None)
urls = url_list_parsing(args, urls, url, pkg)
# Print URLs
for url in sorted(urls):
if args.color or args.extrapolation:
print(color_url(url, subs=args.extrapolation, errors=True))
else:
print(url)
# Return the number of URLs that were printed, only for testing purposes
return len(urls)
[docs]def url_summary(args):
# Collect statistics on how many URLs were correctly parsed
total_urls = 0
correct_names = 0
correct_versions = 0
# Collect statistics on which regexes were matched and how often
name_regex_dict = dict()
right_name_count = defaultdict(int)
wrong_name_count = defaultdict(int)
version_regex_dict = dict()
right_version_count = defaultdict(int)
wrong_version_count = defaultdict(int)
tty.msg('Generating a summary of URL parsing in Spack...')
# Loop through all packages
for pkg in spack.repo.path.all_packages():
urls = set()
url = getattr(pkg, 'url', None)
if url:
urls.add(url)
for params in pkg.versions.values():
url = params.get('url', None)
if url:
urls.add(url)
# Calculate statistics
for url in urls:
total_urls += 1
# Parse versions
version = None
try:
version, vs, vl, vi, vregex = parse_version_offset(url)
version_regex_dict[vi] = vregex
if version_parsed_correctly(pkg, version):
correct_versions += 1
right_version_count[vi] += 1
else:
wrong_version_count[vi] += 1
except UndetectableVersionError:
pass
# Parse names
try:
name, ns, nl, ni, nregex = parse_name_offset(url, version)
name_regex_dict[ni] = nregex
if name_parsed_correctly(pkg, name):
correct_names += 1
right_name_count[ni] += 1
else:
wrong_name_count[ni] += 1
except UndetectableNameError:
pass
print()
print(' Total URLs found: {0}'.format(total_urls))
print(' Names correctly parsed: {0:>4}/{1:>4} ({2:>6.2%})'.format(
correct_names, total_urls, correct_names / total_urls))
print(' Versions correctly parsed: {0:>4}/{1:>4} ({2:>6.2%})'.format(
correct_versions, total_urls, correct_versions / total_urls))
print()
tty.msg('Statistics on name regular expressions:')
print()
print(' Index Right Wrong Total Regular Expression')
for ni in sorted(name_regex_dict.keys()):
print(' {0:>5} {1:>5} {2:>5} {3:>5} r{4!r}'.format(
ni,
right_name_count[ni],
wrong_name_count[ni],
right_name_count[ni] + wrong_name_count[ni],
name_regex_dict[ni])
)
print()
tty.msg('Statistics on version regular expressions:')
print()
print(' Index Right Wrong Total Regular Expression')
for vi in sorted(version_regex_dict.keys()):
print(' {0:>5} {1:>5} {2:>5} {3:>5} r{4!r}'.format(
vi,
right_version_count[vi],
wrong_version_count[vi],
right_version_count[vi] + wrong_version_count[vi],
version_regex_dict[vi])
)
print()
# Return statistics, only for testing purposes
return (total_urls, correct_names, correct_versions,
right_name_count, right_version_count)
[docs]def url_stats(args):
# dictionary of issue type -> package -> descriptions
issues = defaultdict(lambda: defaultdict(lambda: []))
class UrlStats(object):
def __init__(self):
self.total = 0
self.schemes = defaultdict(lambda: 0)
self.checksums = defaultdict(lambda: 0)
self.url_type = defaultdict(lambda: 0)
self.git_type = defaultdict(lambda: 0)
def add(self, pkg_name, fetcher):
self.total += 1
url_type = fetcher.url_attr
self.url_type[url_type or 'no code'] += 1
if url_type == 'url':
digest = getattr(fetcher, 'digest', None)
if digest:
algo = crypto.hash_algo_for_digest(digest)
else:
algo = 'no checksum'
self.checksums[algo] += 1
if algo == "md5":
md5_hashes = issues["md5 hashes"]
md5_hashes[pkg_name].append(fetcher.url)
