Eric Wong [Sun, 20 Jan 2019 12:33:20 +0000 (12:33 +0000)]
extmsg: don't bother partial matching with <16 chars
It's not worth it, and attempts to wildcard off
single-character Message-IDs(*) causes Xapian to error
out in unpredictable ways:
something terrible happened at /usr/lib/x86_64-linux-gnu/perl5/5.24/Search/Xapian/Enquire.pm line 54.
...propagated at lib/PublicInbox/Search.pm line 209.
So don't bother.
(*) because people blindly hit 'y' or 'n' when git-send-email
prompted them for In-Reply-To.
Eric Wong [Fri, 18 Jan 2019 12:50:59 +0000 (12:50 +0000)]
git: git_unquote handles double-quote and backslash
We need to work with 0x22 (double-quote) and 0x5c (backslash);
even if they're oddball characters in filenames which wouldn't
be used by projects I'd want to work on.
Eric Wong [Fri, 18 Jan 2019 19:40:07 +0000 (19:40 +0000)]
t/git.t: avoid passing read-only value to git_unquote
Older versions of Perl (tested 5.14.2 on Debian wheezy(*),
reported by Konstantin on Perl 5.16.3) considered the result of
concatenating two string literals to be a constant value.
(*) not that other stuff works on wheezy, but t/git.t should.
Reported-by: Konstantin Ryabitsev <konstantin@linuxfoundation.org>
Eric Wong [Wed, 16 Jan 2019 04:51:28 +0000 (04:51 +0000)]
nntp: header responses use CRLF consistently
Alpine is apparently stricter than other clients I've tried
w.r.t. using CRLF for headers. So do the same thing we do for
bodies to ensure we only emit CRLFs and no bare LFs.
Reported-by: Wang Kang <i@scateu.me>
https://public-inbox.org/meta/alpine.DEB.2.21.99.1901161043430.29788@la.scateu.me/
Eric Wong [Wed, 9 Jan 2019 11:43:26 +0000 (11:43 +0000)]
config: inbox name checking matches git.git more closely
Actually, it turns out git.git/remote.c::valid_remote_nick
rules alone are insufficient. More checking is performed as
part of the refname in the git.git/refs.c::check_refname_component
I also considered rejecting URL-unfriendly inbox names entirely,
but realized some users may intentionally configure names not
handled by our WWW endpoint for archives they don't want
accessible over HTTP.
Eric Wong [Tue, 15 Jan 2019 02:42:09 +0000 (02:42 +0000)]
git_unquote: perform modifications in-place
This function doesn't have a lot of callers at the moment so
none of them are affected by this change. But the plan is to
use this in our WWW code for things, so do it now before we
call it in more places.
Results from a Thinkpad X200 with a Core2Duo P8600 @ 2.4GHz:
Note: I mainly care about unquoted performance because
that's the common case for the target audience of public-inbox.
Script used to get benchmark results against the Linux source tree:
==> bench_unquote.perl <==
use strict;
use warnings;
use Benchmark ':hireswallclock';
my $nr = 50;
my %GIT_ESC = (
a => "\a",
b => "\b",
f => "\f",
n => "\n",
r => "\r",
t => "\t",
v => "\013",
);
Eric Wong [Thu, 10 Jan 2019 21:41:55 +0000 (21:41 +0000)]
Merge commit 'mem'
* commit 'mem':
view: more culling for search threads
over: cull unneeded fields for get_thread
searchmsg: remove unused fields for PSGI in Xapian results
searchview: drop unused {seen} hashref
searchmsg: remove Xapian::Document field
searchmsg: get rid of termlist scanning for mid
httpd: remove psgix.harakiri reference
Eric Wong [Thu, 10 Jan 2019 03:26:15 +0000 (03:26 +0000)]
t/v2writable.t: force more consistent "git log" output
This should probably use lower-level git plumbing, but until
then, consistently add a bunch of --no-* options to "git log"
to get more consistent output.
