Eric Wong [Thu, 3 Sep 2015 03:00:28 +0000 (03:00 +0000)]
get rid of Message-ID compression entirely
Provide a fallback for legacy SHA-1 messages, but do not
advertise shorter URLs anymore for data portability concerns.
This fixes a regression introduced in
commit 81a9c1b476987d845b340ab9013d26cf4487cb9a
("search: disable Message-ID compression in Xapian")
which ended up breaking thread-related endpoints for
large Message-IDs, as lookups on the SHA-1 message no longer
worked.
Eric Wong [Thu, 3 Sep 2015 01:57:11 +0000 (01:57 +0000)]
search: disable Message-ID compression in Xapian
We'll continue to compress long Message-IDs in URLs (which we know
about), but we will store entire Message-IDs in the Xapian database
to facilitate ease-of-lookups in external databases.
Eric Wong [Wed, 2 Sep 2015 02:37:23 +0000 (02:37 +0000)]
implement external Message-ID finder
Currently, this looks at other public-inbox configurations
served in the same process. In the future, it will generate
links to other Message-ID lookup endpoints.
Eric Wong [Wed, 2 Sep 2015 02:37:22 +0000 (02:37 +0000)]
view: avoid links to unknown compressed Message-IDs
Compressed Message-IDs are irreversible and may not be used
at other sites. So avoid compressing Message-IDs we do not
know about so users have a chance of finding the message in
other archives by doing a Message-ID lookup.
Eric Wong [Wed, 2 Sep 2015 02:37:17 +0000 (02:37 +0000)]
view: close possible race condition in thread view
It's possible that the Xapian index and git HEAD can be out-of-sync
and a message which existed when we did the search is no longer
accessible by the time we get to rendering it.
Eric Wong [Tue, 1 Sep 2015 08:55:28 +0000 (08:55 +0000)]
view: more robust link generation
We must avoid double-escaping in cases where we have URLs anchored
by "<>" in the plain-text as is common (and AFAIK recommended)
convention. So we must use a two step linkification process
to prevent double-escaping.
Eric Wong [Tue, 1 Sep 2015 08:55:18 +0000 (08:55 +0000)]
search: reduce redundant doc data
Redundant document data increases our database size, pull the
smsg->mid off the unique term, the smsg->ts off the value, and
only generate the formatted display date off smsg->ts.
Eric Wong [Sun, 30 Aug 2015 10:12:54 +0000 (10:12 +0000)]
www: avoid BEGIN block for config loading
It fails the syntax check if a user does not have
~/.public-inbox/config setup. Anyways we can safely
use ||= on a global since we do not support threads.
Eric Wong [Sun, 30 Aug 2015 01:51:22 +0000 (01:51 +0000)]
view: remove "threadlink" from thread view
We're already inside the thread, and our thread summary inside
/m/$MESSAGE_ID/ is already sufficient got navigate back to the
/t/$MESSAGE_ID/ page. So I think it's sufficient to keep the
/t/$MESSAGE_ID/ page lighter with fewer links and avoid
introducing strange terminology.
In contrast, "permalink" is relatively well-known and
not an alien term to readers:
Eric Wong [Sun, 30 Aug 2015 01:26:46 +0000 (01:26 +0000)]
mid2path: clean MID of angle brackets '<>'
We screwed up and needed to fix URL generation with '<>'
in them. Regardless, users may attempt to copy and paste
URLs with '<>' in them, do not punish them for that.
Eric Wong [Sun, 30 Aug 2015 01:04:31 +0000 (01:04 +0000)]
public-inbox-index: resolve git directory if run inside one
I often forget to pass the correct path to a git directory
or run from inside one. Fortunately git is script-friendly
and allows easily resolving the correct GIT_DIR path.
Eric Wong [Sun, 30 Aug 2015 00:38:05 +0000 (00:38 +0000)]
search: do not index references and inreplyto terms
We no longer need them, as we can rely on index-time thread
resolution and thread merging. This allows us to index less
data and hopefully increase efficiency.
Eric Wong [Sun, 30 Aug 2015 00:22:43 +0000 (00:22 +0000)]
view: display thread outline in single-message view
If Xapian search is available, we can load most of the
entire thread and show a more meaningful navigation tree
than the References: and In-Reply-To: headers. Searching
on those headers themselves is unreliable because it is
possible for clients to omit some references.
Eric Wong [Fri, 28 Aug 2015 00:00:47 +0000 (00:00 +0000)]
search: do not iterate through entire termlist
A document may have many terms, so this hurts performance
if we blindly iterate. Unfortunately, we can't rely on the
order of the termlist just yet, either, so we must repeatedly
restart the search for now until we're ready to bump schema
versions.
