]> Sergey Matveev's repositories - public-inbox.git/log
public-inbox.git
7 years agoremove incorrect comment about strftime + locales
Eric Wong [Sat, 7 Jan 2017 01:44:50 +0000 (01:44 +0000)]
remove incorrect comment about strftime + locales

We only need strftime to be locale-independent when generating
dates for email and HTTP headers.  Purely numeric dates can
use strftime for ease-of-readability.

7 years agoconfig: allow per-inbox nntpserver
Eric Wong [Sat, 7 Jan 2017 01:44:49 +0000 (01:44 +0000)]
config: allow per-inbox nntpserver

This allows certain inboxes to override the global nntpserver
(perhaps under a different domain).

7 years agoinbox: eliminate weaken usage entirely
Eric Wong [Sat, 7 Jan 2017 01:44:48 +0000 (01:44 +0000)]
inbox: eliminate weaken usage entirely

We can do a better job initializing the data structure
so we no longer need to rely on weak references to cleanup
when we ditch the config on reload.

7 years agoinbox: describe the full key name
Eric Wong [Sat, 7 Jan 2017 01:44:47 +0000 (01:44 +0000)]
inbox: describe the full key name

Hopefully make this easier for future generations to understand.

7 years agoconfig: remove unused get() method
Eric Wong [Sat, 7 Jan 2017 01:44:46 +0000 (01:44 +0000)]
config: remove unused get() method

This seems like an unnecessary abstraction, or an abstraction
on the wrong level.

7 years agoconfig: always use namespaced "publicinboxlimiter"
Eric Wong [Sat, 7 Jan 2017 01:44:45 +0000 (01:44 +0000)]
config: always use namespaced "publicinboxlimiter"

I'm not sure if we'll ever support sharing a config file
with other tools, but maybe we will, and "limiter" is
too generic.

7 years agoqspawn: prepare to support runtime reloading of Limiter
Eric Wong [Sat, 7 Jan 2017 01:44:44 +0000 (01:44 +0000)]
qspawn: prepare to support runtime reloading of Limiter

We may allow the {max} value of a limiter to be changed
in the future, so lets start accounting for it before we
spawn followup processes.

7 years agohttp: remove weaken usage, reduce anonsub capture scope
Eric Wong [Wed, 4 Jan 2017 11:20:51 +0000 (11:20 +0000)]
http: remove weaken usage, reduce anonsub capture scope

Avoiding weaken here is no more dangerous than the existing
circular refs (e.g. psgix.io) we create and manage throughout
the lifetime of the connection.  So, trust ourselves to maintain
the data structure properly and avoid triggering extra memory
usage.

While we're at it, avoid having anonymous subroutines capture
more variables than necessary to simplify reference auditing.

7 years agohttpd/async: remove weaken usage
Eric Wong [Wed, 4 Jan 2017 11:20:50 +0000 (11:20 +0000)]
httpd/async: remove weaken usage

We do not need to use weaken() here, so avoid it to simplify our
interactions with Perl; as weaken requires additional storage
and (it seems) time complexity.

7 years agohttp: fix spelling error
Eric Wong [Wed, 4 Jan 2017 11:20:49 +0000 (11:20 +0000)]
http: fix spelling error

Oops.  And we'll be fixing circular references from now...

7 years agowatch: watchspam affects all configured inboxes
Eric Wong [Mon, 2 Jan 2017 13:16:15 +0000 (13:16 +0000)]
watch: watchspam affects all configured inboxes

If a message is spam in one mailbox, it is spam in all others a
particular user/group will care about.

7 years agodoc: minor updates to design notes
Eric Wong [Mon, 26 Dec 2016 21:41:15 +0000 (21:41 +0000)]
doc: minor updates to design notes

ssoma is not worth marketing, but perhaps our mirror of
the git mailing list archives is...

7 years agoevcleanup: ensure deferred close from timers are handled ASAP
Eric Wong [Mon, 26 Dec 2016 03:05:15 +0000 (03:05 +0000)]
evcleanup: ensure deferred close from timers are handled ASAP

Danga::Socket defers close() syscalls until the end of the event
loop to avoid FD recycling.  Unfortunately, this is dependent on
IO events firing and waking the process up from
poll/kevent/epoll_wait.