Noticed-by: Johannes Berg
https://public-inbox.org/meta/1538164205.14416.76.camel@sipsolutions.net/
Eric Wong [Tue, 8 Jan 2019 11:13:33 +0000 (11:13 +0000)]
over: cull unneeded fields for get_thread
On a certain ugly /$INBOX/$MESSAGE_ID/T/ endpoint with 1000
messages in the thread, this cuts memory usage from 2.5M to 1.9M
(which still isn't great, but it's a start).
Eric Wong [Tue, 8 Jan 2019 11:13:31 +0000 (11:13 +0000)]
searchmsg: remove unused fields for PSGI in Xapian results
These fields are only necessary in NNTP and not even stored in
Xapian; so keeping them around for the PSGI web UI search
results wastes nearly 80K when loading large result sets.
Eric Wong [Tue, 8 Jan 2019 11:13:26 +0000 (11:13 +0000)]
httpd: remove psgix.harakiri reference
We don't need to set "psgix." extension fields for things
we don't support. This saves 138 bytes per-client in $env
as measured by Devel::Size::total_size
Eric Wong [Tue, 8 Jan 2019 00:41:12 +0000 (00:41 +0000)]
view: stop storing all MIME objects on large threads
While we try to discard the $smsg (SearchMsg) objects quickly,
they remain referenced via $node (SearchThread::Msg) objects,
which are stored forever in $ctx->{mapping} to cull redundant
words out of subjects in the thread skeleton.
This significantly cuts memory bloat with large search results
with '&x=t'. Now, the search results overhead of
SearchThread::Msg and linked objects are stable at around 350K
instead of ~7M per response in a rough test (there's more
savings to be had in the same areas).
Several hundred kilobytes is still huge and a large per-client
cost; but it's far better than MEGABYTES per-client.
Eric Wong [Sat, 5 Jan 2019 10:41:15 +0000 (10:41 +0000)]
filter/rubylang: fix SQLite DB lifetime problems
Clearly the AltId stuff was never tested for v2. Ensure
this tricky filter (which reuses Msgmap to avoid introducing
new serial numbers) doesn't trigger deadlocks SQLite due
to opening a DB for writing multiple times.
I went through several iterations of this change before
going with this one, which is the least intrusive I could
fine.
Eric Wong [Fri, 4 Jan 2019 11:53:02 +0000 (11:53 +0000)]
t/cgi.t: remove more redundant tests
Most of these test cases are in t/plack.t, already; and that
runs much faster. Just ensure the slashy corner case and search
stuff works. While we're at it, avoid using the
public-inbox-index command and just use the internal API to
index.
Eric Wong [Wed, 2 Jan 2019 00:50:55 +0000 (00:50 +0000)]
t/v2reindex: use the larger text to increase test reliability
libxapian30:amd64 1.4.9-1 on Debian sid seems to give an 8KB
position.glass database with "hello world" as the document
regardless of our indexlevel. Use the text of the AGPL-3.0 for
a more realisitic Xapian database size.
And perhaps tying our tests to the AGPL will make life more
difficult for would-be copyright violators :>
With the mboxrd downloaded, mutt is able to view them without
difficulty.
Note: this change would require reindexing of Xapian to pick up
the changes. But it's only two ancient messages, the first was
resent by the original sender and the second is too old to be
relevant.
Eric Wong [Wed, 26 Dec 2018 09:07:49 +0000 (09:07 +0000)]
tests: consolidate process spawning code.
IPC::Run provides a nice simplification in several places; and
we already use it (optionally) on a lot of tests.
For the non-test code, we still rely on our vfork-capable
Inline::C stuff since real-world server processes can get large
enough to where vfork is an advantage. Maybe Perl5 can use
CLONE_VFORK somehow, one day:
Eric Wong [Fri, 28 Dec 2018 06:22:55 +0000 (06:22 +0000)]
examples/cgit-commit-filter.lua: update URLs
Let's Encrypt is working out nicely, so we can rely on HTTPS,
now. Use 80x24.org instead of bogomips.org while we're at it,
since I don't think the latter will remain.
Eric Wong [Fri, 28 Dec 2018 10:16:11 +0000 (10:16 +0000)]
init: allow --skip of old epochs for -V2 repos
This allows archivists to publish incomplete archives with newer
mail while allowing "0.git" (or "1.git" and so on) epochs to be
added-after-the-fact (without affecting "git clone" followers).