# parse out the URL scheme (https/http/ftp/etc.)
urlinfo = urllib_parse.urlparse(fetcher.url)
self.schemes[urlinfo.scheme] += 1
if urlinfo.scheme == "http":
http_urls = issues["http urls"]
http_urls[pkg_name].append(fetcher.url)
elif url_type == 'git':
if getattr(fetcher, 'commit', None):
self.git_type['commit'] += 1
elif getattr(fetcher, 'branch', None):
self.git_type['branch'] += 1
elif getattr(fetcher, 'tag', None):
self.git_type['tag'] += 1
else:
self.git_type['no ref'] += 1
npkgs = 0
version_stats = UrlStats()
resource_stats = UrlStats()
for pkg in spack.repo.path.all_packages():
npkgs += 1
for v in pkg.versions:
try:
fetcher = fs.for_package_version(pkg, v)
except (fs.InvalidArgsError, fs.FetcherConflict):
continue
version_stats.add(pkg.name, fetcher)
for _, resources in pkg.resources.items():
for resource in resources:
resource_stats.add(pkg.name, resource.fetcher)
# print a nice summary table
tty.msg("URL stats for %d packages:" % npkgs)
def print_line():
print("-" * 62)
def print_stat(indent, name, stat_name=None):
width = 20 - indent
fmt = " " * indent
fmt += "%%-%ds" % width
if stat_name is None:
print(fmt % name)
else:
fmt += "%12d%8.1f%%%12d%8.1f%%"
v = getattr(version_stats, stat_name).get(name, 0)
r = getattr(resource_stats, stat_name).get(name, 0)
print(fmt % (name,
v, v / version_stats.total * 100,
r, r / resource_stats.total * 100))
print_line()
print("%-20s%12s%9s%12s%9s" % ("stat", "versions", "%", "resources", "%"))
print_line()
print_stat(0, "url", "url_type")
print_stat(4, "schemes")
schemes = set(version_stats.schemes) | set(resource_stats.schemes)
for scheme in schemes:
print_stat(8, scheme, "schemes")
print_stat(4, "checksums")
checksums = set(version_stats.checksums) | set(resource_stats.checksums)
for checksum in checksums:
print_stat(8, checksum, "checksums")
print_line()
types = set(version_stats.url_type) | set(resource_stats.url_type)
types -= set(["url", "git"])
for url_type in sorted(types):
print_stat(0, url_type, "url_type")
print_line()
print_stat(0, "git", "url_type")
git_types = set(version_stats.git_type) | set(resource_stats.git_type)
for git_type in sorted(git_types):
print_stat(4, git_type, "git_type")
print_line()
if args.show_issues:
total_issues = sum(
len(issues)
for _, pkg_issues in issues.items()
for _, issues in pkg_issues.items()
)
print()
tty.msg("Found %d issues." % total_issues)
for issue_type, pkgs in issues.items():
tty.msg("Package URLs with %s" % issue_type)
for pkg, pkg_issues in pkgs.items():
color.cprint(" @*C{%s}" % pkg)
for issue in pkg_issues:
print(" %s" % issue)
[docs]def print_name_and_version(url):
"""Prints a URL. Underlines the detected name with dashes and
the detected version with tildes.
Args:
url (str): The url to parse
"""
name, ns, nl, ntup, ver, vs, vl, vtup = substitution_offsets(url)
underlines = [' '] * max(ns + nl, vs + vl)
for i in range(ns, ns + nl):
underlines[i] = '-'
for i in range(vs, vs + vl):
underlines[i] = '~'
print(' {0}'.format(url))
print(' {0}'.format(''.join(underlines)))
[docs]def url_list_parsing(args, urls, url, pkg):
"""Helper function for :func:`url_list`.
Args:
args (argparse.Namespace): The arguments given to ``spack url list``
urls (set): List of URLs that have already been added
url (str or None): A URL to potentially add to ``urls`` depending on
``args``
pkg (spack.package.PackageBase): The Spack package
Returns:
set: The updated set of ``urls``
"""
if url:
if args.correct_name or args.incorrect_name:
# Attempt to parse the name
try:
name = parse_name(url)
if (args.correct_name and
name_parsed_correctly(pkg, name)):
# Add correctly parsed URLs
urls.add(url)
elif (args.incorrect_name and
not name_parsed_correctly(pkg, name)):
# Add incorrectly parsed URLs
urls.add(url)
except UndetectableNameError:
if args.incorrect_name:
# Add incorrectly parsed URLs
urls.add(url)
elif args.correct_version or args.incorrect_version:
# Attempt to parse the version
try:
version = parse_version(url)
if (args.correct_version and
version_parsed_correctly(pkg, version)):
# Add correctly parsed URLs
urls.add(url)
elif (args.incorrect_version and
not version_parsed_correctly(pkg, version)):
# Add incorrectly parsed URLs
urls.add(url)
except UndetectableVersionError:
if args.incorrect_version:
# Add incorrectly parsed URLs
urls.add(url)
else:
urls.add(url)
return urls
[docs]def name_parsed_correctly(pkg, name):
"""Determine if the name of a package was correctly parsed.
Args:
pkg (spack.package.PackageBase): The Spack package
name (str): The name that was extracted from the URL
Returns:
bool: True if the name was correctly parsed, else False
"""
pkg_name = remove_prefix(pkg.name)
name = simplify_name(name)
return name == pkg_name
[docs]def version_parsed_correctly(pkg, version):
"""Determine if the version of a package was correctly parsed.
Args:
pkg (spack.package.PackageBase): The Spack package
version (str): The version that was extracted from the URL
Returns:
bool: True if the name was correctly parsed, else False
"""
version = remove_separators(version)
# If the version parsed from the URL is listed in a version()
# directive, we assume it was correctly parsed
for pkg_version in pkg.versions:
pkg_version = remove_separators(pkg_version)
if pkg_version == version:
return True
return False
[docs]def remove_prefix(pkg_name):
"""Remove build system prefix ('py-', 'perl-', etc.) from a package name.
After determining a name, `spack create` determines a build system.
Some build systems prepend a special string to the front of the name.
Since this can't be guessed from the URL, it would be unfair to say
that these names are incorrectly parsed, so we remove them.
Args:
pkg_name (str): the name of the package
Returns:
str: the name of the package with any build system prefix removed
"""
prefixes = [
'r-', 'py-', 'tcl-', 'lua-', 'perl-', 'ruby-', 'llvm-',
'intel-', 'votca-', 'octave-', 'gtkorvo-'
]
prefix = next((p for p in prefixes if pkg_name.startswith(p)), '')
return pkg_name[len(prefix):]
[docs]def remove_separators(version):
"""Remove separator characters ('.', '_', and '-') from a version.
A version like 1.2.3 may be displayed as 1_2_3 in the URL.
Make sure 1.2.3, 1-2-3, 1_2_3, and 123 are considered equal.
Unfortunately, this also means that 1.23 and 12.3 are equal.
Args:
version (str or spack.version.Version): A version
Returns:
str: The version with all separator characters removed
"""
version = str(version)
version = version.replace('.', '')
version = version.replace('_', '')
version = version.replace('-', '')
return version