Without any I/O activity, a socket could remain in the
@Danga::Socket::ToClose array indefinitely.  Thus, we will
trigger a fake IO event after running all timers to trigger
the deferred close in Danga::Socket::PostEventLoop.

7 years agohttpd/async: improve variable naming
Eric Wong [Sun, 25 Dec 2016 08:09:48 +0000 (08:09 +0000)]
httpd/async: improve variable naming

We only refer to PublicInbox::HTTP objects here, so '$io'
was a bad name.

7 years agogithttpbackend: minor cleanups to improve readability
Eric Wong [Sun, 25 Dec 2016 07:33:02 +0000 (07:33 +0000)]
githttpbackend: minor cleanups to improve readability

Fewer returns improves readability and the diffstat agrees.

7 years agogithttpbackend: simplify compatibility code
Eric Wong [Sun, 25 Dec 2016 06:52:03 +0000 (06:52 +0000)]
githttpbackend: simplify compatibility code

Fewer conditionals means theres fewer code paths to test
and makes things easier-to-read.

7 years agogithttpbackend: minor readability improvement
Eric Wong [Sun, 25 Dec 2016 06:39:13 +0000 (06:39 +0000)]
githttpbackend: minor readability improvement

Use a more meaningful variable name for the Qspawn
object, since this module is the reference for its
use.

7 years agohttp: fix clobbering of $null_io
Eric Wong [Sun, 25 Dec 2016 09:40:25 +0000 (09:40 +0000)]
http: fix clobbering of $null_io

Oops, this would be disatrous if we started handling
bigger request bodies or slow clients.

Fixes: c008654229a9 ("avoid IO::File for anonymous temporary files")
7 years agolinkify: modify argument in place
Eric Wong [Sat, 24 Dec 2016 11:52:44 +0000 (11:52 +0000)]
linkify: modify argument in place

This results in over 1% speedup doing $MESSAGE_ID/T/ HTML
generation for a 368-message thread.

7 years agoview: do not modify array during iteration
Eric Wong [Sat, 24 Dec 2016 11:52:43 +0000 (11:52 +0000)]
view: do not modify array during iteration

This results in a half percent speedup or so doing
$MESSAGE_ID/T/ HTML generation for a 368 message thread.

7 years agoview: stop chomping off whitespace at ends of messages
Eric Wong [Sat, 24 Dec 2016 11:52:42 +0000 (11:52 +0000)]
view: stop chomping off whitespace at ends of messages

This allows a 3-4% speedup in $MESSAGE_ID/T/ page generation
speed for a 368+ message thread.  It also more faithfully
preserves the message as intended; even if the it makes the
sender look like a space-wasting slob :P

7 years agoview: remove unused parameter
Eric Wong [Sat, 24 Dec 2016 11:52:41 +0000 (11:52 +0000)]
view: remove unused parameter

And add a comment about it to remind our future selves.

7 years agosearch: lookup_mail handles modified DBs
Eric Wong [Thu, 22 Dec 2016 08:00:26 +0000 (08:00 +0000)]
search: lookup_mail handles modified DBs

We call lookup_mail all over the place, be sure we can handle
database modifications in those cases.

7 years agodoc: various comments on async handling
Eric Wong [Thu, 22 Dec 2016 07:29:17 +0000 (07:29 +0000)]
doc: various comments on async handling

Notes for future developers (myself included) since we
can't assume people can read my mind.

7 years agosearchthread: simplify API and remove needless OO
Eric Wong [Tue, 20 Dec 2016 23:42:36 +0000 (23:42 +0000)]
searchthread: simplify API and remove needless OO

This simplifies callers to prevent errors and avoids
needless object-orientation in favor of a single procedure
call to handle threading and ordering.

7 years agosearchthread: update comment about loop prevention
Eric Wong [Tue, 20 Dec 2016 23:42:35 +0000 (23:42 +0000)]
searchthread: update comment about loop prevention

It definitely is necessary to prevent looping with the
%seen hash.

7 years agosearchmsg: remove ensure_metadata
Eric Wong [Tue, 20 Dec 2016 03:03:57 +0000 (03:03 +0000)]
searchmsg: remove ensure_metadata

Instead, only preload the ->mid field for threading,
as we only need ->thread and ->path once in Search->get_thread
(but we will need the ->mid field repeatedly).