A reindex will be necessary for Xapian and SQLite to catch up
once the old epochs are added; but the reindexing code is also
capable of tolerating missing epochs.
Eric Wong [Wed, 12 Dec 2018 23:18:13 +0000 (23:18 +0000)]
doc/hosted: add glibc and bug-gnulib mirrors
These have existed for a while, actually, so, we might as well
publicize them. While we're at it, add a disclaimer to
discourage reliance on single points of failure.
Eric Wong [Thu, 6 Dec 2018 02:40:06 +0000 (02:40 +0000)]
nntp: prevent event_read from firing twice in a row
When a client starts pipelining requests to us which trigger
long responses, we need to keep socket readiness checks disabled
and only enable them when our socket rbuf is drained.
Failure to do this caused aborted clients with
"BUG: nested long response" when Danga::Socket calls event_read
for read-readiness after our "next_tick" sub fires in the
same event loop iteration.
Reported-by: Jonathan Corbet <corbet@lwn.net>
cf. https://public-inbox.org/meta/20181013124658.23b9f9d2@lwn.net/
Jonathan Corbet [Sat, 13 Oct 2018 21:42:21 +0000 (15:42 -0600)]
Add Xrefs to over/xover lines
Putting the Xref field into xover lines allows newsreaders to mark
cross-posted messages read when catching up a group. That, in turn,
massively improves the life of crazy people who try to follow dozens of
kernel lists, where emails are often heavily cross-posted.
Jonathan Corbet [Sat, 13 Oct 2018 21:42:20 +0000 (15:42 -0600)]
Put the NNTP server name into Xref lines
RFC 5536 sec 3.2.14 says that the server-name in an Xref line is "which
news server generated the header field"; indeed, that is necessary for
newsreaders like gnus to handle references properly. So pick up the server
name from the config if available (the first name if there's more than
one), from the host name otherwise, and use it rather than the domain
name of the list server.
Tests have been adjusted to match the new behavior.
Eric W. Biederman [Fri, 10 Aug 2018 00:08:22 +0000 (19:08 -0500)]
Import.pm: When purging replace a purged file with a zero length file
This ensures that the number of added files remains the same and thus
the article numbers derived from a repository will remain the same.
I think this is the last place in public-inbox that has to be tweaked to
guarantee the generated article number will remain the same in an public
inbox archive.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Eric Wong [Sun, 5 Aug 2018 08:19:25 +0000 (08:19 +0000)]
overidx: preserve `tid' column on re-indexing
Otherwise, walking backwards through history could mean the root
message in a thread forgets its `tid' and it prevents messages
from being looked up by it.
This bug was hidden by the fact that `sid' matches were often
good enough to link threads together.
Eric Wong [Sun, 5 Aug 2018 06:04:40 +0000 (06:04 +0000)]
view: distinguish strict and loose thread matches
The "loose" (Subject:-based) thread matching yields too many
hits for some common subjects (e.g. "[GIT] Networking" on LKML)
and causes thread skeletons to not show the current messages.
Favor strict matches in the query and only add loose matches
if there's space.
While working on this, I noticed the backwards --reindex walk
breaks `tid' on v1 repositories, at least. That bug was hidden
by the Subject: match logic and not discovered until now. It
will be fixed separately.
Reported-by: Konstantin Ryabitsev <konstantin@linuxfoundation.org>
Eric Wong [Fri, 3 Aug 2018 20:05:24 +0000 (20:05 +0000)]
Merge branch 'eb/index-incremental'
Incremental indexing fixes from Eric W. Biederman.
These prevents the highest message number in msgmap from
being reassigned after deletes in rare cases and ensures
messages are deleted from msgmap in v2.