Eric Wong [Fri, 28 Aug 2015 00:21:46 +0000 (00:21 +0000)]
GitCatFile: remove unnecessary FD_CLOEXEC setting
Unless some idiot raises $^F, we should not have to care about
the close-on-exec flag. Everything since Perl 3.0 seems to set
it by default, and 5.6 got more consistent about it.
Eric Wong [Thu, 27 Aug 2015 04:34:01 +0000 (04:34 +0000)]
wire up to display non-suffixed Message-ID links
These URLs are preferable in case somebody decides to get cute and
use a suffix we would've used to prevent others from linking to
their message. The common /m/$MESSAGE_ID/ URLs are now 4 characters
shorter so may fit better on terminals.
Eric Wong [Thu, 27 Aug 2015 04:34:00 +0000 (04:34 +0000)]
mid: extract Message-ID from inside '<>'
This is necessary for some mailers which include comment text
in in the In-Reply-To header, merely assuming there is nothing
outside of '<>' as we were doing is not enough.
Eric Wong [Thu, 27 Aug 2015 04:33:59 +0000 (04:33 +0000)]
wire up shorter, less ambiguous URLs
We will prefer URLs without suffixes for now to avoid ambiguity
in case a Message-ID ends with ".html", ".txt", ".mbox.gz" or
any other suffix we may use.
Static file compatibility is preserved by using a trailing slash
as most servers can/will fall back to an index.html file in this
case.
For raw text files, we will follow gmane's lead with "/raw"
Eric Wong [Mon, 24 Aug 2015 02:25:46 +0000 (02:25 +0000)]
view: refactor $state as a hash
Using hash means we no longer have to document and remember what
every field does. The original array form was insane premature
optimization and crazy. Who wrote that? Oh wait, I was on
drugs :<
Eric Wong [Sun, 23 Aug 2015 20:05:41 +0000 (20:05 +0000)]
cleanup calls to header_obj
Dereference header_obj only once when performance may be
critical, or simplify our code by calling "header" directly on
the Email::{Simple,MIME} object if not.
Eric Wong [Sun, 23 Aug 2015 19:41:28 +0000 (19:41 +0000)]
hopefully fix broken permissions for search
We must preserve the umask for the entirety of the indexing
operation, as Xapian transactions replace entire files
atomically instead of writing them in place.
Eric Wong [Sun, 23 Aug 2015 00:31:28 +0000 (00:31 +0000)]
mbox: clarify our use of the the mboxrd variant
Commenting it in the From: line seems appropriate and
reduces compatibility problems in case a MUA cannot handle
trailing comments after the timestamp.
Eric Wong [Sun, 23 Aug 2015 00:02:34 +0000 (00:02 +0000)]
.txt links return an mbox instead
This improves compatibility and allows individual messages
to be concatenated into an existing mbox without further
modifications. "git format-patch" does something similar
(but does not do "From " line escaping(!))
Eric Wong [Sat, 22 Aug 2015 11:41:24 +0000 (11:41 +0000)]
view: wire up mbox.gz links
To reduce clutter, we will not link to uncompressed versions.
Users should be able to download entire threads for offline
reading, enable this feature for them.
Eric Wong [Sat, 22 Aug 2015 08:00:37 +0000 (08:00 +0000)]
remove XML::Atom::SimpleFeed dependency
We will attempt to generate Atom feeds "by hand" as the
XML::Atom::SimpleFeed API does not support streaming output.
Since email is large and servers are small, this should prevent
wasting memory when we generate larger feeds.
Of course, we hope clients use SAX parsers capable of handling
large streams without slurping.
Eric Wong [Sat, 22 Aug 2015 00:06:45 +0000 (00:06 +0000)]
stream HTML views as much as possible
This should allow progressive rendering on the client and reduce
memory usage on the server. Unfortunately XML::Atom::SimpleFeed
does not yet support streaming, so we may not use it in the
future.
Eric Wong [Thu, 20 Aug 2015 10:17:34 +0000 (10:17 +0000)]
search: preserve References: order in document data
We need proper ordering of References to thread messages
correctly. We would lose this order if we load the terms
from the database, so set it directly document data.
Do not bother with a separate In-Reply-To, since Mail::Thread
just merges the IRT into References. This bumps our schema
version once again.
Eric Wong [Thu, 20 Aug 2015 08:54:32 +0000 (08:54 +0000)]
avoid using header_raw for Message-ID retrieval
This is for consistency with ssoma. I doubt it makes
a difference in practice, but in case somebody decides
any of the Message-ID-containing headers should have
strange characters, we'll decode and attempt to thread
them. This isn't an attack vector, just a way to
make messages thread improperly which is pointless...