This more than doubles View->load_results performance on
according to thread-all on an inbox with over 300K messages.

7 years agotests: add thread-all testing for benchmarking
Eric Wong [Tue, 20 Dec 2016 03:03:56 +0000 (03:03 +0000)]
tests: add thread-all testing for benchmarking

I'll be using this to improve message threading performance.

7 years agosearchmsg: do not memoize {date} field
Eric Wong [Sat, 17 Dec 2016 12:04:11 +0000 (12:04 +0000)]
searchmsg: do not memoize {date} field

We only generate the ->date once in NNTP, so creating
the hash entry is a waste.

7 years agosearchmsg: remove locale-dependency for ->date
Eric Wong [Sat, 17 Dec 2016 12:04:10 +0000 (12:04 +0000)]
searchmsg: remove locale-dependency for ->date

strftime is locale-dependent, which can cause surprising
failures for some users.

7 years agot/config.t: fix feedmax default
Eric Wong [Sat, 17 Dec 2016 05:50:30 +0000 (05:50 +0000)]
t/config.t: fix feedmax default

Oops :x

7 years agowwwtext: link to RFC4685 (Atom Threading)
Eric Wong [Wed, 14 Dec 2016 21:00:13 +0000 (21:00 +0000)]
wwwtext: link to RFC4685 (Atom Threading)

This should give this feature some more visibility.

7 years agoatom: implement message threading per RFC 4685
Eric Wong [Tue, 13 Dec 2016 02:33:30 +0000 (02:33 +0000)]
atom: implement message threading per RFC 4685

This will allows certain feed readers to render a message thread
as described in <https://www.jwz.org/doc/threading.html>.

Feed readers with knowledge of of RFC 4685 are unknown to us at
this time, but perhaps this will encourage future implementations.

Existing feed readers I've tested (newsbeuter, feed2imap) seem
to ignore these tags gracefully without degradation.

7 years agofeed: support publicinbox.<name>.feedmax
Eric Wong [Sat, 17 Dec 2016 04:27:52 +0000 (04:27 +0000)]
feed: support publicinbox.<name>.feedmax

This allows users to customize by using smaller or larger Atom
feeds than the default value of 25 entries.

7 years agoTODO: note IO::KQueue for the ticket
Eric Wong [Wed, 14 Dec 2016 23:53:06 +0000 (23:53 +0000)]
TODO: note IO::KQueue for the ticket

Do not require users to have network access to know what
the link refers to.

7 years agot/thread-cycle: no need for Xapian to run this test
Eric Wong [Wed, 14 Dec 2016 19:28:53 +0000 (19:28 +0000)]
t/thread-cycle: no need for Xapian to run this test

We don't actually use anything from SearchMsg,
just the class name.

7 years agowwwtext: remove outdated comment
Eric Wong [Wed, 14 Dec 2016 20:58:00 +0000 (20:58 +0000)]
wwwtext: remove outdated comment

I originally envisioned wwwtext being more flexible and able to
serve arbitrary blobs; but at this point I consider it redundant
and public-inbox is not wiki software.

7 years agosearchmsg: remove unused EPOCH_822 constant
Eric Wong [Tue, 13 Dec 2016 03:10:13 +0000 (03:10 +0000)]
searchmsg: remove unused EPOCH_822 constant

This hasn't been needed since our Email::Abstract removal
for message threading.

7 years agonntp: avoid useless use of strftime
Eric Wong [Tue, 13 Dec 2016 03:10:12 +0000 (03:10 +0000)]
nntp: avoid useless use of strftime

There's no need to use strftime if we'll be converting the date
by hand, anyways.

7 years agonntp: add test case for the "DATE" command
Eric Wong [Tue, 13 Dec 2016 03:10:11 +0000 (03:10 +0000)]
nntp: add test case for the "DATE" command

We may not always use strftime and may implement caching.
But for now, just add a test.

7 years agodaemon: set $now time for NNTP shutdown
Eric Wong [Mon, 12 Dec 2016 12:14:02 +0000 (12:14 +0000)]
daemon: set $now time for NNTP shutdown

commit 6e238ee3396719e578d6a90e177a71ce9f8c1ca0
("nntp: respect 3 minute idle time for shutdown")
was incomplete, and needed this change to Daemon
to be effective.