* eb/index-incremental:
V2Writeable.pm: In unindex_oid delete the message from msgmap
V2Writeable.pm: Ensure that a found message number is in the msgmap
SearchIdx,V2Writeable: Update num_highwater on optimized deletes
t/v[12]reindex.t: Verify the num highwater is as expected
t/v[12]reindex.t Verify num_highwater
Msgmap.pm: Track the largest value of num ever assigned
SearchIdx.pm: Always assign numbers backwards during incremental indexing
t/v[12]reindex.t: Test incremental indexing works
t/v[12]reindex.t: Test that the resulting msgmap is as expected
t/v[12]reindex.t: Place expected second in Xapian tests
t/v2reindex.t: Isolate the test cases more
t/v1reindex.t: Isolate the test cases
Import.pm: Don't assume {in} and {out} always exist
Eric W. Biederman [Wed, 1 Aug 2018 16:43:44 +0000 (11:43 -0500)]
V2Writeable.pm: In unindex_oid delete the message from msgmap
Now that we track the num highwater mark it is safe to remove messages
from msgmap that have been previously allocated. Removing even the
highest numbered article will no longer cause new message numbers to
move backwards.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Eric W. Biederman [Wed, 1 Aug 2018 16:43:43 +0000 (11:43 -0500)]
V2Writeable.pm: Ensure that a found message number is in the msgmap
The lookup to see if a num has already been assigned to a message
happens in a temporary copy of message map. It is possible that the
number has been removed from the current message map. The
unindex/reindex after a history rewrite triggered by a purge should be
one such case. Therefore add the number to the msgmap in case it is
not currently present.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Eric W. Biederman [Wed, 1 Aug 2018 16:43:42 +0000 (11:43 -0500)]
SearchIdx,V2Writeable: Update num_highwater on optimized deletes
When performing an incremental index update with index_sync if a message is seen
to be both added and deleted update the num_highwater mark even though the
message is not otherwise indexed.
This ensures index_sync generates the same msgmap no matter which commit
it stops at during incremental syncs.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Eric W. Biederman [Wed, 1 Aug 2018 16:43:39 +0000 (11:43 -0500)]
Msgmap.pm: Track the largest value of num ever assigned
Today the only thing that prevents public-inbox not reusing the
message numbers of deleted messages is the sqlite autoincrement magic
and that only works part of the time. The new incremental indexing
test has revealed areas where today public-inbox does try to reuse
numbers of deleted messages.
Reusing the message numbers of existing messages is a problem because
if a client ever sees messages that are subsequently deleted the
client will not see the new messages with their old numbers.
In practice this is difficult to trigger because it requires the most
recently added message to be removed and have the removal show up in a
separate pull request. Still it can happen and it should be handled.
Instead of infering the highset number ever used by finding the maximum
number in the message map, track the largest number ever assigned directly.
Update Msgmap to track this value and update the indexers to use this
value.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Eric Wong [Mon, 30 Jul 2018 08:23:51 +0000 (08:23 +0000)]
search: (really) match the behavior of WWW for indexing text
Not sure what was going through my mind when I made my first
attempt at this, but we really want to make sure we index all
the text we display in the web view (and presumably anything a
reasonable mail client can display).
Followup-to: 0cf6196025d4e4880cd1ed859257ce21dd3cdcf6
("search: match the behavior of WWW for indexing text")
Eric W. Biederman [Wed, 1 Aug 2018 16:43:38 +0000 (11:43 -0500)]
SearchIdx.pm: Always assign numbers backwards during incremental indexing
When walking messages newest to oldest, assigning the larger numbers
before smaller numbers ensures older messages get smaller numbers.
This leads to the possibility of a msgmap that can be regenerated when
needed.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Eric W. Biederman [Wed, 1 Aug 2018 16:43:37 +0000 (11:43 -0500)]
t/v[12]reindex.t: Test incremental indexing works
Capture interesting commits of the test repository in mark variables.
Use those marks to build interesting scenarios where index_sync proceeds
as if those marks are the heads of the repositor. Use this capability to
test what happens when adds and deletes are mixed within a repository.
