From: "Sam James" <sam@gentoo.org>
To: gentoo-commits@lists.gentoo.org
Subject: [gentoo-commits] proj/portage:master commit in: lib/portage/dbapi/
Date: Thu, 11 Sep 2025 18:07:29 +0000 (UTC) [thread overview]
Message-ID: <1757614014.443d5765431e3d9ada2bd9b940ca5c14040a06e6.sam@gentoo> (raw)
commit: 443d5765431e3d9ada2bd9b940ca5c14040a06e6
Author: Florian Schmaus <flow <AT> gentoo <DOT> org>
AuthorDate: Tue Sep 9 19:23:32 2025 +0000
Commit: Sam James <sam <AT> gentoo <DOT> org>
CommitDate: Thu Sep 11 18:06:54 2025 +0000
URL: https://gitweb.gentoo.org/proj/portage.git/commit/?id=443d5765
bintree: Accelerate index fetch by requesting Packages.gz first
Fetching the compressed package index (Packages.gz) significantly
reduces the time required to transfer the index. At the time of
writing this, the size of the uncompressed package index of the
official Gentoo binhost is ~16MiB, while the compressed variant's size
is ~1.7MiB.
Portage got support writing the a compressed package index with
11c0619c63b5 ("Portage writes a compressed copy of 'Packages' index
file."). However, this did not include *fetching* the compressed
package index, as the idea back then was to piggy-back on HTTP
compression. Unfortunately, many binhosts do not configure their HTTP
service accordingly.
This commit adjusts Portage logic to first try to fetch the compressed
package index with a fallback to the uncompressed one.
The diff of this change looks wild, but essentially it introduced only
a few new lines of code:
- a new loop: for remote_pkgindex_file in ("Packages.gz", "Packages"):
- which required indenting the existing code by one level
- a break statement at the end of the loop (if it was successful)
- wrapping the existing file object into GzipFile if required via
if remote_pkgindex_file == "Packages.gz":
f = GzipFile(fileobj=f, mode="rb")
- a check if the received a 404 for Packages.gz, issuing the fallback
to Packages in this case
Note that when ssh is used to fetch the file, we currently do not
request Packages.gz as it may not exist and the fallback is
non-trivial in this case.
Bug: https://bugs.gentoo.org/248089
Closes: https://bugs.gentoo.org/962719
Signed-off-by: Florian Schmaus <flow <AT> gentoo.org>
Part-of: https://github.com/gentoo/portage/pull/1454
Closes: https://github.com/gentoo/portage/pull/1454
Signed-off-by: Sam James <sam <AT> gentoo.org>
lib/portage/dbapi/bintree.py | 404 +++++++++++++++++++++++--------------------
1 file changed, 220 insertions(+), 184 deletions(-)
diff --git a/lib/portage/dbapi/bintree.py b/lib/portage/dbapi/bintree.py
index 7152e72eba..6f6074cd56 100644
--- a/lib/portage/dbapi/bintree.py
+++ b/lib/portage/dbapi/bintree.py
@@ -72,6 +72,7 @@ import tempfile
import textwrap
import time
import traceback
+import urllib
import warnings
from gzip import GzipFile
from itertools import chain
@@ -1412,213 +1413,248 @@ class binarytree:
rmt_idx = self._new_pkgindex()
proc = None
tmp_filename = None
- try:
- # urlparse.urljoin() only works correctly with recognized
- # protocols and requires the base url to have a trailing
- # slash, so join manually...
- url = base_url.rstrip("/") + "/Packages"
- f = None
-
- if local_timestamp and (repo.frozen or not getbinpkg_refresh):
- raise UseCachedCopyOfRemoteIndex()
-
+ for remote_pkgindex_file in ("Packages.gz", "Packages"):
try:
- ttl = float(pkgindex.header.get("TTL", 0))
- except ValueError:
- pass
- else:
- if (
- download_timestamp
- and ttl
- and download_timestamp + ttl > time.time()
- ):
+ # urlparse.urljoin() only works correctly with recognized
+ # protocols and requires the base url to have a trailing
+ # slash, so join manually...
+ url = base_url.rstrip("/") + "/" + remote_pkgindex_file
+ f = None
+
+ if local_timestamp and (repo.frozen or not getbinpkg_refresh):
raise UseCachedCopyOfRemoteIndex()
- # Set proxy settings for _urlopen -> urllib_request
- proxies = {}
- for proto in ("http", "https"):
- value = self.settings.get(proto + "_proxy")
- if value is not None:
- proxies[proto] = value
-
- # Don't use urlopen for https, unless
- # PEP 476 is supported (bug #469888).