In the future, there will be more common code between
NNTP.pm and HTTP.pm

7 years agodoc: simplify makefile snippet
Eric Wong [Mon, 12 Dec 2016 12:07:21 +0000 (12:07 +0000)]
doc: simplify makefile snippet

We have these manpages, and will always have them, so stop
trying to pretend we're doing something about maintainability,
here.

7 years agoinit: preserve permissions of existing config file
Eric Wong [Mon, 12 Dec 2016 12:02:45 +0000 (12:02 +0000)]
init: preserve permissions of existing config file

This matches git-config(1) behavior, and implied user
intent when it comes to programatically editing files.

7 years agosearch: retry document loading from Xapian
Eric Wong [Sat, 10 Dec 2016 23:35:43 +0000 (23:35 +0000)]
search: retry document loading from Xapian

In addition to needing to retry enquire queries, we also need
to protect document loading from the Xapian DB and retry on
modification, as it seems to throw the same errors.

Checking the $@ ref for Search::Xapian::DatabaseModifiedError
is actually in the test suite for both the XS and SWIG Xapian
bindings, so we should be good as far as forward/backwards
compatibility.

7 years agosearch: always sort thread results in ascending time order
Eric Wong [Sat, 10 Dec 2016 01:09:51 +0000 (01:09 +0000)]
search: always sort thread results in ascending time order

This makes life easier for the threading algorithm, as we can
use the implied ordering of timestamps to avoid temporary ghosts
and resulting container vivication.

This would've also allowed us to hide the bug (in most cases)
fixed by the patch titled "thread: last Reference always wins",
in case that needs to be reverted due to infinite looping.

7 years agothread: last Reference always wins
Eric Wong [Sat, 10 Dec 2016 01:09:50 +0000 (01:09 +0000)]
thread: last Reference always wins

Since we use SearchMsg from Xapian data, we can be
assured we do not get self-referential {references}
field.

However, we may need to be more careful when checking
has_descendent for loops, as blindly calling add_child
could open us up to that possibility...

7 years agoview: skip ghosts with no direct children
Eric Wong [Sat, 10 Dec 2016 01:09:49 +0000 (01:09 +0000)]
view: skip ghosts with no direct children

Otherwise, a malicious or broken client could populate the
thread skeleton with invalid References.  We only care about
ghosts which messages correctly refer to, not totally bogus ones
which may be the result of long line or token truncation +
wrapping in MUA headers.

7 years agoview: reduce indentation for skeleton generation
Eric Wong [Sat, 10 Dec 2016 01:09:48 +0000 (01:09 +0000)]
view: reduce indentation for skeleton generation

This should reduce the number of subroutine calls needed
for the common case of real (non-ghost) messages as well
as shortening code.

7 years agothread: fix comment describing its existence
Eric Wong [Sat, 10 Dec 2016 01:09:47 +0000 (01:09 +0000)]
thread: fix comment describing its existence

Mail::Thread is UNavailable on many distros, meaning ordinary
users will have to rely on CPAN, a Perl-specific packaging tool.

7 years agoview: favor SearchMsg for In-Reply-To over Email::MIME
Eric Wong [Sat, 10 Dec 2016 03:21:29 +0000 (03:21 +0000)]
view: favor SearchMsg for In-Reply-To over Email::MIME

This should avoid warnings during thread skeleton generation if
ever the Xapian database disagrees with View.pm about which is
the proper direct parent of a message.  We will treat the data
in Xapian as the truth (if Xapian is available).

7 years agosearch: favor In-Reply-To over last References iff IRT exists
Eric Wong [Sat, 10 Dec 2016 01:09:46 +0000 (01:09 +0000)]
search: favor In-Reply-To over last References iff IRT exists

Some email clients set the References headers backwards, so
trust the In-Reply-To header if (and only if) it exists and
is parseable as direct parent of the current message.

For affected repos, this will require reindexing (via
"public-inbox-index --reindex"), but there will be no
version bump for this bugfix.