Be sad because things don't yet work as they should.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Eric W. Biederman [Wed, 1 Aug 2018 16:43:34 +0000 (11:43 -0500)]
t/v2reindex.t: Isolate the test cases more
While inspecting the tests I realized that because we have been
reusing variables there can be a memory between one test case and
another. Add scopes and local variables to prevent an unintended
memory between one test cases.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Eric W. Biederman [Wed, 1 Aug 2018 16:43:33 +0000 (11:43 -0500)]
t/v1reindex.t: Isolate the test cases
While inspecting the tests I realized that because we have been
reusing variables there can be a memory between one test case and
another. Add scopes and local variables to prevent an unintended
memory between one test and another.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Eric W. Biederman [Wed, 1 Aug 2018 16:43:32 +0000 (11:43 -0500)]
Import.pm: Don't assume {in} and {out} always exist
While working on one of the tests I did:
my $im = PublicInbox::V2Writable->new($ibx, 1);
my $im0 = $im->importer();
$im->add($mime);
Which resulted in a warning of the use of an undefined value from
atfork_child, and the test failing nastily. Inspection of the code
reveals this can happen anytime gfi_start has not been called.
So just fix atfork_child to skip closing file descriptors that have
not yet been setup.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Eric W. Biederman [Mon, 30 Jul 2018 05:04:45 +0000 (00:04 -0500)]
ProcessPipe.pm: Use read not sysread
While playing with git fast export I discovered that mixing <> and
read would give inconsistent results. I tracked the issue down to
using sysread in ProcessPipe instead of plain read.
If it is desirable to use readline I can't see how using sysread
can work as readline to be efficient needs to use buffered I/O.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Eric Wong [Sun, 29 Jul 2018 10:05:13 +0000 (10:05 +0000)]
mda: allow configuring globally without spamc support
This reuses some of the configuration from -watch, but remains
independent since some configurations will use -watch for some
inboxes and -mda for others.
The default remains "spamc" for -mda users so nothing changes
without explicit configuration.
Per-inbox configurations may also be supported in the future.
Eric Wong [Sun, 29 Jul 2018 09:34:41 +0000 (09:34 +0000)]
mda: v2: ensure message bodies are indexed
We must not clobber the original message string, as Email::MIME(*)
still needs it for iterating through parts in SearchIdx (but not
when handing it as a raw string to git-fast-import).
I've noticed message bodies (especially dfpre/dpost) were not
getting indexed when going through -mda (no problems with
-watch). This also did not affect v1 repos, since indexing is a
separate process for v1 and requires re-reading the data from
git.
Eric Wong [Fri, 20 Jul 2018 06:16:12 +0000 (06:16 +0000)]
search: use boolean prefixes for git blob queries
I've hit some case where probabilistic searches don't work when
using dfpre:/dfpost:/dfblob: search prefixes because stemming in
the query parser interferes.
In any case, our indexing code indexes longer/unabbreviated blob
names down to its 7 character abbreviation, so there should be
no need to do wildcard searches on git blob names.
Eric Wong [Thu, 19 Jul 2018 21:43:10 +0000 (21:43 +0000)]
tests: fixup indexlevel setting in tests
The correct field is underscore-less for consistency with
git-config naming conventions. While we're at it, beef up
the v2 tests with actual size checks, too.
I also noticed phrase searching still seems to work for
the limited test case, so I left it documented; but the
size checking verifies the space savings.
Eric W. Biederman [Thu, 19 Jul 2018 19:36:31 +0000 (14:36 -0500)]
Import.pm: Deal with potentially missing From and Sender headers
Use ||= '' to ensure that if the From or Sender header is not present
the code sees an empty string and instead of undefined.
I had some email messages with a From field without an @ (because the
sender was local) and without a Sender which were causing errors when
imported. I think this was bad enough that the email messages were
failing to be imported.
Signed-off-by: Eric Biederamn <ebiederm@xmission.com>
Eric Wong [Thu, 19 Jul 2018 03:21:38 +0000 (03:21 +0000)]
searchidx: respect XAPIAN_FLUSH_THRESHOLD env if set
Xapian documents and respect XAPIAN_FLUSH_THRESHOLD to define
the interval in documents to flush, so don't override it with
our own BATCH_BYTES. This is helpful for initial indexing for
those on slower storage but enough RAM.
It is unnecessary for -watch and frequent incremental indexing;
and it increases transaction times if -watch is playing "catch-up"
if it was stopped for a while.
The original BATCH_BYTES was tuned for a machine with little
memory as the default XAPIAN_FLUSH_THRESHOLD of 10000 documents
was causing swap storms. Using document counts also proved an
innaccurate estimator of RAM usage compared to the actual bytes
processed.