Date: Wed, 9 Oct 2013 10:36:58 -0400
From: Austin Clements <amdragon@MIT.EDU>
To: Jani Nikula <jani@nikula.org>
Subject: Re: [PATCH 08/11] search: Add stable queries to thread search results
Message-ID: <20131009143658.GQ21611@mit.edu>
References: <1381185201-25197-1-git-send-email-amdragon@mit.edu>
	<1381185201-25197-9-git-send-email-amdragon@mit.edu>
	<87fvsaao2q.fsf@nikula.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <87fvsaao2q.fsf@nikula.org>
User-Agent: Mutt/1.5.21 (2010-09-15)
Cc: notmuch@notmuchmail.org
Precedence: list

Quoth Jani Nikula on Oct 09 at  9:41 am:
> On Tue, 08 Oct 2013, Austin Clements <amdragon@MIT.EDU> wrote:
> > These queries will match exactly the set of messages currently in the
> > thread, even if more messages later arrive.  Two queries are provided:
> > one for matched messages and one for unmatched messages.
> >
> > This can be used to fix race conditions with tagging threads from
> > search results.  While tagging based on a thread: query can affect
> > messages that arrived after the search, tagging based on stable
> > queries affects only the messages the user was shown in the search UI.
> >
> > Since we want clients to be able to depend on the presence of these
> > queries, this ushers in schema version 2.
> > ---
> >  devel/schemata       | 22 +++++++++++++++++--
> >  notmuch-client.h     |  2 +-
> >  notmuch-search.c     | 60 ++++++++++++++++++++++++++++++++++++++++++++++++++++
> >  test/json            |  2 ++
> >  test/missing-headers |  6 ++++--
> >  test/sexp            |  4 ++--
> >  6 files changed, 89 insertions(+), 7 deletions(-)
> >
> > diff --git a/devel/schemata b/devel/schemata
> > index cdd0e43..41dc4a6 100644
> > --- a/devel/schemata
> > +++ b/devel/schemata
> > @@ -14,7 +14,17 @@ are interleaved. Keys are printed as keywords (symbols preceded by a
> >  colon), e.g. (:id "123" :time 54321 :from "foobar"). Null is printed as
> >  nil, true as t and false as nil.
> >  
> > -This is version 1 of the structured output format.
> > +This is version 2 of the structured output format.
> > +
> > +Version history
> > +---------------
> > +
> > +v1
> > +- First versioned schema release.
> > +- Added part.content-length and part.content-transfer-encoding fields.
> > +
> > +v2
> > +- Added the thread_summary.query field.
> >  
> >  Common non-terminals
> >  --------------------
> > @@ -145,7 +155,15 @@ thread_summary = {
> >      authors:        string,   # comma-separated names with | between
> >                                # matched and unmatched
> >      subject:        string,
> > -    tags:           [string*]
> > +    tags:           [string*],
> > +
> > +    # Two stable query strings identifying exactly the matched and
> > +    # unmatched messages currently in this thread.  The messages
> > +    # matched by these queries will not change even if more messages
> > +    # arrive in the thread.  If there are no matched or unmatched
> > +    # messages, the corresponding query will be null (there is no
> > +    # query that matches nothing).  (Added in schema version 2.)
> > +    query:          [string|null, string|null],
> >  }
> >  
> >  notmuch reply schema
> > diff --git a/notmuch-client.h b/notmuch-client.h
> > index 8d986f4..1b14910 100644
> > --- a/notmuch-client.h
> > +++ b/notmuch-client.h
> > @@ -138,7 +138,7 @@ chomp_newline (char *str)
> >   * this.  New (required) map fields can be added without increasing
> >   * this.
> >   */
> > -#define NOTMUCH_FORMAT_CUR 1
> > +#define NOTMUCH_FORMAT_CUR 2
> >  /* The minimum supported structured output format version.  Requests
> >   * for format versions below this will return an error. */
> >  #define NOTMUCH_FORMAT_MIN 1
> > diff --git a/notmuch-search.c b/notmuch-search.c
> > index d9d39ec..1d14651 100644
> > --- a/notmuch-search.c
> > +++ b/notmuch-search.c
> > @@ -20,6 +20,7 @@
> >  
> >  #include "notmuch-client.h"
> >  #include "sprinter.h"
> > +#include "string-util.h"
> >  
> >  typedef enum {
> >      OUTPUT_SUMMARY,
> > @@ -46,6 +47,46 @@ sanitize_string (const void *ctx, const char *str)
> >      return out;
> >  }
> >  
> > +/* Return two stable query strings that identify exactly the matched
> > + * and unmatched messages currently in thread.  If there are no
> > + * matched or unmatched messages, the returned buffers will be
> > + * NULL. */
> > +static int
> > +get_thread_query (notmuch_thread_t *thread,
> > +		  char **matched_out, char **unmached_out)
> > +{
> > +    notmuch_messages_t *messages;
> > +    char *escaped = NULL;
> > +    size_t escaped_len = 0;
> > +
> > +    *matched_out = *unmached_out = NULL;
> > +
> > +    for (messages = notmuch_thread_get_messages (thread);
> > +	 notmuch_messages_valid (messages);
> > +	 notmuch_messages_move_to_next (messages))
> > +    {
> > +	notmuch_message_t *message = notmuch_messages_get (messages);
> > +	const char *mid = notmuch_message_get_message_id (message);
> > +	/* Determine which query buffer to extend */
> > +	char **buf = notmuch_message_get_flag (
> > +	    message, NOTMUCH_MESSAGE_FLAG_MATCH) ? matched_out : unmached_out;
> > +	/* Allocate the query buffer is this is the first message */
> > +	if (!*buf && (*buf = talloc_strdup (thread, "")) == NULL)
> > +	    return -1;
> 
> I think it would improve clarity if you dropped the above...
> 
> > +	/* Add this message's id: query.  Since "id" is an exclusive
> > +	 * prefix, it is implicitly 'or'd together, so we only need to
> > +	 * join queries with a space. */
> > +	if (make_boolean_term (thread, "id", mid, &escaped, &escaped_len) < 0)
> > +	    return -1;
> > +	*buf = talloc_asprintf_append_buffer (
> > +	    *buf, "%s%s", **buf ? " " : "", escaped);
> 
> ...and turned this into:
> 
> 	if (*buf)
> 	    *buf = talloc_asprintf_append_buffer (*buf, " %s", escaped);
> 	else
> 	    *buf = talloc_strdup (thread, escaped);