7 years agolinkify: implement Markdown link compatibility (again)
Eric Wong [Tue, 6 Dec 2016 23:40:33 +0000 (23:40 +0000)]
linkify: implement Markdown link compatibility (again)

Although unescaped parentheses in URLs are technically allowed,
they are uncommon.  However, Markdown-like syntaxes are
unfortunately common for URLs, so we might as well support them.

This fixes parentheses detection at sentence endings, as seen
in practice on emails.

7 years agoRevert "linkify: implement Markdown link compatibility"
Eric Wong [Tue, 6 Dec 2016 23:15:02 +0000 (23:15 +0000)]
Revert "linkify: implement Markdown link compatibility"

This reverts commit 130d0c4e33c5c73dc69e270fc698735d49e0f159.

7 years agolinkify: implement Markdown link compatibility
Eric Wong [Tue, 6 Dec 2016 23:01:39 +0000 (23:01 +0000)]
linkify: implement Markdown link compatibility

Although unescaped parentheses in URLs are technically allowed,
they are uncommon.  However, Markdown-like syntaxes are
unfortunately common for URLs, so we might as well support them.

7 years agoatom: switch to getline/close for response bodies
Eric Wong [Sat, 3 Dec 2016 00:24:06 +0000 (00:24 +0000)]
atom: switch to getline/close for response bodies

This will let us stream larger Atom documents bodies without
wasting too much memory and reduce the amount of round-trip
requests needed to get necessary information.

Hopefully clients are using streaming (SAX) parsers, too.

This is the final transition in the core public-inbox
code to allow migrating to a "pull"-based body streaming
scheme which allows a HTTP server to respond appropriately
to backpressure from slow clients.

7 years agowwwstream: improve documentation and variable naming
Eric Wong [Sat, 3 Dec 2016 00:24:05 +0000 (00:24 +0000)]
wwwstream: improve documentation and variable naming

Hopefully this makes the code more readable for newbies.

7 years agosearchview: fix <title> tag in Atom feed
Eric Wong [Sat, 3 Dec 2016 00:24:51 +0000 (00:24 +0000)]
searchview: fix <title> tag in Atom feed

This only affects the Atom feed for search results.
"xmlstarlet val" failed to detect or warn about this,
and I only noticed this bug while working on another
patch.

7 years agonote the source code is AGPL for cloning
Eric Wong [Tue, 29 Nov 2016 21:40:35 +0000 (21:40 +0000)]
note the source code is AGPL for cloning

This should be adequate warning for folks who may be
uncomfortable or uncertain about even possessing AGPL
source code due to employer agreements and such.

Disclaimer: I remain completely in favor of AGPL and strong
copyleft, and am more than willing to risk my own future on it.
However, I refuse to even nudge people into downloading AGPL
source code if it presents any legal risk to them.

7 years agoavoid IO::File for anonymous temporary files
Eric Wong [Sat, 26 Nov 2016 08:52:50 +0000 (08:52 +0000)]
avoid IO::File for anonymous temporary files

We do not need to import IO::File into the main programs
since Perl 5.8+ supports literal "undef" for generating
anonymous temporary file handles.

7 years agogithttpbackend: error checking for input handling
Eric Wong [Sat, 26 Nov 2016 08:34:34 +0000 (08:34 +0000)]
githttpbackend: error checking for input handling

This was sloppy code, all calls need to be checked
for failure.

7 years agoview: fix spaces in mailto: link
Eric Wong [Tue, 22 Nov 2016 02:49:40 +0000 (02:49 +0000)]
view: fix spaces in mailto: link

Some mail clients do not seem to handle '+' as a space in query
parameters for the mail subject, use the more common '%20' for
compatibility.

7 years agoindex: allow indexing before configuration
Eric Wong [Fri, 4 Nov 2016 21:11:35 +0000 (21:11 +0000)]
index: allow indexing before configuration

One may build the initial index on a powerful host and transfer
it to a weaker one for incremental indexing.  Thus there is
no requirement to have a configured public-inbox for building
the index unless a user needs altid support or some such.

7 years agoimport: failed GC runs are non-fatal
Eric Wong [Sun, 16 Oct 2016 00:36:14 +0000 (00:36 +0000)]
import: failed GC runs are non-fatal

We should not completely kill a process if "git gc --auto"
errors out due to a warning or whatnot.