- if (
- repo.fetchcommand is None or parsed_url.scheme in ("", "file")
- ) and (parsed_url.scheme not in ("https",) or _have_pep_476()):
try:
- if parsed_url.scheme in ("", "file"):
- f = open(f"{parsed_url.path.rstrip('/')}/Packages", "rb")
- else:
- f = _urlopen(
- url, if_modified_since=local_timestamp, proxies=proxies
- )
- if hasattr(f, "headers") and f.headers.get("timestamp", ""):
- remote_timestamp = f.headers.get("timestamp")
- except OSError as err:
+ ttl = float(pkgindex.header.get("TTL", 0))
+ except ValueError:
+ pass
+ else:
if (
- hasattr(err, "code") and err.code == 304
- ): # not modified (since local_timestamp)
+ download_timestamp
+ and ttl
+ and download_timestamp + ttl > time.time()
+ ):
raise UseCachedCopyOfRemoteIndex()
- if parsed_url.scheme in ("ftp", "http", "https"):
- # This protocol is supposedly supported by urlopen,
- # so apparently there's a problem with the url
- # or a bug in urlopen.
- if self.settings.get("PORTAGE_DEBUG", "0") != "0":
- traceback.print_exc()
+ # Set proxy settings for _urlopen -> urllib_request
+ proxies = {}
+ for proto in ("http", "https"):
+ value = self.settings.get(proto + "_proxy")
+ if value is not None:
+ proxies[proto] = value
- raise
- except ValueError:
- raise ParseError(
- f"Invalid Portage BINHOST value '{url.lstrip()}'"
- )
+ # Don't use urlopen for https, unless
+ # PEP 476 is supported (bug #469888).
+ if (
+ repo.fetchcommand is None or parsed_url.scheme in ("", "file")
+ ) and (parsed_url.scheme not in ("https",) or _have_pep_476()):
+ try:
+ if parsed_url.scheme in ("", "file"):
+ f = open(
+ f"{parsed_url.path.rstrip('/')}/{remote_pkgindex_file}",
+ "rb",
+ )
+ else:
+ f = _urlopen(
+ url,
+ if_modified_since=local_timestamp,
+ proxies=proxies,
+ )
+ if hasattr(f, "headers") and f.headers.get(
+ "timestamp", ""
+ ):
+ remote_timestamp = f.headers.get("timestamp")
+ except OSError as err:
+ if (
+ hasattr(err, "code") and err.code == 304
+ ): # not modified (since local_timestamp)
+ raise UseCachedCopyOfRemoteIndex()
+
+ if parsed_url.scheme in ("ftp", "http", "https"):
+ # This protocol is supposedly supported by urlopen,
+ # so apparently there's a problem with the url
+ # or a bug in urlopen.
+ if self.settings.get("PORTAGE_DEBUG", "0") != "0":
+ traceback.print_exc()
+
+ raise
+ except ValueError:
+ raise ParseError(
+ f"Invalid Portage BINHOST value '{url.lstrip()}'"
+ )
- if f is None:
- path = parsed_url.path.rstrip("/") + "/Packages"
-
- if repo.fetchcommand is None and parsed_url.scheme == "ssh":
- # Use a pipe so that we can terminate the download
- # early if we detect that the TIMESTAMP header
- # matches that of the cached Packages file.
- ssh_args = ["ssh"]
- if port is not None:
- ssh_args.append(f"-p{port}")
- # NOTE: shlex evaluates embedded quotes
- ssh_args.extend(
- shlex.split(self.settings.get("PORTAGE_SSH_OPTS", ""))
- )
- ssh_args.append(user_passwd + host)
- ssh_args.append("--")
- ssh_args.append("cat")
- ssh_args.append(path)
+ if f is None:
+ path = parsed_url.path.rstrip("/") + "/" + remote_pkgindex_file
- proc = subprocess.Popen(ssh_args, stdout=subprocess.PIPE)
- f = proc.stdout
- else:
- if repo.fetchcommand is None:
- setting = "FETCHCOMMAND_" + parsed_url.scheme.upper()
- fcmd = self.settings.get(setting)
- if not fcmd:
- fcmd = self.settings.get("FETCHCOMMAND")
- if not fcmd:
- raise OSError("FETCHCOMMAND is unset")
+ if repo.fetchcommand is None and parsed_url.scheme == "ssh":
+ if remote_pkgindex_file == "Packages.gz":
+ # TODO: Check first if Packages.gz exist before
+ # cat'ing it. Until this is done, never try to retrieve
+ # Packages.gz as it is not guaranteed to exist.