Much nicer!

> Also one talloc less. Which brings me to the main worry:
> performance. What's the impact?

Seems to be about 1%-3% for CLI search (tested on the medium corpus).
It's hard to measure what the effect on Emacs search is, though I
would expect it to be similarly negligible.  Some work I did several
attempts at this ago suggests that this slows down tagging (though I
doubt it would be noticeable for single threads), but I also found
that switching to docid-based queries significantly sped things up:
id:CAH-f9WsPj=1Eu=g3sOePJgCTBFs6HrLdLq18xMEnJ8aZ00yCEg@mail.gmail.com
Actually, docid queries probably make tagging faster than it is *now*,
but I didn't measure that when I did the experiments.

> BR,
> Jani.
> 
> 
> > +	if (!*buf)
> > +	    return -1;
> > +    }
> > +    talloc_free (escaped);
> > +    return 0;
> > +}
> > +
> >  static int
> >  do_search_threads (sprinter_t *format,
> >  		   notmuch_query_t *query,
> > @@ -131,6 +172,25 @@ do_search_threads (sprinter_t *format,
> >  		format->string (format, authors);
> >  		format->map_key (format, "subject");
> >  		format->string (format, subject);
> > +		if (notmuch_format_version >= 2) {
> > +		    char *matched_query, *unmatched_query;
> > +		    if (get_thread_query (thread, &matched_query,
> > +					  &unmatched_query) < 0) {
> > +			fprintf (stderr, "Out of memory\n");
> > +			return 1;
> > +		    }
> > +		    format->map_key (format, "query");
> > +		    format->begin_list (format);
> > +		    if (matched_query)
> > +			format->string (format, matched_query);
> > +		    else
> > +			format->null (format);
> > +		    if (unmatched_query)
> > +			format->string (format, unmatched_query);
> > +		    else
> > +			format->null (format);
> > +		    format->end (format);
> > +		}
> >  	    }
> >  
> >  	    talloc_free (ctx_quote);
> > diff --git a/test/json b/test/json
> > index b87b7f6..e07a290 100755
> > --- a/test/json
> > +++ b/test/json
> > @@ -26,6 +26,7 @@ test_expect_equal_json "$output" "[{\"thread\": \"XXX\",
> >   \"total\": 1,
> >   \"authors\": \"Notmuch Test Suite\",
> >   \"subject\": \"json-search-subject\",
> > + \"query\": [\"id:$gen_msg_id\", null],
> >   \"tags\": [\"inbox\",
> >   \"unread\"]}]"
> >  
> > @@ -59,6 +60,7 @@ test_expect_equal_json "$output" "[{\"thread\": \"XXX\",
> >   \"total\": 1,
> >   \"authors\": \"Notmuch Test Suite\",
> >   \"subject\": \"json-search-utf8-body-sübjéct\",
> > + \"query\": [\"id:$gen_msg_id\", null],
> >   \"tags\": [\"inbox\",
> >   \"unread\"]}]"
> >  
> > diff --git a/test/missing-headers b/test/missing-headers
> > index f14b878..43e861b 100755
> > --- a/test/missing-headers
> > +++ b/test/missing-headers
> > @@ -43,7 +43,8 @@ test_expect_equal_json "$output" '
> >          ],
> >          "thread": "XXX",
> >          "timestamp": 978709437,
> > -        "total": 1
> > +        "total": 1,
> > +        "query": ["id:notmuch-sha1-7a6e4eac383ef958fcd3ebf2143db71b8ff01161", null]