7 years agothread: reinstates stable ordering when ghosts are present
Eric Wong [Fri, 14 Oct 2016 09:00:01 +0000 (09:00 +0000)]
thread: reinstates stable ordering when ghosts are present

This reverts commit 3c9dd6619f825f0515e7e4afa1bd55c99c1a68d3
("thread: fix sorting without topmost")
and reinstates the "topmost" routine for sorting purposes.

7 years agothread: fix parent/child relationships
Eric Wong [Thu, 13 Oct 2016 03:59:03 +0000 (03:59 +0000)]
thread: fix parent/child relationships

The ordering change in add_child is critical if $self == $parent
as the {children} hash was lost before this change.

has_descendent can be simplified by walking upwards from the child
instead of downwards from the parent.

This fixes threading regressions introduced in
commit 30100c46326e2eac275e0af13116636701d2537e
("thread: use hash + array instead of hand-rolled linked list")

7 years agothread: reduce indentation level
Eric Wong [Thu, 13 Oct 2016 03:59:02 +0000 (03:59 +0000)]
thread: reduce indentation level

This should reduce differences from the original Mail::Thread
code and hopefully make things easier-to-follow.

7 years agothread: remove weaken dependency
Eric Wong [Wed, 5 Oct 2016 23:47:32 +0000 (23:47 +0000)]
thread: remove weaken dependency

We have to walk through all the messages after threading
anyways to build the rootset, so we can just delete all
the parent references at that point.

7 years agot/thread-cycle: test self-referential messages
Eric Wong [Wed, 5 Oct 2016 23:47:31 +0000 (23:47 +0000)]
t/thread-cycle: test self-referential messages

Some broken (or malicious) mailers may include a generated
Message-ID in its References header, so be prepared for it.

7 years agoview: remove redundant children array in thread views
Eric Wong [Wed, 5 Oct 2016 23:47:30 +0000 (23:47 +0000)]
view: remove redundant children array in thread views

Each node has an entire arrayref of its children nowadays, so
there's no need to waste time and memory creating another one.

7 years agothread: use hash + array instead of hand-rolled linked list
Eric Wong [Wed, 5 Oct 2016 23:47:29 +0000 (23:47 +0000)]
thread: use hash + array instead of hand-rolled linked list

This starts to show noticeable performance improvements when
attempting to thread over 400 messages; but the improvement
may not be measurable with less.

However, the resulting code is much shorter and (IMHO)
much easier to understand.

7 years agothread: fix sorting without topmost
Eric Wong [Wed, 5 Oct 2016 23:47:28 +0000 (23:47 +0000)]
thread: fix sorting without topmost

This bug was hidden, and we may not be able to efficiently
implement a topmost subroutine with the hash-based (vs
linked-list) based container for threading in the next
commit.

7 years agothread: inline and remove recurse_down logic
Eric Wong [Wed, 5 Oct 2016 23:47:27 +0000 (23:47 +0000)]
thread: inline and remove recurse_down logic

We no longer recurse, and it's too hard to come up with
a new name for a sub we will only use once.

7 years agothread: order_children no longer cares about depth
Eric Wong [Wed, 5 Oct 2016 23:47:26 +0000 (23:47 +0000)]
thread: order_children no longer cares about depth

We never use the depth anywhere in this sub

7 years agothread: avoid incrementing undefined value
Eric Wong [Wed, 5 Oct 2016 23:47:25 +0000 (23:47 +0000)]
thread: avoid incrementing undefined value

It is pointless to increment when setting a true value is
simpler as there is no need to read before writing.

7 years agothread: remove iterate_down
Eric Wong [Wed, 5 Oct 2016 23:47:24 +0000 (23:47 +0000)]
thread: remove iterate_down

Unnecessary subs and complexity.  This was hiding the fact
that $before is never used.

7 years agothread: simplify
Eric Wong [Wed, 5 Oct 2016 23:47:23 +0000 (23:47 +0000)]
thread: simplify

Single use subroutines actually make the code more complex in
this case, and there's never a {seen} field in $self.