+ continue
+
+ # Use a pipe so that we can terminate the download
+ # early if we detect that the TIMESTAMP header
+ # matches that of the cached Packages file.
+ ssh_args = ["ssh"]
+ if port is not None:
+ ssh_args.append(f"-p{port}")
+ # NOTE: shlex evaluates embedded quotes
+ ssh_args.extend(
+ shlex.split(self.settings.get("PORTAGE_SSH_OPTS", ""))
+ )
+ ssh_args.append(user_passwd + host)
+ ssh_args.append("--")
+ ssh_args.append("cat")
+ ssh_args.append(path)
+
+ proc = subprocess.Popen(ssh_args, stdout=subprocess.PIPE)
+ f = proc.stdout
else:
- fcmd = repo.fetchcommand
+ if repo.fetchcommand is None:
+ setting = "FETCHCOMMAND_" + parsed_url.scheme.upper()
+ fcmd = self.settings.get(setting)
+ if not fcmd:
+ fcmd = self.settings.get("FETCHCOMMAND")
+ if not fcmd:
+ raise OSError("FETCHCOMMAND is unset")
+ else:
+ fcmd = repo.fetchcommand
- fd, tmp_filename = tempfile.mkstemp()
- tmp_dirname, tmp_basename = os.path.split(tmp_filename)
- os.close(fd)
+ fd, tmp_filename = tempfile.mkstemp()
+ tmp_dirname, tmp_basename = os.path.split(tmp_filename)
+ os.close(fd)
- fcmd_vars = {
- "DISTDIR": tmp_dirname,
- "FILE": tmp_basename,
- "URI": url,
- }
+ fcmd_vars = {
+ "DISTDIR": tmp_dirname,
+ "FILE": tmp_basename,
+ "URI": url,
+ }
- for k in ("PORTAGE_SSH_OPTS",):
- v = self.settings.get(k)
- if v is not None:
- fcmd_vars[k] = v
+ for k in ("PORTAGE_SSH_OPTS",):
+ v = self.settings.get(k)
+ if v is not None:
+ fcmd_vars[k] = v
- success = portage.getbinpkg.file_get(
- fcmd=fcmd, fcmd_vars=fcmd_vars
- )
- if not success:
- raise OSError(f"{setting} failed")
- f = open(tmp_filename, "rb")
+ success = portage.getbinpkg.file_get(
+ fcmd=fcmd, fcmd_vars=fcmd_vars
+ )
+ if not success:
+ if remote_pkgindex_file == "Packages.gz":
+ # Ignore failures for Packages.gz, as the file is
+ # not guaranteed to exist.
+ continue
+ raise OSError(f"{setting} failed")
+ f = open(tmp_filename, "rb")
- f_dec = codecs.iterdecode(
- f, _encodings["repo.content"], errors="replace"
- )
- try:
- rmt_idx.readHeader(f_dec)
- if (
- not remote_timestamp
- ): # in case it had not been read from HTTP header
- remote_timestamp = rmt_idx.header.get("TIMESTAMP", None)
- if not remote_timestamp:
- # no timestamp in the header, something's wrong
- pkgindex = None
- writemsg(
- _(
- "\n\n!!! Binhost package index "
- " has no TIMESTAMP field.\n"
- ),
- noiselevel=-1,
- )
- else:
- if not self._pkgindex_version_supported(rmt_idx):
+ if remote_pkgindex_file == "Packages.gz":
+ f = GzipFile(fileobj=f, mode="rb")
+
+ f_dec = codecs.iterdecode(
+ f, _encodings["repo.content"], errors="replace"
+ )
+ try:
+ rmt_idx.readHeader(f_dec)
+ if (
+ not remote_timestamp
+ ): # in case it had not been read from HTTP header
+ remote_timestamp = rmt_idx.header.get("TIMESTAMP", None)
+ if not remote_timestamp:
+ # no timestamp in the header, something's wrong
+ pkgindex = None
writemsg(
_(
- "\n\n!!! Binhost package index version"
- " is not supported: '%s'\n"
- )
- % rmt_idx.header.get("VERSION"),
+ "\n\n!!! Binhost package index "
+ " has no TIMESTAMP field.\n"
+ ),
noiselevel=-1,
)
- pkgindex = None
- elif not local_timestamp or int(local_timestamp) < int(
- remote_timestamp
- ):
- rmt_idx.readBody(f_dec)
- pkgindex = rmt_idx
- finally:
- # Timeout after 5 seconds, in case close() blocks
- # indefinitely (see bug #350139).