> >      },
> >      {
> >          "authors": "Notmuch Test Suite",
> > @@ -56,7 +57,8 @@ test_expect_equal_json "$output" '
> >          ],
> >          "thread": "XXX",
> >          "timestamp": 0,
> > -        "total": 1
> > +        "total": 1,
> > +        "query": ["id:notmuch-sha1-ca55943aff7a72baf2ab21fa74fab3d632401334", null]
> >      }
> >  ]'
> >  
> > diff --git a/test/sexp b/test/sexp
> > index 492a82f..be815e1 100755
> > --- a/test/sexp
> > +++ b/test/sexp
> > @@ -19,7 +19,7 @@ test_expect_equal "$output" "((((:id \"${gen_msg_id}\" :match t :excluded nil :f
> >  test_begin_subtest "Search message: sexp"
> >  add_message "[subject]=\"sexp-search-subject\"" "[date]=\"Sat, 01 Jan 2000 12:00:00 -0000\"" "[body]=\"sexp-search-message\""
> >  output=$(notmuch search --format=sexp "sexp-search-message" | notmuch_search_sanitize)
> > -test_expect_equal "$output" "((:thread \"0000000000000002\" :timestamp 946728000 :date_relative \"2000-01-01\" :matched 1 :total 1 :authors \"Notmuch Test Suite\" :subject \"sexp-search-subject\" :tags (\"inbox\" \"unread\")))"
> > +test_expect_equal "$output" "((:thread \"0000000000000002\" :timestamp 946728000 :date_relative \"2000-01-01\" :matched 1 :total 1 :authors \"Notmuch Test Suite\" :subject \"sexp-search-subject\" :query (\"id:$gen_msg_id\" nil) :tags (\"inbox\" \"unread\")))"
> >  
> >  test_begin_subtest "Show message: sexp, utf-8"
> >  add_message "[subject]=\"sexp-show-utf8-body-sübjéct\"" "[date]=\"Sat, 01 Jan 2000 12:00:00 -0000\"" "[body]=\"jsön-show-méssage\""
> > @@ -44,7 +44,7 @@ test_expect_equal "$output" "((((:id \"$id\" :match t :excluded nil :filename \"
> >  test_begin_subtest "Search message: sexp, utf-8"
> >  add_message "[subject]=\"sexp-search-utf8-body-sübjéct\"" "[date]=\"Sat, 01 Jan 2000 12:00:00 -0000\"" "[body]=\"jsön-search-méssage\""
> >  output=$(notmuch search --format=sexp "jsön-search-méssage" | notmuch_search_sanitize)
> > -test_expect_equal "$output" "((:thread \"0000000000000005\" :timestamp 946728000 :date_relative \"2000-01-01\" :matched 1 :total 1 :authors \"Notmuch Test Suite\" :subject \"sexp-search-utf8-body-sübjéct\" :tags (\"inbox\" \"unread\")))"
> > +test_expect_equal "$output" "((:thread \"0000000000000005\" :timestamp 946728000 :date_relative \"2000-01-01\" :matched 1 :total 1 :authors \"Notmuch Test Suite\" :subject \"sexp-search-utf8-body-sübjéct\" :query (\"id:$gen_msg_id\" nil) :tags (\"inbox\" \"unread\")))"
> >  
> >  
> >  test_done
> >
> > _______________________________________________
> > notmuch mailing list
> > notmuch@notmuchmail.org
> > http://notmuchmail.org/mailman/listinfo/notmuch

-- 
Austin Clements                                      MIT/'06/PhD/CSAIL
amdragon@mit.edu                           http://web.mit.edu/amdragon
       Somewhere in the dream we call reality you will find me,
              searching for the reality we call dreams.