7 years agothread: remove rootset accessor method
Eric Wong [Wed, 5 Oct 2016 23:47:22 +0000 (23:47 +0000)]
thread: remove rootset accessor method

It doesn't buy us much and copying to a new array is slower;
but probably not measurable in real-world use.

7 years agothread: remove Email::Abstract wrapping
Eric Wong [Wed, 5 Oct 2016 23:47:21 +0000 (23:47 +0000)]
thread: remove Email::Abstract wrapping

This roughly doubles performance due to the reduction in
object creation and abstraction layers.

7 years agoinbox: deal with ghost smsg
Eric Wong [Wed, 5 Oct 2016 23:47:20 +0000 (23:47 +0000)]
inbox: deal with ghost smsg

smsg will be undef for ghost messages in a subsequent commit

7 years agothread: remove accessor usage in internals
Eric Wong [Wed, 5 Oct 2016 23:47:19 +0000 (23:47 +0000)]
thread: remove accessor usage in internals

This improves top-level index generation performance by 3-4%.

7 years agothread: pass array refs instead of entire arrays
Eric Wong [Wed, 5 Oct 2016 23:47:18 +0000 (23:47 +0000)]
thread: pass array refs instead of entire arrays

Copying large arrays is expensive, so avoid it.
This reduces /$INBOX/ time by around 1%.

7 years agothread: remove Mail::Thread dependency
Eric Wong [Wed, 5 Oct 2016 23:47:17 +0000 (23:47 +0000)]
thread: remove Mail::Thread dependency

Introduce our own SearchThread class for threading messages.
This should allow us to specialize and optimize away objects
in future commits.

7 years agoview: remove "subject dummy" references
Eric Wong [Wed, 5 Oct 2016 23:47:16 +0000 (23:47 +0000)]
view: remove "subject dummy" references

We will not care for inexact threading by subject or pruning.

7 years agohelp: document new search prefixes
Eric Wong [Tue, 13 Sep 2016 01:18:30 +0000 (01:18 +0000)]
help: document new search prefixes

Support (and document) 'a:' after all, as "mairix -h" uses it,
so this should reduce the learning curve for mairix users.

7 years agonntp: cleanup: move use statements out of sub scope
Eric Wong [Fri, 9 Sep 2016 18:43:55 +0000 (18:43 +0000)]
nntp: cleanup: move use statements out of sub scope

This clarifies the code somewhat, and we don't care to lazy-load
in NNTP.pm anyways since this is only used for a long-lived
daemon.

7 years agoTODO: updates for done items
Eric Wong [Fri, 9 Sep 2016 09:05:18 +0000 (09:05 +0000)]
TODO: updates for done items

The existing string -> number date range Xapian query is good
enough, and having too much flexibility is probably bad for
caching (as well as increasing our attack surface, because
parsing queries is tricky).

Tags-as-skiplists are probably not worth the effort given
Xapian, and we may have to import old messages after-the-fact,
anyways, and message delivery for mirrors is never orderly.

Other items are all done and need to be maintained (like the
search engine docs for the mairix-compatibility features that
just got pushed out)

7 years agot/httpd-unix: warn about connection failure
Eric Wong [Fri, 9 Sep 2016 03:09:00 +0000 (03:09 +0000)]
t/httpd-unix: warn about connection failure

Output $! for diagnostic purposes since I've noticed this on
two slow machines, today (and seemingly, never prior).

7 years agosearch: index attachment filenames
Eric Wong [Fri, 9 Sep 2016 00:01:31 +0000 (00:01 +0000)]
search: index attachment filenames

And while we're at it, ensure searching inside displayable
attachment bodies works.

7 years agosearch: match the behavior of WWW for indexing text
Eric Wong [Fri, 9 Sep 2016 00:01:30 +0000 (00:01 +0000)]
search: match the behavior of WWW for indexing text

The basic rule is that if it is displayable via our WWW
interface, it should be indexable text for Xapian search.

7 years agosearch: avoid mindlessly calling body_set
Eric Wong [Fri, 9 Sep 2016 00:01:29 +0000 (00:01 +0000)]
search: avoid mindlessly calling body_set

It's not worth entering a complex codepath in Email::MIME to
save some (probably immeasurable amount of) memory, here.  We've
already stopped doing this in our WWW code a while back, too.
If we really cared enough about it, we'd prioritize work on a
streaming replacement for Email::MIME.