- try:
+ else:
+ if not self._pkgindex_version_supported(rmt_idx):
+ writemsg(
+ _(
+ "\n\n!!! Binhost package index version"
+ " is not supported: '%s'\n"
+ )
+ % rmt_idx.header.get("VERSION"),
+ noiselevel=-1,
+ )
+ pkgindex = None
+ elif not local_timestamp or int(local_timestamp) < int(
+ remote_timestamp
+ ):
+ rmt_idx.readBody(f_dec)
+ pkgindex = rmt_idx
+ finally:
+ # Timeout after 5 seconds, in case close() blocks
+ # indefinitely (see bug #350139).
try:
- AlarmSignal.register(5)
- f.close()
- finally:
- AlarmSignal.unregister()
- except AlarmSignal:
- writemsg(
- "\n\n!!! %s\n"
- % _("Timed out while closing connection to binhost"),
- noiselevel=-1,
- )
- except UseCachedCopyOfRemoteIndex:
- changed = False
- rmt_idx = pkgindex
- if getbinpkg_refresh or repo.frozen:
- desc = "frozen" if repo.frozen else "up-to-date"
- writemsg_stdout("\n")
- writemsg_stdout(
- colorize(
- "GOOD",
- _("Local copy of remote index is %s and will be used.")
- % desc,
+ try:
+ AlarmSignal.register(5)
+ f.close()
+ finally:
+ AlarmSignal.unregister()
+ except AlarmSignal:
+ writemsg(
+ "\n\n!!! %s\n"
+ % _("Timed out while closing connection to binhost"),
+ noiselevel=-1,
+ )
+ break
+ except UseCachedCopyOfRemoteIndex:
+ changed = False
+ rmt_idx = pkgindex
+ if getbinpkg_refresh or repo.frozen:
+ desc = "frozen" if repo.frozen else "up-to-date"
+ writemsg_stdout("\n")
+ writemsg_stdout(
+ colorize(
+ "GOOD",
+ _("Local copy of remote index is %s and will be used.")
+ % desc,
+ )
+ + "\n"
)
- + "\n"
+ except OSError as e:
+ if (
+ remote_pkgindex_file == "Packages.gz"
+ and isinstance(e, urllib.error.HTTPError)
+ and e.code == 404
+ ):
+ # Ignore 404s for Packages.gz, as the file is
+ # not guaranteed to exist.
+ continue
+
+ # This includes URLError which is raised for SSL
+ # certificate errors when PEP 476 is supported.
+ writemsg(
+ _("\n\n!!! Error fetching binhost package" " info from '%s'\n")
+ % _hide_url_passwd(base_url)
)
- except OSError as e:
- # This includes URLError which is raised for SSL
- # certificate errors when PEP 476 is supported.
- writemsg(
- _("\n\n!!! Error fetching binhost package" " info from '%s'\n")
- % _hide_url_passwd(base_url)
- )
- # With Python 2, the EnvironmentError message may
- # contain bytes or unicode, so use str to ensure
- # safety with all locales (bug #532784).
- try:
- error_msg = str(e)
- except UnicodeDecodeError as uerror:
- error_msg = str(uerror.object, encoding="utf_8", errors="replace")
- writemsg(f"!!! {error_msg}\n\n")
- del e
- pkgindex = None
- if proc is not None:
- if proc.poll() is None:
- proc.kill()
- proc.wait()
- proc = None
- if tmp_filename is not None:
- try:
- os.unlink(tmp_filename)
- except OSError:
- pass
+ # With Python 2, the EnvironmentError message may
+ # contain bytes or unicode, so use str to ensure
+ # safety with all locales (bug #532784).