7 years agosearch: fix compatibility with Debian wheezy
Eric Wong [Fri, 9 Sep 2016 00:01:28 +0000 (00:01 +0000)]
search: fix compatibility with Debian wheezy

Specifying the "d:" field only worked for
NumberValueRangeProcessor in older versions of Xapian, such
as the one in Debian wheezy (libsearch-xapian-perl=1.2.10.0-1)

This slipped through since I rarely use wheezy, anymore, and
perhaps nobody else does, either.  Perhaps wheezy support may be
dropped, soon.

Unfortunately, this requires a schema version bump.

7 years agosearch: increase term positions for each quoted hunk
Eric Wong [Fri, 9 Sep 2016 00:01:27 +0000 (00:01 +0000)]
search: increase term positions for each quoted hunk

We pay a storage cost for storing positional information
in Xapian, make good use of it by attempting to preserve
it for (hopefully) better search results.

7 years agosearch: match quote detection behavior of view
Eric Wong [Fri, 9 Sep 2016 00:01:26 +0000 (00:01 +0000)]
search: match quote detection behavior of view

This is stricter than the mutt quote_regexp default
("^([ \t]*[|>:}#])+" on Debian jessie),
but matches what we have in View.pm.

I prefer the stricter quote detection since it is less ambiguous
and less likely to hide/obscure important details.

7 years agosearch: fix space regressions from recent changes
Eric Wong [Fri, 9 Sep 2016 00:01:25 +0000 (00:01 +0000)]
search: fix space regressions from recent changes

As of Xapian 1.0.4 (from 2007) is possible to use
Search::Xapian::QueryParser::add_prefix multiple times with the
same user field name but different term prefixes.

This brings my current git@vger mirror from 6.5GB to 2.1GB
(both sizes are after xapian-compact).

7 years agosearch: more granular message body searching
Eric Wong [Fri, 9 Sep 2016 00:01:24 +0000 (00:01 +0000)]
search: more granular message body searching

"bs:" and "b:" are adapted from mairix(1)

We will also support searching explicitly for quoted vs
non-quoted text via "q:" and "nq:" prefixes since sometimes
readers will not care for quoted text.

In the future, we will support parsing diffs (perhaps when
repobrowse integration is complete).

Note: this roughly doubles the size of the Xapian database due
to the additional information; so this change may not be worth
it.

7 years agosearch: drop longer subject: prefix for search
Eric Wong [Fri, 9 Sep 2016 00:01:23 +0000 (00:01 +0000)]
search: drop longer subject: prefix for search

We only document the "s:" anyways.  While the long name is more
descriptive, the ambiguity makes agnostic caching (by Varnish or
similar) slightly harder and longer URLs are more likely to be
accidentally truncated when shared.

7 years agosearch: allow searching user fields (To/Cc/From)
Eric Wong [Fri, 9 Sep 2016 00:01:22 +0000 (00:01 +0000)]
search: allow searching user fields (To/Cc/From)

Sometimes it can be useful to search based on who the
message was sent to, sent by, or Cc:-ed.  Of course,
headers can be faked, but they usually are not...

Anyways this mostly matches the behavior of mairix(1).

7 years agoimport: run "git gc --auto" when done
Eric Wong [Thu, 8 Sep 2016 22:42:42 +0000 (22:42 +0000)]
import: run "git gc --auto" when done

We need to prevent excessive repository growth for
public-inbox-watch and public-inbox-mda users.

7 years agoimport: hoist out common run_die subroutine
Eric Wong [Thu, 8 Sep 2016 22:41:01 +0000 (22:41 +0000)]
import: hoist out common run_die subroutine

We will be reusing this in the next commit, too.

7 years agodoc: document PERL_INLINE_DIRECTORY usage
Eric Wong [Thu, 8 Sep 2016 20:15:25 +0000 (20:15 +0000)]
doc: document PERL_INLINE_DIRECTORY usage

For now, we will document this since it allows better
performance without the burden of extensions.  Perhaps one day
far in the future Perl can natively support vfork(2) AND that
version of Perl will be widely available, but I suspect that day
is at least a decade away, if not two:

https://rt.perl.org/Ticket/Display.html?id=128227