+ try:
+ error_msg = str(e)
+ except UnicodeDecodeError as uerror:
+ error_msg = str(
+ uerror.object, encoding="utf_8", errors="replace"
+ )
+ writemsg(f"!!! {error_msg}\n\n")
+ del e
+ pkgindex = None
+ finally:
+ if proc is not None:
+ if proc.poll() is None:
+ proc.kill()
+ proc.wait()
+ proc = None
+ if tmp_filename is not None:
+ try:
+ os.unlink(tmp_filename)
+ except OSError:
+ pass
+
if pkgindex is rmt_idx and changed:
pkgindex.modified = False # don't update the header
pkgindex.header["DOWNLOAD_TIMESTAMP"] = "%d" % time.time()
next reply other threads:[~2025-09-11 18:08 UTC|newest]
Thread overview: 99+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-09-11 18:07 Sam James [this message]
-- strict thread matches above, loose matches on Subject: below --
2025-11-03 15:59 [gentoo-commits] proj/portage:master commit in: lib/portage/dbapi/ Sam James
2025-11-03 15:59 Sam James
2025-11-01 19:54 Zac Medico
2025-11-01 9:49 Sam James
2025-11-01 9:49 Sam James
2025-10-30 3:06 Zac Medico
2025-10-28 5:27 Zac Medico
2025-09-16 15:40 Sam James
2025-09-15 12:21 Sam James
2025-09-15 12:21 Sam James
2025-09-15 11:14 Sam James
2025-09-15 11:14 Sam James
2025-09-15 11:14 Sam James
2025-09-15 11:14 Sam James
2025-09-14 9:33 Sam James
2025-09-14 0:15 Sam James
2025-09-14 0:15 Sam James
2025-09-12 21:18 Sam James
2025-09-12 21:18 Sam James
2025-09-11 20:37 Sam James
2025-09-11 11:27 Sam James
2025-09-11 3:17 Sam James
2025-09-11 0:07 Sam James
2025-06-24 1:09 Sam James
2025-06-17 2:58 Sam James
2025-01-24 21:29 Zac Medico
2025-01-09 23:35 Zac Medico
2024-06-01 19:20 Zac Medico
2024-05-27 18:13 Zac Medico
2024-03-02 22:55 Zac Medico
2024-02-25 8:25 Sam James
2024-01-16 7:52 Sam James
2024-01-16 5:26 Zac Medico
2024-01-16 5:16 Sam James
2024-01-16 5:16 Sam James
2024-01-03 5:57 Zac Medico
2023-12-10 1:28 Zac Medico
2023-10-24 18:37 Zac Medico
2023-10-23 14:28 Zac Medico
2023-10-22 21:30 Zac Medico
2023-10-22 20:58 Zac Medico
2023-10-22 15:53 Zac Medico
2023-10-20 0:34 Zac Medico
2023-10-15 22:02 Zac Medico
2023-10-08 19:48 Zac Medico
2023-10-05 5:45 Zac Medico
2023-10-04 4:29 Zac Medico
2023-09-26 21:09 Sam James
2023-09-23 22:49 Sam James
2023-09-23 22:38 Sam James
2023-09-23 22:31 Sam James
2023-09-20 18:02 Mike Gilbert
2023-09-15 4:28 Sam James
2023-07-29 3:57 Sam James
2023-05-23 0:26 Sam James
2023-05-23 0:26 Sam James
2023-05-23 0:26 Sam James
2022-12-21 1:30 Sam James
2022-12-21 1:28 Sam James
2022-11-08 23:07 Sam James
2022-09-25 19:12 Mike Gilbert
2022-09-09 10:16 Michał Górny
2022-08-18 19:00 Mike Gilbert
2022-04-20 20:24 Zac Medico
2022-04-13 15:34 Sam James
2022-04-13 15:34 Sam James
2021-11-26 21:09 Mike Gilbert
2021-09-21 5:51 Zac Medico
2021-09-21 5:51 Zac Medico
2021-09-21 5:51 Zac Medico
2021-06-05 18:08 Zac Medico
2021-03-07 11:42 Zac Medico
2021-02-23 21:31 Zac Medico
2021-01-18 9:20 Zac Medico
2021-01-17 13:31 Zac Medico
2021-01-17 8:49 Zac Medico
2021-01-17 8:49 Zac Medico
2020-09-08 2:52 Zac Medico
2020-08-09 0:15 Zac Medico
2020-08-04 3:16 Zac Medico
2020-08-03 23:28 Zac Medico
2020-08-03 23:28 Zac Medico
2020-08-03 21:42 Zac Medico
2020-08-03 21:42 Zac Medico
2020-08-03 19:30 Zac Medico
2020-07-22 20:14 Zac Medico
2020-07-22 19:52 Zac Medico
2020-07-22 17:46 Zac Medico
2020-06-07 3:26 Zac Medico
2020-02-20 9:55 Zac Medico
2020-02-03 3:04 Zac Medico
2019-08-24 3:15 Zac Medico
2019-06-20 19:43 Zac Medico
2019-05-11 21:16 Zac Medico
2019-01-20 6:55 Zac Medico
2019-01-11 10:14 Fabian Groffen
2018-09-24 7:30 Zac Medico
2018-09-24 0:46 Zac Medico
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1757614014.443d5765431e3d9ada2bd9b940ca5c14040a06e6.sam@gentoo \
--to=sam@gentoo.org \
--cc=gentoo-commits@lists.gentoo.org \
--cc=gentoo-dev@lists.gentoo.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox