[PATCH v4 0/9] notmuch search date:since..until query support

[PATCH v4 0/9] notmuch search date:since..until query support

[Jani Nikula]

Hi all, v4 of id:"cover.1347484177.git.jani@nikula.org", addressing most
of Michal Sojka's comments and issues in [1] and [2], and incorporating
his idea for improved testability and the tests from [2]. Thanks Michal!

[1] id:"8739262u4i.fsf@steelpick.2x.cz"
[2] id:"87zk4e1f5k.fsf@steelpick.2x.cz"

BR,
Jani.

Jani Nikula (9):
  build: drop the -Wswitch-enum warning
  parse-time-string: add a date/time parser to notmuch
  test: add new test tool parse-time for date/time parser
  test: add smoke tests for the date/time parser module
  build: build parse-time-string as part of the notmuch lib and static
    cli
  lib: add date range query support
  test: add tests for date:since..until range queries
  man: document the date:since..until range queries
  NEWS: date range search support

 Makefile                              |    2 +-
 Makefile.local                        |    2 +-
 NEWS                                  |   14 +
 configure                             |    2 +-
 lib/Makefile.local                    |    3 +-
 lib/database-private.h                |    1 +
 lib/database.cc                       |    5 +
 lib/parse-time-vrp.cc                 |   40 +
 lib/parse-time-vrp.h                  |   19 +
 man/man7/notmuch-search-terms.7       |  147 +++-
 parse-time-string/Makefile            |    5 +
 parse-time-string/Makefile.local      |   12 +
 parse-time-string/README              |    9 +
 parse-time-string/parse-time-string.c | 1492 +++++++++++++++++++++++++++++++++
 parse-time-string/parse-time-string.h |  102 +++
 test/Makefile.local                   |    7 +-
 test/basic                            |    2 +-
 test/notmuch-test                     |    2 +
 test/parse-time-string                |   71 ++
 test/parse-time.c                     |  281 +++++++
 test/search-date                      |   21 +
 21 files changed, 2221 insertions(+), 18 deletions(-)
 create mode 100644 lib/parse-time-vrp.cc
 create mode 100644 lib/parse-time-vrp.h
 create mode 100644 parse-time-string/Makefile
 create mode 100644 parse-time-string/Makefile.local
 create mode 100644 parse-time-string/README
 create mode 100644 parse-time-string/parse-time-string.c
 create mode 100644 parse-time-string/parse-time-string.h
 create mode 100755 test/parse-time-string
 create mode 100644 test/parse-time.c
 create mode 100755 test/search-date

-- 
1.7.9.5

[PATCH v4 1/9] build: drop the -Wswitch-enum warning

[Jani Nikula]

-Wswitch-enum is a bit awkward if a switch statement is intended to
handle just some of the named codes of an enumeration especially, and
leave the rest to the default label.

We already have -Wall, which enables -Wswitch by default, and per GCC
documentation, "The only difference between -Wswitch and this option
[-Wswitch-enum] is that this option gives a warning about an omitted
enumeration code even if there is a default label."

Drop -Wswitch-enum to not force listing all named codes of
enumerations in switch statements that have a default label.

---

This will be useful in the next patch.
---
 configure |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/configure b/configure
index acb90a8..afa5c16 100755
--- a/configure
+++ b/configure
@@ -532,7 +532,7 @@ fi
 
 WARN_CXXFLAGS=""
 printf "Checking for available C++ compiler warning flags... "
-for flag in -Wall -Wextra -Wwrite-strings -Wswitch-enum; do
+for flag in -Wall -Wextra -Wwrite-strings; do
     if ${CC} $flag -o minimal minimal.c > /dev/null 2>&1
     then
 	WARN_CXXFLAGS="${WARN_CXXFLAGS}${WARN_CXXFLAGS:+ }${flag}"
-- 
1.7.9.5

[PATCH v4 2/9] parse-time-string: add a date/time parser to notmuch

[Jani Nikula]

Add a date/time parser to notmuch, to be used for adding date range
query support for notmuch lib later on. Add the parser to a directory
of its own to make it independent of the rest of the notmuch code
base.

Signed-off-by: Jani Nikula <jani@nikula.org>
---
 Makefile                              |    2 +-
 parse-time-string/Makefile            |    5 +
 parse-time-string/Makefile.local      |   12 +
 parse-time-string/README              |    9 +
 parse-time-string/parse-time-string.c | 1492 +++++++++++++++++++++++++++++++++
 parse-time-string/parse-time-string.h |  102 +++
 6 files changed, 1621 insertions(+), 1 deletion(-)
 create mode 100644 parse-time-string/Makefile
 create mode 100644 parse-time-string/Makefile.local
 create mode 100644 parse-time-string/README
 create mode 100644 parse-time-string/parse-time-string.c
 create mode 100644 parse-time-string/parse-time-string.h

diff --git a/Makefile b/Makefile
index e5e2e3a..bb9c316 100644
--- a/Makefile
+++ b/Makefile
@@ -3,7 +3,7 @@
 all:
 
 # List all subdirectories here. Each contains its own Makefile.local
-subdirs = compat completion emacs lib man util test
+subdirs = compat completion emacs lib man parse-time-string util test
 
 # We make all targets depend on the Makefiles themselves.
 global_deps = Makefile Makefile.config Makefile.local \
diff --git a/parse-time-string/Makefile b/parse-time-string/Makefile
new file mode 100644
index 0000000..fa25832
--- /dev/null
+++ b/parse-time-string/Makefile
@@ -0,0 +1,5 @@
+all:
+	$(MAKE) -C .. all
+
+.DEFAULT:
+	$(MAKE) -C .. $@
diff --git a/parse-time-string/Makefile.local b/parse-time-string/Makefile.local
new file mode 100644
index 0000000..53534f3
--- /dev/null
+++ b/parse-time-string/Makefile.local
@@ -0,0 +1,12 @@
+dir := parse-time-string
+extra_cflags += -I$(srcdir)/$(dir)
+
+libparse-time-string_c_srcs := $(dir)/parse-time-string.c
+
+libparse-time-string_modules := $(libparse-time-string_c_srcs:.c=.o)
+
+$(dir)/libparse-time-string.a: $(libparse-time-string_modules)
+	$(call quiet,AR) rcs $@ $^
+
+SRCS := $(SRCS) $(libparse-time-string_c_srcs)
+CLEAN := $(CLEAN) $(libparse-time-string_modules) $(dir)/libparse-time-string.a
diff --git a/parse-time-string/README b/parse-time-string/README
new file mode 100644
index 0000000..300ff1f
--- /dev/null
+++ b/parse-time-string/README
@@ -0,0 +1,9 @@
+PARSE TIME STRING
+=================
+
+parse_time_string() is a date/time parser originally written for
+notmuch by Jani Nikula <jani@nikula.org>. However, there is nothing
+notmuch specific in it, and it should be kept reusable for other
+projects, and ready to be packaged on its own as needed. Please do not
+add dependencies on or references to anything notmuch specific. The
+parser should only depend on the C library.
diff --git a/parse-time-string/parse-time-string.c b/parse-time-string/parse-time-string.c
new file mode 100644
index 0000000..861a705
--- /dev/null
+++ b/parse-time-string/parse-time-string.c
@@ -0,0 +1,1492 @@
+/*
+ * parse time string - user friendly date and time parser
+ * Copyright © 2012 Jani Nikula
+ *
+ * This program is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program.  If not, see <http://www.gnu.org/licenses/>.
+ *
+ * Author: Jani Nikula <jani@nikula.org>
+ */
+
+#include <assert.h>
+#include <ctype.h>
+#include <errno.h>
+#include <limits.h>
+#include <stdio.h>
+#include <stdarg.h>
+#include <stdbool.h>
+#include <stdlib.h>
+#include <string.h>
+#include <strings.h>
+#include <time.h>
+#include <sys/time.h>
+#include <sys/types.h>
+
+#include "parse-time-string.h"
+
+/*
+ * IMPLEMENTATION DETAILS
+ *
+ * At a high level, the parsing is done in two phases: 1) actual
+ * parsing of the input string and storing the parsed data into
+ * 'struct state', and 2) processing of the data in 'struct state'
+ * according to current time (or provided reference time) and
+ * rounding. This is evident in the main entry point function
+ * parse_time_string().
+ *
+ * 1) The parsing phase - parse_input()
+ *
+ * Parsing is greedy and happens from left to right. The parsing is as
+ * unambiguous as possible; only unambiguous date/time formats are
+ * accepted. Redundant or contradictory absolute date/time in the
+ * input (e.g. date specified multiple times/ways) is not
+ * accepted. Relative date/time on the other hand just accumulates if
+ * present multiple times (e.g. "5 days 5 days" just turns into 10
+ * days).
+ *
+ * Parsing decisions are made on the input format, not value. For
+ * example, "20/5/2005" fails because the recognized format here is
+ * MM/D/YYYY, even though the values would suggest DD/M/YYYY.
+ *
+ * Parsing is mostly stateless in the sense that parsing decisions are
+ * not made based on the values of previously parsed data, or whether
+ * certain data is present in the first place. (There are a few
+ * exceptions to the latter part, though, such as parsing of time zone
+ * that would otherwise look like plain time.)
+ *
+ * When the parser encounters a number that is not greedily parsed as
+ * part of a format, the interpretation is postponed until the next
+ * token is parsed. The parser for the next token may consume the
+ * previously postponed number. For example, when parsing "20 May" the
+ * meaning of "20" is not known until "May" is parsed. If the parser
+ * for the next token does not consume the postponed number, the
+ * number is handled as a "lone" number before parser for the next
+ * token finishes.
+ *
+ * 2) The processing phase - create_output()
+ *
+ * Once the parser in phase 1 has finished, 'struct state' contains
+ * all the information from the input string, and it's no longer
+ * needed. Since the parser does not even handle the concept of "now",
+ * the processing initializes the fields referring to the current
+ * date/time.
+ *
+ * If requested, the result is rounded towards past or future. The
+ * idea behind rounding is to support parsing date/time ranges in an
+ * obvious way. For example, for a range defined as two dates (without
+ * time), one would typically want to have an inclusive range from the
+ * beginning of start date to the end of the end date. The caller
+ * would use rounding towards past in the start date, and towards
+ * future in the end date.
+ *
+ * The absolute date and time is shifted by the relative date and
+ * time, and time zone adjustments are made. Daylight saving time
+ * (DST) is specifically *not* handled at all.
+ *
+ * Finally, the result is stored to time_t.
+ */
+
+#define unused(x) x __attribute__ ((unused))
+
+/* XXX: Redefine these to add i18n support. The keyword table uses
+ * N_() to mark strings to be translated; they are accessed
+ * dynamically using _(). */
+#define _(s) (s)	/* i18n: define as gettext (s) */
+#define N_(s) (s)	/* i18n: define as gettext_noop (s) */
+
+#define ARRAY_SIZE(a) (sizeof (a) / sizeof (a[0]))
+
+/*
+ * Field indices in the tm and set arrays of struct state.
+ *
+ * NOTE: There's some code that depends on the ordering of this enum.
+ */
+enum field {
+    /* Keep SEC...YEAR in this order. */
+    TM_ABS_SEC,		/* seconds */
+    TM_ABS_MIN,		/* minutes */
+    TM_ABS_HOUR,	/* hours */
+    TM_ABS_MDAY,	/* day of the month */
+    TM_ABS_MON,		/* month */
+    TM_ABS_YEAR,	/* year */
+
+    TM_ABS_WDAY,	/* day of the week. special: may be relative */
+    TM_ABS_ISDST,	/* daylight saving time */
+
+    TM_AMPM,		/* am vs. pm */
+    TM_TZ,		/* timezone in minutes */
+
+    /* Keep SEC...YEAR in this order. */
+    TM_REL_SEC,		/* seconds relative to absolute or reference time */
+    TM_REL_MIN,		/* minutes ... */
+    TM_REL_HOUR,	/* hours ... */
+    TM_REL_DAY,		/* days ... */
+    TM_REL_MON,		/* months ... */
+    TM_REL_YEAR,	/* years ... */
+    TM_REL_WEEK,	/* weeks ... */
+
+    TM_NONE,		/* not a field */
+
+    TM_SIZE = TM_NONE,
+    TM_FIRST_ABS = TM_ABS_SEC,
+    TM_FIRST_REL = TM_REL_SEC,
+};
+
+/* Values for the set array of struct state. */
+enum field_set {
+    FIELD_UNSET,	/* The field has not been touched by parser. */
+    FIELD_SET,		/* The field has been set by parser. */
+    FIELD_NOW,		/* The field will be set to reference time. */
+};
+
+static enum field
+next_abs_field (enum field field)
+{
+    /* NOTE: Depends on the enum ordering. */
+    return field < TM_ABS_YEAR ? field + 1 : TM_NONE;
+}
+
+static enum field
+abs_to_rel_field (enum field field)
+{
+    assert (field <= TM_ABS_YEAR);
+
+    /* NOTE: Depends on the enum ordering. */
+    return field + (TM_FIRST_REL - TM_FIRST_ABS);
+}
+
+/* Get epoch value for field. */
+static int
+field_epoch (enum field field)
+{
+    if (field == TM_ABS_MDAY || field == TM_ABS_MON)
+	return 1;
+    else if (field == TM_ABS_YEAR)
+	return 1970;
+    else
+	return 0;
+}
+
+/* The parsing state. */
+struct state {
+    int tm[TM_SIZE];			/* parsed date and time */
+    enum field_set set[TM_SIZE];	/* set status of tm */
+
+    enum field last_field;	/* Previously set field. */
+    enum field next_field;	/* Next field for parse_postponed_number() */
+    char delim;
+
+    int postponed_length;	/* Number of digits in postponed value. */
+    int postponed_value;
+    char postponed_delim;	/* The delimiter preceding postponed number. */
+};
+
+/*
+ * Helpers for postponed numbers.
+ *
+ * postponed_length is the number of digits in postponed value. 0
+ * means there is no postponed number. -1 means there is a postponed
+ * number, but it comes from a keyword, and it doesn't have digits.
+ */
+static int
+get_postponed_length (struct state *state)
+{
+    return state->postponed_length;
+}
+
+/*
+ * Consume a previously postponed number. Return true if a number was
+ * in fact postponed, false otherwise. Store the postponed number's
+ * value in *v, length in the input string in *n (or -1 if the number
+ * was written out and parsed as a keyword), and the preceding
+ * delimiter to *d.
+ */
+static bool
+get_postponed_number (struct state *state, int *v, int *n, char *d)
+{
+    if (!state->postponed_length)
+	return false;
+
+    if (n)
+	*n = state->postponed_length;
+
+    if (v)
+	*v = state->postponed_value;
+
+    if (d)
+	*d = state->postponed_delim;
+
+    state->postponed_length = 0;
+    state->postponed_value = 0;
+    state->postponed_delim = 0;
+
+    return true;
+}
+
+/* Parse a previously postponed number if one exists. */
+static int parse_postponed_number (struct state *state, int v, int n, char d);
+static int
+handle_postponed_number (struct state *state, enum field next_field)
+{
+    int v = state->postponed_value;
+    int n = state->postponed_length;
+    char d = state->postponed_delim;
+    int r;
+
+    if (!n)
+	return 0;
+
+    state->postponed_value = 0;
+    state->postponed_length = 0;
+    state->postponed_delim = 0;
+
+    state->next_field = next_field;
+    r = parse_postponed_number (state, v, n, d);
+    state->next_field = TM_NONE;
+
+    return r;
+}
+
+/*
+ * Postpone a number to be handled later. If one exists already,
+ * handle it first. n may be -1 to indicate a keyword that has no
+ * number length.
+ */
+static int
+set_postponed_number (struct state *state, int v, int n)
+{
+    int r;
+    char d = state->delim;
+
+    /* Parse a previously postponed number, if any. */
+    r = handle_postponed_number (state, TM_NONE);
+    if (r)
+	return r;
+
+    state->postponed_length = n;
+    state->postponed_value = v;
+    state->postponed_delim = d;
+
+    return 0;
+}
+
+static void
+set_delim (struct state *state, char delim)
+{
+    state->delim = delim;
+}
+
+static void
+unset_delim (struct state *state)
+{
+    state->delim = 0;
+}
+
+/*
+ * Field set/get/mod helpers.
+ */
+
+/* Return true if field has been set. */
+static bool
+is_field_set (struct state *state, enum field field)
+{
+    assert (field < ARRAY_SIZE (state->tm));
+
+    return field < ARRAY_SIZE (state->set) &&
+	   state->set[field] != FIELD_UNSET;
+}
+
+static void
+unset_field (struct state *state, enum field field)
+{
+    assert (field < ARRAY_SIZE (state->tm));
+
+    state->set[field] = FIELD_UNSET;
+    state->tm[field] = 0;
+}
+
+/*
+ * Set field to value. A field can only be set once to ensure the
+ * input does not contain redundant and potentially conflicting data.
+ */
+static int
+set_field (struct state *state, enum field field, int value)
+{
+    int r;
+
+    assert (field < ARRAY_SIZE (state->tm));
+
+    /* Fields can only be set once. */
+    if (field < ARRAY_SIZE (state->set) && state->set[field] != FIELD_UNSET)
+	return -PARSE_TIME_ERR_ALREADYSET;
+
+    state->set[field] = FIELD_SET;
+
+    /* Parse postponed number, if any. */
+    r = handle_postponed_number (state, field);
+    if (r)
+	return r;
+
+    unset_delim (state);
+
+    state->tm[field] = value;
+    state->last_field = field;
+
+    return 0;
+}
+
+/*
+ * Mark n fields in fields to be set to the reference date/time in the
+ * specified time zone, or local timezone if not specified. The fields
+ * will be initialized after parsing is complete and timezone is
+ * known.
+ */
+static int
+set_fields_to_now (struct state *state, enum field *fields, size_t n)
+{
+    size_t i;
+    int r;
+
+    for (i = 0; i < n; i++) {
+	r = set_field (state, fields[i], 0);
+	if (r)
+	    return r;
+	state->set[fields[i]] = FIELD_NOW;
+    }
+
+    return 0;
+}
+
+/* Modify field by adding value to it. To be used on relative fields,
+ * which can be modified multiple times (to accumulate). */
+static int
+mod_field (struct state *state, enum field field, int value)
+{
+    int r;
+
+    assert (field < ARRAY_SIZE (state->tm));   /* assert relative??? */
+
+    if (field < ARRAY_SIZE (state->set))
+	state->set[field] = FIELD_SET;
+
+    /* Parse postponed number, if any. */
+    r = handle_postponed_number (state, field);
+    if (r)
+	return r;
+
+    unset_delim (state);
+
+    state->tm[field] += value;
+    state->last_field = field;
+
+    return 0;
+}
+
+/*
+ * Get field value. Make sure the field is set before query. It's most
+ * likely an error to call this while parsing (for example fields set
+ * as FIELD_NOW will only be set to some value after parsing).
+ */
+static int
+get_field (struct state *state, enum field field)
+{
+    assert (field < ARRAY_SIZE (state->tm));
+
+    return state->tm[field];
+}
+
+/*
+ * Validity checkers.
+ */
+static bool is_valid_12hour (int h)
+{
+    return h >= 0 && h <= 12;
+}
+
+static bool is_valid_time (int h, int m, int s)
+{
+    /* Allow 24:00:00 to denote end of day. */
+    if (h == 24 && m == 0 && s == 0)
+	return true;
+
+    return h >= 0 && h <= 23 && m >= 0 && m <= 59 && s >= 0 && s <= 59;
+}
+
+static bool is_valid_mday (int mday)
+{
+    return mday >= 1 && mday <= 31;
+}
+
+static bool is_valid_mon (int mon)
+{
+    return mon >= 1 && mon <= 12;
+}
+
+static bool is_valid_year (int year)
+{
+    return year >= 1970;
+}
+
+static bool is_valid_date (int year, int mon, int mday)
+{
+    return is_valid_year (year) && is_valid_mon (mon) && is_valid_mday (mday);
+}
+
+/* Unset indicator for time and date set helpers. */
+#define UNSET -1
+
+/* Time set helper. No input checking. Use UNSET (-1) to leave unset. */
+static int
+set_abs_time (struct state *state, int hour, int min, int sec)
+{
+    int r;
+
+    if (hour != UNSET) {
+	if ((r = set_field (state, TM_ABS_HOUR, hour)))
+	    return r;
+    }
+
+    if (min != UNSET) {
+	if ((r = set_field (state, TM_ABS_MIN, min)))
+	    return r;
+    }
+
+    if (sec != UNSET) {
+	if ((r = set_field (state, TM_ABS_SEC, sec)))
+	    return r;
+    }
+
+    return 0;
+}
+
+/* Date set helper. No input checking. Use UNSET (-1) to leave unset. */
+static int
+set_abs_date (struct state *state, int year, int mon, int mday)
+{
+    int r;
+
+    if (year != UNSET) {
+	if ((r = set_field (state, TM_ABS_YEAR, year)))
+	    return r;
+    }
+
+    if (mon != UNSET) {
+	if ((r = set_field (state, TM_ABS_MON, mon)))
+	    return r;
+    }
+
+    if (mday != UNSET) {
+	if ((r = set_field (state, TM_ABS_MDAY, mday)))
+	    return r;
+    }
+
+    return 0;
+}
+
+/*
+ * Keyword parsing and handling.
+ */
+struct keyword;
+typedef int (*setter_t)(struct state *state, struct keyword *kw);
+
+struct keyword {
+    const char *name;	/* keyword */
+    enum field field;	/* field to set, or FIELD_NONE if N/A */
+    int value;		/* value to set, or 0 if N/A */
+    setter_t set;	/* function to use for setting, if non-NULL */
+};
+
+/*
+ * Setter callback functions for keywords.
+ */
+static int
+kw_set_default (struct state *state, struct keyword *kw)
+{
+    return set_field (state, kw->field, kw->value);
+}
+
+static int
+kw_set_rel (struct state *state, struct keyword *kw)
+{
+    int multiplier = 1;
+
+    /* Get a previously set multiplier, if any. */
+    get_postponed_number (state, &multiplier, NULL, NULL);
+
+    /* Accumulate relative field values. */
+    return mod_field (state, kw->field, multiplier * kw->value);
+}
+
+static int
+kw_set_number (struct state *state, struct keyword *kw)
+{
+    /* -1 = no length, from keyword. */
+    return set_postponed_number (state, kw->value, -1);
+}
+
+static int
+kw_set_month (struct state *state, struct keyword *kw)
+{
+    int n = get_postponed_length (state);
+
+    /* Consume postponed number if it could be mday. This handles "20
+     * January". */
+    if (n == 1 || n == 2) {
+	int r, v;
+
+	get_postponed_number (state, &v, NULL, NULL);
+
+	if (!is_valid_mday (v))
+	    return -PARSE_TIME_ERR_INVALIDDATE;
+
+	r = set_field (state, TM_ABS_MDAY, v);
+	if (r)
+	    return r;
+    }
+
+    return set_field (state, kw->field, kw->value);
+}
+
+static int
+kw_set_ampm (struct state *state, struct keyword *kw)
+{
+    int n = get_postponed_length (state);
+
+    /* Consume postponed number if it could be hour. This handles
+     * "5pm". */
+    if (n == 1 || n == 2) {
+	int r, v;
+
+	get_postponed_number (state, &v, NULL, NULL);
+
+	if (!is_valid_12hour (v))
+	    return -PARSE_TIME_ERR_INVALIDTIME;
+
+	r = set_abs_time (state, v, 0, 0);
+	if (r)
+	    return r;
+    }
+
+    return set_field (state, kw->field, kw->value);
+}
+
+static int
+kw_set_timeofday (struct state *state, struct keyword *kw)
+{
+    return set_abs_time (state, kw->value, 0, 0);
+}
+
+static int
+kw_set_today (struct state *state, unused (struct keyword *kw))
+{
+    enum field fields[] = { TM_ABS_YEAR, TM_ABS_MON, TM_ABS_MDAY };
+
+    return set_fields_to_now (state, fields, ARRAY_SIZE (fields));
+}
+
+static int
+kw_set_now (struct state *state, unused (struct keyword *kw))
+{
+    enum field fields[] = { TM_ABS_HOUR, TM_ABS_MIN, TM_ABS_SEC };
+
+    return set_fields_to_now (state, fields, ARRAY_SIZE (fields));
+}
+
+static int
+kw_set_ordinal (struct state *state, struct keyword *kw)
+{
+    int n, v;
+
+    /* Require a postponed number. */
+    if (!get_postponed_number (state, &v, &n, NULL))
+	return -PARSE_TIME_ERR_DATEFORMAT;
+
+    /* Ordinals are mday. */
+    if (n != 1 && n != 2)
+	return -PARSE_TIME_ERR_DATEFORMAT;
+
+    /* Be strict about st, nd, rd, and lax about th. */
+    if (strcasecmp (kw->name, "st") == 0 && v != 1 && v != 21 && v != 31)
+	return -PARSE_TIME_ERR_INVALIDDATE;
+    else if (strcasecmp (kw->name, "nd") == 0 && v != 2 && v != 22)
+	return -PARSE_TIME_ERR_INVALIDDATE;
+    else if (strcasecmp (kw->name, "rd") == 0 && v != 3 && v != 23)
+	return -PARSE_TIME_ERR_INVALIDDATE;
+    else if (strcasecmp (kw->name, "th") == 0 && !is_valid_mday (v))
+	return -PARSE_TIME_ERR_INVALIDDATE;
+
+    return set_field (state, TM_ABS_MDAY, v);
+}
+
+/*
+ * Accepted keywords.
+ *
+ * A keyword may optionally contain a '|' to indicate the minimum
+ * match length. Without one, full match is required. It's advisable
+ * to keep the minimum match parts unique across all keywords.
+ *
+ * If keyword begins with upper case letter, then the matching will be
+ * case sensitive. Otherwise the matching is case insensitive.
+ *
+ * If setter is NULL, set_default will be used.
+ *
+ * Note: Order matters. Matching is greedy, longest match is used, but
+ * of equal length matches the first one is used, unless there's an
+ * equal length case sensitive match which trumps case insensitive
+ * matches.
+ */
+static struct keyword keywords[] = {
+    /* Weekdays. */
+    { N_("sun|day"),	TM_ABS_WDAY,	0,	NULL },
+    { N_("mon|day"),	TM_ABS_WDAY,	1,	NULL },
+    { N_("tue|sday"),	TM_ABS_WDAY,	2,	NULL },
+    { N_("wed|nesday"),	TM_ABS_WDAY,	3,	NULL },
+    { N_("thu|rsday"),	TM_ABS_WDAY,	4,	NULL },
+    { N_("fri|day"),	TM_ABS_WDAY,	5,	NULL },
+    { N_("sat|urday"),	TM_ABS_WDAY,	6,	NULL },
+
+    /* Months. */
+    { N_("jan|uary"),	TM_ABS_MON,	1,	kw_set_month },
+    { N_("feb|ruary"),	TM_ABS_MON,	2,	kw_set_month },
+    { N_("mar|ch"),	TM_ABS_MON,	3,	kw_set_month },
+    { N_("apr|il"),	TM_ABS_MON,	4,	kw_set_month },
+    { N_("may"),	TM_ABS_MON,	5,	kw_set_month },
+    { N_("jun|e"),	TM_ABS_MON,	6,	kw_set_month },
+    { N_("jul|y"),	TM_ABS_MON,	7,	kw_set_month },
+    { N_("aug|ust"),	TM_ABS_MON,	8,	kw_set_month },
+    { N_("sep|tember"),	TM_ABS_MON,	9,	kw_set_month },
+    { N_("oct|ober"),	TM_ABS_MON,	10,	kw_set_month },
+    { N_("nov|ember"),	TM_ABS_MON,	11,	kw_set_month },
+    { N_("dec|ember"),	TM_ABS_MON,	12,	kw_set_month },
+
+    /* Durations. */
+    { N_("y|ears"),	TM_REL_YEAR,	1,	kw_set_rel },
+    { N_("w|eeks"),	TM_REL_WEEK,	1,	kw_set_rel },
+    { N_("d|ays"),	TM_REL_DAY,	1,	kw_set_rel },
+    { N_("h|ours"),	TM_REL_HOUR,	1,	kw_set_rel },
+    { N_("hr|s"),	TM_REL_HOUR,	1,	kw_set_rel },
+    { N_("m|inutes"),	TM_REL_MIN,	1,	kw_set_rel },
+    /* M=months, m=minutes */
+    { N_("M"),		TM_REL_MON,	1,	kw_set_rel },
+    { N_("mins"),	TM_REL_MIN,	1,	kw_set_rel },
+    { N_("mo|nths"),	TM_REL_MON,	1,	kw_set_rel },
+    { N_("s|econds"),	TM_REL_SEC,	1,	kw_set_rel },
+    { N_("secs"),	TM_REL_SEC,	1,	kw_set_rel },
+
+    /* Numbers. */
+    { N_("one"),	TM_NONE,	1,	kw_set_number },
+    { N_("two"),	TM_NONE,	2,	kw_set_number },
+    { N_("three"),	TM_NONE,	3,	kw_set_number },
+    { N_("four"),	TM_NONE,	4,	kw_set_number },
+    { N_("five"),	TM_NONE,	5,	kw_set_number },
+    { N_("six"),	TM_NONE,	6,	kw_set_number },
+    { N_("seven"),	TM_NONE,	7,	kw_set_number },
+    { N_("eight"),	TM_NONE,	8,	kw_set_number },
+    { N_("nine"),	TM_NONE,	9,	kw_set_number },
+    { N_("ten"),	TM_NONE,	10,	kw_set_number },
+    { N_("dozen"),	TM_NONE,	12,	kw_set_number },
+    { N_("hundred"),	TM_NONE,	100,	kw_set_number },
+
+    /* Special number forms. */
+    { N_("this"),	TM_NONE,	0,	kw_set_number },
+    { N_("last"),	TM_NONE,	1,	kw_set_number },
+
+    /* Other special keywords. */
+    { N_("yesterday"),	TM_REL_DAY,	1,	kw_set_rel },
+    { N_("today"),	TM_NONE,	0,	kw_set_today },
+    { N_("now"),	TM_NONE,	0,	kw_set_now },
+    { N_("noon"),	TM_NONE,	12,	kw_set_timeofday },
+    { N_("midnight"),	TM_NONE,	0,	kw_set_timeofday },
+    { N_("am"),		TM_AMPM,	0,	kw_set_ampm },
+    { N_("a.m."),	TM_AMPM,	0,	kw_set_ampm },
+    { N_("pm"),		TM_AMPM,	1,	kw_set_ampm },
+    { N_("p.m."),	TM_AMPM,	1,	kw_set_ampm },
+    { N_("st"),		TM_NONE,	0,	kw_set_ordinal },
+    { N_("nd"),		TM_NONE,	0,	kw_set_ordinal },
+    { N_("rd"),		TM_NONE,	0,	kw_set_ordinal },
+    { N_("th"),		TM_NONE,	0,	kw_set_ordinal },
+
+    /* Timezone codes: offset in minutes. XXX: Add more codes. */
+    { N_("pst"),	TM_TZ,		-8*60,	NULL },
+    { N_("mst"),	TM_TZ,		-7*60,	NULL },
+    { N_("cst"),	TM_TZ,		-6*60,	NULL },
+    { N_("est"),	TM_TZ,		-5*60,	NULL },
+    { N_("ast"),	TM_TZ,		-4*60,	NULL },
+    { N_("nst"),	TM_TZ,		-(3*60+30),	NULL },
+
+    { N_("gmt"),	TM_TZ,		0,	NULL },
+    { N_("utc"),	TM_TZ,		0,	NULL },
+
+    { N_("wet"),	TM_TZ,		0,	NULL },
+    { N_("cet"),	TM_TZ,		1*60,	NULL },
+    { N_("eet"),	TM_TZ,		2*60,	NULL },
+    { N_("fet"),	TM_TZ,		3*60,	NULL },
+
+    { N_("wat"),	TM_TZ,		1*60,	NULL },
+    { N_("cat"),	TM_TZ,		2*60,	NULL },
+    { N_("eat"),	TM_TZ,		3*60,	NULL },
+};
+
+/*
+ * Compare strings s and keyword. Return number of matching chars on
+ * match, 0 for no match. Match must be at least n chars, or all of
+ * keyword if n < 0, otherwise it's not a match. Use match_case for
+ * case sensitive matching.
+ */
+static size_t
+stringcmp (const char *s, const char *keyword, ssize_t n, bool match_case)
+{
+    ssize_t i;
+
+    if (!n)
+	return 0;
+
+    for (i = 0; *s && *keyword; i++, s++, keyword++) {
+	if (match_case) {
+	    if (*s != *keyword)
+		break;
+	} else {
+	    if (tolower ((unsigned char) *s) !=
+		tolower ((unsigned char) *keyword))
+		break;
+	}
+    }
+
+    if (n > 0)
+	return i < n ? 0 : i;
+    else
+	return *keyword ? 0 : i;
+}
+
+/*
+ * Parse a keyword. Return < 0 on error, number of parsed chars on
+ * success.
+ */
+static ssize_t
+parse_keyword (struct state *state, const char *s)
+{
+    unsigned int i;
+    size_t n, max_n = 0;
+    struct keyword *kw = NULL;
+    int r;
+
+    /* Match longest keyword */
+    for (i = 0; i < ARRAY_SIZE (keywords); i++) {
+	/* Match case if keyword begins with upper case letter. */
+	bool mcase = isupper ((unsigned char) keywords[i].name[0]);
+	ssize_t minlen = -1;
+	char keyword[128];
+	char *p;
+
+	strncpy (keyword, _(keywords[i].name), sizeof (keyword));
+
+	/* Truncate too long keywords. XXX: Make this dynamic? */
+	keyword[sizeof (keyword) - 1] = '\0';
+
+	/* Minimum match length. */
+	p = strchr (keyword, '|');
+	if (p) {
+	    minlen = p - keyword;
+
+	    /* Remove the minimum match length separator. */
+	    memmove (p, p + 1, strlen (p + 1) + 1);
+	}
+
+	n = stringcmp (s, keyword, minlen, mcase);
+	if (n > max_n || (n == max_n && mcase)) {
+	    max_n = n;
+	    kw = &keywords[i];
+	}
+    }
+
+    if (!kw)
+	return -PARSE_TIME_ERR_KEYWORD;
+
+    if (kw->set)
+	r = kw->set (state, kw);
+    else
+	r = kw_set_default (state, kw);
+
+    if (r < 0)
+	return r;
+
+    return max_n;
+}
+
+/*
+ * Non-keyword parsers and their helpers.
+ */
+
+static int
+set_user_tz (struct state *state, char sign, int hour, int min)
+{
+    int tz = hour * 60 + min;
+
+    assert (sign == '+' || sign == '-');
+
+    if (hour < 0 || hour > 14 || min < 0 || min > 59 || min % 15)
+	return -PARSE_TIME_ERR_INVALIDTIME;
+
+    if (sign == '-')
+	tz = -tz;
+
+    return set_field (state, TM_TZ, tz);
+}
+
+/*
+ * Independent parsing of a postponed number when it wasn't consumed
+ * during parsing of the following token.
+ */
+static int
+parse_postponed_number (struct state *state, int v, int n, char d)
+{
+    if (n == 1 || n == 2) {
+	/* Notable exception: Previous field affects parsing. This
+	 * handles "January 20". */
+	if (state->last_field == TM_ABS_MON) {
+	    /* D[D] */
+	    if (!is_valid_mday (v))
+		return -PARSE_TIME_ERR_INVALIDDATE;
+
+	    return set_field (state, TM_ABS_MDAY, v);
+	} else if (n == 2) {
+	    /* XXX: Only allow if last field is hour, min, or sec? */
+	    if (d == '+' || d == '-') {
+		/* +/-HH */
+		return set_user_tz (state, d, v, 0);
+	    }
+	}
+    } else if (n == 4) {
+	/* Notable exception: Value affects parsing. Time zones are
+	 * always at most 1400 and we don't understand years before
+	 * 1970. */
+	if (!is_valid_year (v)) {
+	    if (d == '+' || d == '-') {
+		/* +/-HHMM */
+		return set_user_tz (state, d, v / 100, v % 100);
+	    }
+	} else {
+	    /* YYYY */
+	    return set_field (state, TM_ABS_YEAR, v);
+	}
+    } else if (n == 6) {
+	/* HHMMSS */
+	int hour = v / 10000;
+	int min = (v / 100) % 100;
+	int sec = v % 100;
+
+	if (!is_valid_time (hour, min, sec))
+	    return -PARSE_TIME_ERR_INVALIDTIME;
+
+	return set_abs_time (state, hour, min, sec);
+    } else if (n == 8) {
+	/* YYYYMMDD */
+	int year = v / 10000;
+	int mon = (v / 100) % 100;
+	int mday = v % 100;
+
+	if (!is_valid_date (year, mon, mday))
+	    return -PARSE_TIME_ERR_INVALIDDATE;
+
+	return set_abs_date (state, year, mon, mday);
+    } else {
+	return -PARSE_TIME_ERR_FORMAT;
+    }
+
+    return -PARSE_TIME_ERR_FORMAT;
+}
+
+static int tm_get_field (const struct tm *tm, enum field field);
+
+static int
+set_timestamp (struct state *state, time_t t)
+{
+    struct tm tm;
+    enum field f;
+    int r;
+
+    if (gmtime_r (&t, &tm) == NULL)
+	return -PARSE_TIME_ERR_LIB;
+
+    for (f = TM_ABS_SEC; f != TM_NONE; f = next_abs_field (f)) {
+	r = set_field (state, f, tm_get_field (&tm, f));
+	if (r)
+	    return r;
+    }
+
+    r = set_field (state, TM_TZ, 0);
+    if (r)
+	return r;
+
+    /* XXX: Prevent TM_AMPM with timestamp, e.g. "@123456 pm" */
+
+    return 0;
+}
+
+/* Parse a single number. Typically postpone parsing until later. */
+static int
+parse_single_number (struct state *state, unsigned long v,
+		     unsigned long n)
+{
+    assert (n);
+
+    if (state->delim == '@')
+	return set_timestamp (state, (time_t) v);
+
+    if (v > INT_MAX)
+	return -PARSE_TIME_ERR_FORMAT;
+
+    return set_postponed_number (state, v, n);
+}
+
+static bool
+is_time_sep (char c)
+{
+    return c == ':';
+}
+
+static bool
+is_date_sep (char c)
+{
+    return c == '/' || c == '-' || c == '.';
+}
+
+static bool
+is_sep (char c)
+{
+    return is_time_sep (c) || is_date_sep (c);
+}
+
+/* Two-digit year: 00...69 is 2000s, 70...99 1900s, if n == 0 keep
+ * unset. */
+static int
+expand_year (unsigned long year, size_t n)
+{
+    if (n == 2) {
+	return (year < 70 ? 2000 : 1900) + year;
+    } else if (n == 4) {
+	return year;
+    } else {
+	return UNSET;
+    }
+}
+
+/* Parse a date number triplet. */
+static int
+parse_date (struct state *state, char sep,
+	    unsigned long v1, unsigned long v2, unsigned long v3,
+	    size_t n1, size_t n2, size_t n3)
+{
+    int year = UNSET, mon = UNSET, mday = UNSET;
+
+    assert (is_date_sep (sep));
+
+    switch (sep) {
+    case '/': /* Date: M[M]/D[D][/YY[YY]] or M[M]/YYYY */
+	if (n1 != 1 && n1 != 2)
+	    return -PARSE_TIME_ERR_DATEFORMAT;
+
+	if ((n2 == 1 || n2 == 2) && (n3 == 0 || n3 == 2 || n3 == 4)) {
+	    /* M[M]/D[D][/YY[YY]] */
+	    year = expand_year (v3, n3);
+	    mon = v1;
+	    mday = v2;
+	} else if (n2 == 4 && n3 == 0) {
+	    /* M[M]/YYYY */
+	    year = v2;
+	    mon = v1;
+	} else {
+	    return -PARSE_TIME_ERR_DATEFORMAT;
+	}
+	break;
+
+    case '-': /* Date: YYYY-MM[-DD] or DD-MM[-YY[YY]] or MM-YYYY */
+	if (n1 == 4 && n2 == 2 && (n3 == 0 || n3 == 2)) {
+	    /* YYYY-MM[-DD] */
+	    year = v1;
+	    mon = v2;
+	    if (n3)
+		mday = v3;
+	} else if (n1 == 2 && n2 == 2 && (n3 == 0 || n3 == 2 || n3 == 4)) {
+	    /* DD-MM[-YY[YY]] */
+	    year = expand_year (v3, n3);
+	    mon = v2;
+	    mday = v1;
+	} else if (n1 == 2 && n2 == 4 && n3 == 0) {
+	    /* MM-YYYY */
+	    year = v2;
+	    mon = v1;
+	} else {
+	    return -PARSE_TIME_ERR_DATEFORMAT;
+	}
+	break;
+
+    case '.': /* Date: D[D].M[M][.[YY[YY]]] */
+	if ((n1 != 1 && n1 != 2) || (n2 != 1 && n2 != 2) ||
+	    (n3 != 0 && n3 != 2 && n3 != 4))
+	    return -PARSE_TIME_ERR_DATEFORMAT;
+
+	year = expand_year (v3, n3);
+	mon = v2;
+	mday = v1;
+	break;
+    }
+
+    if (year != UNSET && !is_valid_year (year))
+	return -PARSE_TIME_ERR_INVALIDDATE;
+
+    if (mon != UNSET && !is_valid_mon (mon))
+	return -PARSE_TIME_ERR_INVALIDDATE;
+
+    if (mday != UNSET && !is_valid_mday (mday))
+	return -PARSE_TIME_ERR_INVALIDDATE;
+
+    return set_abs_date (state, year, mon, mday);
+}
+
+/* Parse a time number triplet. */
+static int
+parse_time (struct state *state, char sep,
+	    unsigned long v1, unsigned long v2, unsigned long v3,
+	    size_t n1, size_t n2, size_t n3)
+{
+    assert (is_time_sep (sep));
+
+    if ((n1 != 1 && n1 != 2) || n2 != 2 || (n3 != 0 && n3 != 2))
+	return -PARSE_TIME_ERR_TIMEFORMAT;
+
+    /*
+     * Notable exception: Previously set fields affect
+     * parsing. Interpret (+|-)HH:MM as time zone only if hour and
+     * minute have been set.
+     *
+     * XXX: This could be fixed by restricting the delimiters
+     * preceding time. For '+' it would be justified, but for '-' it
+     * might be inconvenient. However prefer to allow '-' as an
+     * insignificant delimiter preceding time for convenience, and
+     * handle '+' the same way for consistency between positive and
+     * negative time zones.
+     */
+    if (is_field_set (state, TM_ABS_HOUR) &&
+	is_field_set (state, TM_ABS_MIN) &&
+	n1 == 2 && n2 == 2 && n3 == 0 &&
+	(state->delim == '+' || state->delim == '-')) {
+	return set_user_tz (state, state->delim, v1, v2);
+    }
+
+    if (!is_valid_time (v1, v2, v3))
+	return -PARSE_TIME_ERR_INVALIDTIME;
+
+    return set_abs_time (state, v1, v2, n3 ? v3 : 0);
+}
+
+/* strtoul helper that assigns length. */
+static unsigned long
+strtoul_len (const char *s, const char **endp, size_t *len)
+{
+    unsigned long val = strtoul (s, (char **) endp, 10);
+
+    *len = *endp - s;
+    return val;
+}
+
+/*
+ * Parse a (group of) number(s). Return < 0 on error, number of parsed
+ * chars on success.
+ */
+static ssize_t
+parse_number (struct state *state, const char *s)
+{
+    int r;
+    unsigned long v1, v2, v3 = 0;
+    size_t n1, n2, n3 = 0;
+    const char *p = s;
+    char sep;
+
+    v1 = strtoul_len (p, &p, &n1);
+
+    if (is_sep (*p) && isdigit ((unsigned char) *(p + 1))) {
+	sep = *p;
+	v2 = strtoul_len (p + 1, &p, &n2);
+    } else {
+	/* A single number. */
+	r = parse_single_number (state, v1, n1);
+	if (r)
+	    return r;
+
+	return p - s;
+    }
+
+    /* A group of two or three numbers? */
+    if (*p == sep && isdigit ((unsigned char) *(p + 1)))
+	v3 = strtoul_len (p + 1, &p, &n3);
+
+    if (is_time_sep (sep))
+	r = parse_time (state, sep, v1, v2, v3, n1, n2, n3);
+    else
+	r = parse_date (state, sep, v1, v2, v3, n1, n2, n3);
+
+    if (r)
+	return r;
+
+    return p - s;
+}
+
+/*
+ * Parse delimiter(s). Throw away all except the last one, which is
+ * stored for parsing the next non-delimiter. Return < 0 on error,
+ * number of parsed chars on success.
+ *
+ * XXX: We might want to be more strict here.
+ */
+static ssize_t
+parse_delim (struct state *state, const char *s)
+{
+    const char *p = s;
+
+    /*
+     * Skip non-alpha and non-digit, and store the last for further
+     * processing.
+     */
+    while (*p && !isalnum ((unsigned char) *p)) {
+	set_delim (state, *p);
+	p++;
+    }
+
+    return p - s;
+}
+
+/*
+ * Parse a date/time string. Return < 0 on error, number of parsed
+ * chars on success.
+ */
+static ssize_t
+parse_input (struct state *state, const char *s)
+{
+    const char *p = s;
+    ssize_t n;
+    int r;
+
+    while (*p) {
+	if (isalpha ((unsigned char) *p)) {
+	    n = parse_keyword (state, p);
+	} else if (isdigit ((unsigned char) *p)) {
+	    n = parse_number (state, p);
+	} else {
+	    n = parse_delim (state, p);
+	}
+
+	if (n <= 0) {
+	    if (n == 0)
+		n = -PARSE_TIME_ERR;
+
+	    return n;
+	}
+
+	p += n;
+    }
+
+    /* Parse postponed number, if any. */
+    r = handle_postponed_number (state, TM_NONE);
+    if (r < 0)
+	return r;
+
+    return p - s;
+}
+
+/*
+ * Processing the parsed input.
+ */
+
+/*
+ * Initialize reference time to tm. Use time zone in state if
+ * specified, otherwise local time. Use now for reference time if
+ * non-NULL, otherwise current time.
+ */
+static int
+initialize_now (struct state *state, struct tm *tm, const time_t *now)
+{
+    time_t t;
+
+    if (now) {
+	t = *now;
+    } else {
+	if (time (&t) == (time_t) -1)
+	    return -PARSE_TIME_ERR_LIB;
+    }
+
+    if (is_field_set (state, TM_TZ)) {
+	/* Some other time zone. */
+
+	/* Adjust now according to the TZ. */
+	t += get_field (state, TM_TZ) * 60;
+
+	/* It's not gm, but this doesn't mess with the TZ. */
+	if (gmtime_r (&t, tm) == NULL)
+	    return -PARSE_TIME_ERR_LIB;
+    } else {
+	/* Local time. */
+	if (localtime_r (&t, tm) == NULL)
+	    return -PARSE_TIME_ERR_LIB;
+    }
+
+    return 0;
+}
+
+/*
+ * Normalize tm according to mktime(3). Both mktime(3) and
+ * localtime_r(3) use local time, but they cancel each other out here,
+ * making this function agnostic to time zone.
+ */
+static int
+normalize_tm (struct tm *tm)
+{
+    time_t t = mktime (tm);
+
+    if (t == (time_t) -1)
+	return -PARSE_TIME_ERR_LIB;
+
+    if (!localtime_r (&t, tm))
+	return -PARSE_TIME_ERR_LIB;
+
+    return 0;
+}
+
+/* Get field out of a struct tm. */
+static int
+tm_get_field (const struct tm *tm, enum field field)
+{
+    switch (field) {
+    case TM_ABS_SEC:	return tm->tm_sec;
+    case TM_ABS_MIN:	return tm->tm_min;
+    case TM_ABS_HOUR:	return tm->tm_hour;
+    case TM_ABS_MDAY:	return tm->tm_mday;
+    case TM_ABS_MON:	return tm->tm_mon + 1; /* 0- to 1-based */
+    case TM_ABS_YEAR:	return 1900 + tm->tm_year;
+    case TM_ABS_WDAY:	return tm->tm_wday;
+    case TM_ABS_ISDST:	return tm->tm_isdst;
+    default:
+	assert (false);
+	break;
+    }
+
+    return 0;
+}
+
+/* Modify hour according to am/pm setting. */
+static int
+fixup_ampm (struct state *state)
+{
+    int hour, hdiff = 0;
+
+    if (!is_field_set (state, TM_AMPM))
+	return 0;
+
+    if (!is_field_set (state, TM_ABS_HOUR))
+	return -PARSE_TIME_ERR_TIMEFORMAT;
+
+    hour = get_field (state, TM_ABS_HOUR);
+    if (!is_valid_12hour (hour))
+	return -PARSE_TIME_ERR_INVALIDTIME;
+
+    if (get_field (state, TM_AMPM)) {
+	/* 12pm is noon. */
+	if (hour != 12)
+	    hdiff = 12;
+    } else {
+	/* 12am is midnight, beginning of day. */
+	if (hour == 12)
+	    hdiff = -12;
+    }
+
+    mod_field (state, TM_REL_HOUR, -hdiff);
+
+    return 0;
+}
+
+/* Combine absolute and relative fields, and round. */
+static int
+create_output (struct state *state, time_t *t_out, const time_t *ref,
+	       int round)
+{
+    struct tm tm = { .tm_isdst = -1 };
+    struct tm now;
+    time_t t;
+    enum field f;
+    int r;
+    int week_round = PARSE_TIME_NO_ROUND;
+
+    r = initialize_now (state, &now, ref);
+    if (r)
+	return r;
+
+    /* Initialize fields flagged as "now" to reference time. */
+    for (f = TM_ABS_SEC; f != TM_NONE; f = next_abs_field (f)) {
+	if (state->set[f] == FIELD_NOW) {
+	    state->tm[f] = tm_get_field (&now, f);
+	    state->set[f] = FIELD_SET;
+	}
+    }
+
+    /*
+     * If WDAY is set but MDAY is not, we consider WDAY relative
+     *
+     * XXX: This fails on stuff like "two months monday" because two
+     * months ago wasn't the same day as today. Postpone until we know
+     * date?
+     */
+    if (is_field_set (state, TM_ABS_WDAY) &&
+	!is_field_set (state, TM_ABS_MDAY)) {
+	int wday = get_field (state, TM_ABS_WDAY);
+	int today = tm_get_field (&now, TM_ABS_WDAY);
+	int rel_days;
+
+	if (today > wday)
+	    rel_days = today - wday;
+	else
+	    rel_days = today + 7 - wday;
+
+	/* This also prevents special week rounding from happening. */
+	mod_field (state, TM_REL_DAY, rel_days);
+
+	unset_field (state, TM_ABS_WDAY);
+    }
+
+    r = fixup_ampm (state);
+    if (r)
+	return r;
+
+    /*
+     * Iterate fields from most accurate to least accurate, and set
+     * unset fields according to requested rounding.
+     */
+    for (f = TM_ABS_SEC; f != TM_NONE; f = next_abs_field (f)) {
+	if (round != PARSE_TIME_NO_ROUND) {
+	    enum field r = abs_to_rel_field (f);
+
+	    if (is_field_set (state, f) || is_field_set (state, r)) {
+		if (round >= PARSE_TIME_ROUND_UP && f != TM_ABS_SEC) {
+		    mod_field (state, r, -1);
+		    if (round == PARSE_TIME_ROUND_UP_INCLUSIVE)
+			mod_field (state, TM_REL_SEC, 1);
+		}
+		round = PARSE_TIME_NO_ROUND; /* No more rounding. */
+	    } else {
+		if (f == TM_ABS_MDAY &&
+		    is_field_set (state, TM_REL_WEEK)) {
+		    /* Week is most accurate. */
+		    week_round = round;
+		    round = PARSE_TIME_NO_ROUND;
+		} else {
+		    set_field (state, f, field_epoch (f));
+		}
+	    }
+	}
+
+	if (!is_field_set (state, f))
+	    set_field (state, f, tm_get_field (&now, f));
+    }
+
+    /* Special case: rounding with week accuracy. */
+    if (week_round != PARSE_TIME_NO_ROUND) {
+	/* Temporarily set more accurate fields to now. */
+	set_field (state, TM_ABS_SEC, tm_get_field (&now, TM_ABS_SEC));
+	set_field (state, TM_ABS_MIN, tm_get_field (&now, TM_ABS_MIN));
+	set_field (state, TM_ABS_HOUR, tm_get_field (&now, TM_ABS_HOUR));
+	set_field (state, TM_ABS_MDAY, tm_get_field (&now, TM_ABS_MDAY));
+    }
+
+    /*
+     * Set all fields. They may contain out of range values before
+     * normalization by mktime(3).
+     */
+    tm.tm_sec = get_field (state, TM_ABS_SEC) - get_field (state, TM_REL_SEC);
+    tm.tm_min = get_field (state, TM_ABS_MIN) - get_field (state, TM_REL_MIN);
+    tm.tm_hour = get_field (state, TM_ABS_HOUR) - get_field (state, TM_REL_HOUR);
+    tm.tm_mday = get_field (state, TM_ABS_MDAY) -
+		 get_field (state, TM_REL_DAY) - 7 * get_field (state, TM_REL_WEEK);
+    tm.tm_mon = get_field (state, TM_ABS_MON) - get_field (state, TM_REL_MON);
+    tm.tm_mon--; /* 1- to 0-based */
+    tm.tm_year = get_field (state, TM_ABS_YEAR) - get_field (state, TM_REL_YEAR) - 1900;
+
+    /*
+     * It's always normal time.
+     *
+     * XXX: This is probably not a solution that universally
+     * works. Just make sure DST is not taken into account. We don't
+     * want rounding to be affected by DST.
+     */
+    tm.tm_isdst = -1;
+
+    /* Special case: rounding with week accuracy. */
+    if (week_round != PARSE_TIME_NO_ROUND) {
+	/* Normalize to get proper tm.wday. */
+	r = normalize_tm (&tm);
+	if (r < 0)
+	    return r;
+
+	/* Set more accurate fields back to zero. */
+	tm.tm_sec = 0;
+	tm.tm_min = 0;
+	tm.tm_hour = 0;
+	tm.tm_isdst = -1;
+
+	/* Monday is the true 1st day of week, but this is easier. */
+	if (week_round >= PARSE_TIME_ROUND_UP) {
+	    tm.tm_mday += 7 - tm.tm_wday;
+	    if (week_round == PARSE_TIME_ROUND_UP_INCLUSIVE)
+		tm.tm_sec--;
+	} else {
+	    tm.tm_mday -= tm.tm_wday;
+	}
+    }
+
+    if (is_field_set (state, TM_TZ)) {
+	/* tm is in specified TZ, convert to UTC for timegm(3). */
+	tm.tm_min -= get_field (state, TM_TZ);
+	t = timegm (&tm);
+    } else {
+	/* tm is in local time. */
+	t = mktime (&tm);
+    }
+
+    if (t == (time_t) -1)
+	return -PARSE_TIME_ERR_LIB;
+
+    *t_out = t;
+
+    return 0;
+}
+
+/* Internally, all errors are < 0. parse_time_string() returns errors > 0. */
+#define EXTERNAL_ERR(r) (-r)
+
+int
+parse_time_string (const char *s, time_t *t, const time_t *ref, int round)
+{
+    struct state state = { .last_field = TM_NONE };
+    int r;
+
+    if (!s || !t)
+	return EXTERNAL_ERR (-PARSE_TIME_ERR);
+
+    r = parse_input (&state, s);
+    if (r < 0)
+	return EXTERNAL_ERR (r);
+
+    r = create_output (&state, t, ref, round);
+    if (r < 0)
+	return EXTERNAL_ERR (r);
+
+    return 0;
+}
diff --git a/parse-time-string/parse-time-string.h b/parse-time-string/parse-time-string.h
new file mode 100644
index 0000000..bfa4ee3
--- /dev/null
+++ b/parse-time-string/parse-time-string.h
@@ -0,0 +1,102 @@
+/*
+ * parse time string - user friendly date and time parser
+ * Copyright © 2012 Jani Nikula
+ *
+ * This program is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program.  If not, see <http://www.gnu.org/licenses/>.
+ *
+ * Author: Jani Nikula <jani@nikula.org>
+ */
+
+#ifndef PARSE_TIME_STRING_H
+#define PARSE_TIME_STRING_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <time.h>
+
+/* return values for parse_time_string() */
+enum {
+    PARSE_TIME_OK = 0,
+    PARSE_TIME_ERR,		/* unspecified error */
+    PARSE_TIME_ERR_LIB,		/* library call failed */
+    PARSE_TIME_ERR_ALREADYSET,	/* attempt to set unit twice */
+    PARSE_TIME_ERR_FORMAT,	/* generic date/time format error */
+    PARSE_TIME_ERR_DATEFORMAT,	/* date format error */
+    PARSE_TIME_ERR_TIMEFORMAT,	/* time format error */
+    PARSE_TIME_ERR_INVALIDDATE,	/* date value error */
+    PARSE_TIME_ERR_INVALIDTIME,	/* time value error */
+    PARSE_TIME_ERR_KEYWORD,	/* unknown keyword */
+};
+
+/* round values for parse_time_string() */
+enum {
+    PARSE_TIME_ROUND_DOWN = -1,
+    PARSE_TIME_NO_ROUND = 0,
+    PARSE_TIME_ROUND_UP = 1,
+    PARSE_TIME_ROUND_UP_INCLUSIVE = 2,
+};
+
+/**
+ * parse_time_string() - user friendly date and time parser
+ * @s:		string to parse
+ * @t:		pointer to time_t to store parsed time in
+ * @ref:	pointer to time_t containing reference date/time, or NULL
+ * @round:	PARSE_TIME_NO_ROUND, PARSE_TIME_ROUND_DOWN, or
+ *		PARSE_TIME_ROUND_UP
+ *
+ * Parse a date/time string 's' and store the parsed date/time result
+ * in 't'.
+ *
+ * A reference date/time is used for determining the "date/time units"
+ * (roughly equivalent to struct tm members) not specified by 's'. If
+ * 'ref' is non-NULL, it must contain a pointer to a time_t to be used
+ * as reference date/time. Otherwise, the current time is used.
+ *
+ * If 's' does not specify a full date/time, the 'round' parameter
+ * specifies if and how the result should be rounded as follows:
+ *
+ *   PARSE_TIME_NO_ROUND: All date/time units that are not specified
+ *   by 's' are set to the corresponding unit derived from the
+ *   reference date/time.
+ *
+ *   PARSE_TIME_ROUND_DOWN: All date/time units that are more accurate
+ *   than the most accurate unit specified by 's' are set to the
+ *   smallest valid value for that unit. Rest of the unspecified units
+ *   are set as in PARSE_TIME_NO_ROUND.
+ *
+ *   PARSE_TIME_ROUND_UP: All date/time units that are more accurate
+ *   than the most accurate unit specified by 's' are set to the
+ *   smallest valid value for that unit. The most accurate unit
+ *   specified by 's' is incremented by one (and this is rolled over
+ *   to the less accurate units as necessary), unless the most
+ *   accurate unit is seconds. Rest of the unspecified units are set
+ *   as in PARSE_TIME_NO_ROUND.
+ *
+ *   PARSE_TIME_ROUND_UP_INCLUSIVE: Same as PARSE_TIME_ROUND_UP, minus
+ *   one second, unless the most accurate unit specified by 's' is
+ *   seconds. This is useful for callers that require a value for
+ *   inclusive comparison of the result.
+ *
+ * Return 0 (PARSE_TIME_OK) for succesfully parsed date/time, or one
+ * of PARSE_TIME_ERR_* on error. 't' is not modified on error.
+ */
+int parse_time_string (const char *s, time_t *t, const time_t *ref, int round);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* PARSE_TIME_STRING_H */
-- 
1.7.9.5

[PATCH v4 3/9] test: add new test tool parse-time for date/time parser

[Jani Nikula]

Add a smoke testing tool to support testing the date/time parser
module directly and independent of the rest of notmuch.

Credits to Michal Sojka <sojkam1@fel.cvut.cz> for the stdin parsing
idea and consequent massive improvement in testability.
---
 test/Makefile.local |    7 +-
 test/basic          |    2 +-
 test/parse-time.c   |  281 +++++++++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 288 insertions(+), 2 deletions(-)
 create mode 100644 test/parse-time.c

diff --git a/test/Makefile.local b/test/Makefile.local
index 45df4c7..9ae130a 100644
--- a/test/Makefile.local
+++ b/test/Makefile.local
@@ -19,9 +19,13 @@ $(dir)/smtp-dummy: $(smtp_dummy_modules)
 $(dir)/symbol-test: $(dir)/symbol-test.o
 	$(call quiet,CXX) $^ -o $@ -Llib -lnotmuch $(XAPIAN_LDFLAGS)
 
+$(dir)/parse-time: $(dir)/parse-time.o parse-time-string/parse-time-string.o
+	$(call quiet,CC) $^ -o $@
+
 .PHONY: test check
 
-test-binaries: $(dir)/arg-test $(dir)/smtp-dummy $(dir)/symbol-test
+test-binaries: $(dir)/arg-test $(dir)/smtp-dummy $(dir)/symbol-test \
+	$(dir)/parse-time
 
 test:	all test-binaries
 	@${dir}/notmuch-test $(OPTIONS)
@@ -32,4 +36,5 @@ SRCS := $(SRCS) $(smtp_dummy_srcs)
 CLEAN := $(CLEAN) $(dir)/smtp-dummy $(dir)/smtp-dummy.o \
 	 $(dir)/symbol-test $(dir)/symbol-test.o \
 	 $(dir)/arg-test $(dir)/arg-test.o \
+	 $(dir)/parse-time $(dir)/parse-time.o \
 	 $(dir)/corpus.mail $(dir)/test-results $(dir)/tmp.*
diff --git a/test/basic b/test/basic
index 3b635c8..c47197c 100755
--- a/test/basic
+++ b/test/basic
@@ -54,7 +54,7 @@ test_begin_subtest 'Ensure that all available tests will be run by notmuch-test'
 eval $(sed -n -e '/^TESTS="$/,/^"$/p' $TEST_DIRECTORY/notmuch-test)
 tests_in_suite=$(for i in $TESTS; do echo $i; done | sort)
 available=$(find "$TEST_DIRECTORY" -maxdepth 1 -type f -perm +111 | \
-    sed -r -e "s,.*/,," -e "/^(aggregate-results.sh|notmuch-test|smtp-dummy|test-verbose|symbol-test|arg-test)$/d" | \
+    sed -r -e "s,.*/,," -e "/^(aggregate-results.sh|notmuch-test|smtp-dummy|test-verbose|symbol-test|arg-test|parse-time)$/d" | \
     sort)
 test_expect_equal "$tests_in_suite" "$available"
 
diff --git a/test/parse-time.c b/test/parse-time.c
new file mode 100644
index 0000000..5f73b85
--- /dev/null
+++ b/test/parse-time.c
@@ -0,0 +1,281 @@
+/*
+ * parse time string - user friendly date and time parser
+ * Copyright © 2012 Jani Nikula
+ *
+ * This program is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program.  If not, see <http://www.gnu.org/licenses/>.
+ *
+ * Author: Jani Nikula <jani@nikula.org>
+ */
+
+#include <assert.h>
+#include <ctype.h>
+#include <getopt.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "parse-time-string.h"
+
+#define ARRAY_SIZE(a) (sizeof (a) / sizeof (a[0]))
+
+/*
+ * concat argv[start]...argv[end - 1], separating them by a single
+ * space, to a malloced string
+ */
+static char *
+concat_args (int start, int end, char *argv[])
+{
+    int i;
+    size_t len = 1;
+    char *p;
+
+    for (i = start; i < end; i++)
+	len += strlen (argv[i]) + 1;
+
+    p = malloc (len);
+    if (!p)
+	return NULL;
+
+    *p = 0;
+
+    for (i = start; i < end; i++) {
+	if (i != start)
+	    strcat (p, " ");
+	strcat (p, argv[i]);
+    }
+
+    return p;
+}
+
+#define DEFAULT_FORMAT "%a %b %d %T %z %Y"
+
+static void
+usage (const char *name)
+{
+    printf ("Usage: %s [options ...] [<date/time>]\n\n", name);
+    printf (
+	"Parse <date/time> and display it in given format. If <date/time> is\n"
+	"not given, parse each line in stdin according to:\n\n"
+	"  <date/time> [(==>|==_>|==^>|==^^>)<ignored>] [#<comment>]\n\n"
+	"and produce output:\n\n"
+	"  <date/time> (==>|==_>|==^>|==^^>) <time in --format=FMT> [#<comment>]\n\n"
+	"preserving whitespace and comment in input. The operators ==>, ==_>,\n"
+	"==^>, and ==^^> define rounding as no rounding, round down, round up\n"
+	"inclusive, and round up, respectively.\n\n"
+
+	"  -f, --format=FMT output format, FMT according to strftime(3)\n"
+	"                   (default: \"%s\")\n"
+	"  -r, --ref=N      use N seconds since epoch as reference time\n"
+	"                   (default: now)\n"
+	"  -u, --^          round result up inclusive (default: no rounding)\n"
+	"  -U, --^^         round result up (default: no rounding)\n"
+	"  -d, --_          round result down (default: no rounding)\n"
+	"  -h, --help       print this help\n",
+	DEFAULT_FORMAT);
+}
+
+struct {
+    const char *operator;
+    int round;
+} operators[] = {
+    { "==>",	PARSE_TIME_NO_ROUND },
+    { "==_>",	PARSE_TIME_ROUND_DOWN },
+    { "==^>",	PARSE_TIME_ROUND_UP_INCLUSIVE },
+    { "==^^>",	PARSE_TIME_ROUND_UP },
+};
+
+static const char *
+find_operator_in_string (char *str, char **ptr, int *round)
+{
+    const char *oper = NULL;
+    unsigned int i;
+
+    for (i = 0; i < ARRAY_SIZE (operators); i++) {
+	char *p = strstr (str, operators[i].operator);
+	if (p) {
+	    if (round)
+		*round = operators[i].round;
+	    if (ptr)
+		*ptr = p;
+
+	    oper = operators[i].operator;
+	    break;
+	}
+    }
+
+    return oper;
+}
+
+static const char *
+get_operator (int round)
+{
+    const char *oper = NULL;
+    unsigned int i;
+
+    for (i = 0; i < ARRAY_SIZE(operators); i++) {
+	if (round == operators[i].round) {
+	    oper = operators[i].operator;
+	    break;
+	}
+    }
+
+    return oper;
+}
+
+static int
+parse_stdin (FILE *infile, time_t *ref, int round, const char *format)
+{
+    char *input = NULL;
+    char result[1024];
+    size_t inputsize;
+    ssize_t len;
+    struct tm tm;
+    time_t t;
+    int r;
+
+    while ((len = getline (&input, &inputsize, infile)) != -1) {
+	const char *oper;
+	char *trail, *tmp;
+
+	/* trail is trailing whitespace and (optional) comment */
+	trail = strchr (input, '#');
+	if (!trail)
+	    trail = input + len;
+
+	while (trail > input && isspace ((unsigned char) *(trail-1)))
+	    trail--;
+
+	if (trail == input) {
+	    printf ("%s", input);
+	    continue;
+	}
+
+	tmp = strdup (trail);
+	if (!tmp) {
+	    fprintf (stderr, "strdup() failed\n");
+	    continue;
+	}
+	*trail = '\0';
+	trail = tmp;
+
+	/* operator */
+	oper = find_operator_in_string (input, &tmp, &round);
+	if (oper) {
+	    *tmp = '\0';
+	} else {
+	    oper = get_operator (round);
+	    assert (oper);
+	}
+
+	r = parse_time_string (input, &t, ref, round);
+	if (!r) {
+	    if (!localtime_r (&t, &tm)) {
+		fprintf (stderr, "localtime_r() failed\n");
+		free (trail);
+		continue;
+	    }
+
+	    strftime (result, sizeof (result), format, &tm);
+	} else {
+	    snprintf (result, sizeof (result), "ERROR: %d", r);
+	}
+
+	printf ("%s%s %s%s", input, oper, result, trail);
+	free (trail);
+    }
+
+    free (input);
+
+    return 0;
+}
+
+int
+main (int argc, char *argv[])
+{
+    int r;
+    struct tm tm;
+    time_t result;
+    time_t now;
+    time_t *nowp = NULL;
+    char *argstr;
+    int round = PARSE_TIME_NO_ROUND;
+    char buf[1024];
+    const char *format = DEFAULT_FORMAT;
+    struct option options[] = {
+	{ "help",	no_argument,		NULL,	'h' },
+	{ "^",		no_argument,		NULL,	'u' },
+	{ "^^",		no_argument,		NULL,	'U' },
+	{ "_",		no_argument,		NULL,	'd' },
+	{ "format",	required_argument,	NULL,	'f' },
+	{ "ref",	required_argument,	NULL,	'r' },
+	{ NULL, 0, NULL, 0 },
+    };
+
+    for (;;) {
+	int c;
+
+	c = getopt_long (argc, argv, "huUdf:r:", options, NULL);
+	if (c == -1)
+	    break;
+
+	switch (c) {
+	case 'f':
+	    /* output format */
+	    format = optarg;
+	    break;
+	case 'u':
+	    round = PARSE_TIME_ROUND_UP_INCLUSIVE;
+	    break;
+	case 'U':
+	    round = PARSE_TIME_ROUND_UP;
+	    break;
+	case 'd':
+	    round = PARSE_TIME_ROUND_DOWN;
+	    break;
+	case 'r':
+	    /* specify now in seconds since epoch */
+	    now = (time_t) strtol (optarg, NULL, 10);
+	    if (now >= (time_t) 0)
+		nowp = &now;
+	    break;
+	case 'h':
+	case '?':
+	default:
+	    usage (argv[0]);
+	    return 1;
+	}
+    }
+
+    if (optind == argc)
+	return parse_stdin (stdin, nowp, round, format);
+
+    argstr = concat_args (optind, argc, argv);
+    if (!argstr)
+	return 1;
+
+    r = parse_time_string (argstr, &result, nowp, round);
+
+    free (argstr);
+
+    if (r)
+	return 1;
+
+    if (!localtime_r (&result, &tm))
+	return 1;
+
+    strftime (buf, sizeof (buf), format, &tm);
+    printf ("%s\n", buf);
+
+    return 0;
+}
-- 
1.7.9.5

[PATCH v4 4/9] test: add smoke tests for the date/time parser module

[Jani Nikula]

Test the date/time parser module directly, independent of notmuch,
using the parse-time test tool.

Credits to Michal Sojka <sojkam1@fel.cvut.cz> for writing most of the
tests.
---
 test/notmuch-test      |    1 +
 test/parse-time-string |   71 ++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 72 insertions(+)
 create mode 100755 test/parse-time-string

diff --git a/test/notmuch-test b/test/notmuch-test
index cc732c3..7eadfdf 100755
--- a/test/notmuch-test
+++ b/test/notmuch-test
@@ -60,6 +60,7 @@ TESTS="
   emacs-hello
   emacs-show
   missing-headers
+  parse-time-string
 "
 TESTS=${NOTMUCH_TESTS:=$TESTS}
 
diff --git a/test/parse-time-string b/test/parse-time-string
new file mode 100755
index 0000000..862e701
--- /dev/null
+++ b/test/parse-time-string
@@ -0,0 +1,71 @@
+#!/usr/bin/env bash
+test_description="date/time parser module"
+. ./test-lib.sh
+
+# Sanity/smoke tests for the date/time parser independent of notmuch
+
+_date ()
+{
+    date -d "$*" +%s
+}
+
+_parse_time ()
+{
+    ${TEST_DIRECTORY}/parse-time --format=%s "$*"
+}
+
+test_begin_subtest "date(1) default format without TZ code"
+test_expect_equal "$(_parse_time Fri Aug 3 23:06:06 2012)" "$(_date Fri Aug 3 23:06:06 2012)"
+
+test_begin_subtest "date(1) --rfc-2822 format"
+test_expect_equal "$(_parse_time Fri, 03 Aug 2012 23:07:46 +0100)" "$(_date Fri, 03 Aug 2012 23:07:46 +0100)"
+
+test_begin_subtest "date(1) --rfc=3339=seconds format"
+test_expect_equal "$(_parse_time 2012-08-03 23:09:37+03:00)" "$(_date 2012-08-03 23:09:37+03:00)"
+
+test_begin_subtest "Date parser tests"
+REFERENCE=$(_date Tue Jan 11 11:11:00 +0000 2011)
+cat <<EOF > INPUT
+now          ==> Tue Jan 11 11:11:00 +0000 2011
+2010-1-1     ==> ERROR: 5
+Jan 2        ==> Sun Jan 02 11:11:00 +0000 2011
+Mon          ==> Mon Jan 10 11:11:00 +0000 2011
+last Friday  ==> ERROR: 4
+2 hours ago  ==> ERROR: 1
+last month   ==> Sat Dec 11 11:11:00 +0000 2010
+month ago    ==> ERROR: 1
+8am          ==> Tue Jan 11 08:00:00 +0000 2011
+9:15         ==> Tue Jan 11 09:15:00 +0000 2011
+12:34        ==> Tue Jan 11 12:34:00 +0000 2011
+monday       ==> Mon Jan 10 11:11:00 +0000 2011
+yesterday    ==> Mon Jan 10 11:11:00 +0000 2011
+tomorrow     ==> ERROR: 1
+             ==> Tue Jan 11 11:11:00 +0000 2011 # empty string is reference time
+
+Aug 3 23:06:06 2012             ==> Fri Aug 03 23:06:06 +0000 2012 # date(1) default format without TZ code
+Fri, 03 Aug 2012 23:07:46 +0100 ==> Fri Aug 03 22:07:46 +0000 2012 # rfc-2822
+2012-08-03 23:09:37+03:00       ==> Fri Aug 03 20:09:37 +0000 2012 # rfc-3339 seconds
+
+10s           ==> Tue Jan 11 11:10:50 +0000 2011
+19701223s     ==> Fri May 28 10:37:17 +0000 2010
+19701223      ==> Wed Dec 23 11:11:00 +0000 1970
+
+19701223 +0100 ==> Wed Dec 23 11:11:00 +0000 1970 # Timezone is ignored without an error
+
+today ==^> Tue Jan 11 23:59:59 +0000 2011
+today ==_> Tue Jan 11 00:00:00 +0000 2011
+
+thisweek ==^> Sat Jan 15 23:59:59 +0000 2011
+thisweek ==_> Sun Jan 09 00:00:00 +0000 2011
+
+two months ago==> ERROR: 1 # "ago" is not supported
+two months ==> Thu Nov 11 11:11:00 +0000 2010
+
+@1348569850 ==> Tue Sep 25 10:44:10 +0000 2012
+@10 ==> Thu Jan 01 00:00:10 +0000 1970
+EOF
+
+${TEST_DIRECTORY}/parse-time --ref=${REFERENCE} < INPUT > OUTPUT
+test_expect_equal_file INPUT OUTPUT
+
+test_done
-- 
1.7.9.5

[PATCH v4 5/9] build: build parse-time-string as part of the notmuch lib and static cli

[Jani Nikula]

---
 Makefile.local     |    2 +-
 lib/Makefile.local |    2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/Makefile.local b/Makefile.local
index 7f2d4f1..2b91946 100644
--- a/Makefile.local
+++ b/Makefile.local
@@ -277,7 +277,7 @@ notmuch_client_srcs =		\
 
 notmuch_client_modules = $(notmuch_client_srcs:.c=.o)
 
-notmuch: $(notmuch_client_modules) lib/libnotmuch.a util/libutil.a
+notmuch: $(notmuch_client_modules) lib/libnotmuch.a util/libutil.a parse-time-string/libparse-time-string.a
 	$(call quiet,CXX $(CFLAGS)) $^ $(FINAL_LIBNOTMUCH_LDFLAGS) -o $@
 
 notmuch-shared: $(notmuch_client_modules) lib/$(LINKER_NAME)
diff --git a/lib/Makefile.local b/lib/Makefile.local
index 8a9aa28..d1635cf 100644
--- a/lib/Makefile.local
+++ b/lib/Makefile.local
@@ -70,7 +70,7 @@ $(dir)/libnotmuch.a: $(libnotmuch_modules)
 	$(call quiet,AR) rcs $@ $^
 
 $(dir)/$(LIBNAME): $(libnotmuch_modules) notmuch.sym
-	$(call quiet,CXX $(CXXFLAGS)) $(libnotmuch_modules) $(FINAL_LIBNOTMUCH_LDFLAGS) $(LIBRARY_LINK_FLAG) -o $@ util/libutil.a
+	$(call quiet,CXX $(CXXFLAGS)) $(libnotmuch_modules) $(FINAL_LIBNOTMUCH_LDFLAGS) $(LIBRARY_LINK_FLAG) -o $@ util/libutil.a parse-time-string/libparse-time-string.a
 
 notmuch.sym: $(srcdir)/$(dir)/notmuch.h $(libnotmuch_modules)
 	sh $(srcdir)/$(lib)/gen-version-script.sh $< $(libnotmuch_modules) > $@
-- 
1.7.9.5

[PATCH v4 6/9] lib: add date range query support

[Jani Nikula]

Add a custom value range processor to enable date and time searches of
the form date:since..until, where "since" and "until" are expressions
understood by the previously added date/time parser, to restrict the
results to messages within a particular time range (based on the Date:
header).

If "since" or "until" describes date/time at an accuracy of days or
less, the values are rounded according to the accuracy, towards past
for "since" and towards future for "until". For example,
date:november..yesterday would match from the beginning of November
until the end of yesterday. Expressions such as date:today..today
means since the beginning of today until the end of today.

Open-ended ranges are supported (since Xapian 1.2.1), i.e. you can
specify date:..until or date:since.. to not limit the start or end
date, respectively.

CAVEATS:

Xapian does not support spaces in range expressions. You can replace
the spaces with '_', or (in most cases) '-', or (in some cases) leave
the spaces out altogether.

Entering date:expr without ".." (for example date:yesterday) will not
work as you might expect. You can achieve the expected result by
duplicating the expr both sides of ".." (for example
date:yesterday..yesterday).

Open-ended ranges won't work with pre-1.2.1 Xapian, but they don't
produce an error either.

Signed-off-by: Jani Nikula <jani@nikula.org>
---
 lib/Makefile.local     |    1 +
 lib/database-private.h |    1 +
 lib/database.cc        |    5 +++++
 lib/parse-time-vrp.cc  |   40 ++++++++++++++++++++++++++++++++++++++++
 lib/parse-time-vrp.h   |   19 +++++++++++++++++++
 5 files changed, 66 insertions(+)
 create mode 100644 lib/parse-time-vrp.cc
 create mode 100644 lib/parse-time-vrp.h

diff --git a/lib/Makefile.local b/lib/Makefile.local
index d1635cf..6c0f42f 100644
--- a/lib/Makefile.local
+++ b/lib/Makefile.local
@@ -58,6 +58,7 @@ libnotmuch_c_srcs =		\
 
 libnotmuch_cxx_srcs =		\
 	$(dir)/database.cc	\
+	$(dir)/parse-time-vrp.cc	\
 	$(dir)/directory.cc	\
 	$(dir)/index.cc		\
 	$(dir)/message.cc	\
diff --git a/lib/database-private.h b/lib/database-private.h
index 88532d5..d3e65fd 100644
--- a/lib/database-private.h
+++ b/lib/database-private.h
@@ -52,6 +52,7 @@ struct _notmuch_database {
     Xapian::QueryParser *query_parser;
     Xapian::TermGenerator *term_gen;
     Xapian::ValueRangeProcessor *value_range_processor;
+    Xapian::ValueRangeProcessor *date_range_processor;
 };
 
 /* Return the list of terms from the given iterator matching a prefix.
diff --git a/lib/database.cc b/lib/database.cc
index 761dc1a..4df3217 100644
--- a/lib/database.cc
+++ b/lib/database.cc
@@ -19,6 +19,7 @@
  */
 
 #include "database-private.h"
+#include "parse-time-vrp.h"
 
 #include <iostream>
 
@@ -710,12 +711,14 @@ notmuch_database_open (const char *path,
 	notmuch->term_gen = new Xapian::TermGenerator;
 	notmuch->term_gen->set_stemmer (Xapian::Stem ("english"));
 	notmuch->value_range_processor = new Xapian::NumberValueRangeProcessor (NOTMUCH_VALUE_TIMESTAMP);
+	notmuch->date_range_processor = new ParseTimeValueRangeProcessor (NOTMUCH_VALUE_TIMESTAMP);
 
 	notmuch->query_parser->set_default_op (Xapian::Query::OP_AND);
 	notmuch->query_parser->set_database (*notmuch->xapian_db);
 	notmuch->query_parser->set_stemmer (Xapian::Stem ("english"));
 	notmuch->query_parser->set_stemming_strategy (Xapian::QueryParser::STEM_SOME);
 	notmuch->query_parser->add_valuerangeprocessor (notmuch->value_range_processor);
+	notmuch->query_parser->add_valuerangeprocessor (notmuch->date_range_processor);
 
 	for (i = 0; i < ARRAY_SIZE (BOOLEAN_PREFIX_EXTERNAL); i++) {
 	    prefix_t *prefix = &BOOLEAN_PREFIX_EXTERNAL[i];
@@ -778,6 +781,8 @@ notmuch_database_close (notmuch_database_t *notmuch)
     notmuch->xapian_db = NULL;
     delete notmuch->value_range_processor;
     notmuch->value_range_processor = NULL;
+    delete notmuch->date_range_processor;
+    notmuch->date_range_processor = NULL;
 }
 
 void
diff --git a/lib/parse-time-vrp.cc b/lib/parse-time-vrp.cc
new file mode 100644
index 0000000..7e4eca4
--- /dev/null
+++ b/lib/parse-time-vrp.cc
@@ -0,0 +1,40 @@
+
+#include "database-private.h"
+#include "parse-time-vrp.h"
+#include "parse-time-string.h"
+
+#define PREFIX "date:"
+
+/* See *ValueRangeProcessor in xapian-core/api/valuerangeproc.cc */
+Xapian::valueno
+ParseTimeValueRangeProcessor::operator() (std::string &begin, std::string &end)
+{
+    time_t t, now;
+
+    /* Require date: prefix in start of the range... */
+    if (STRNCMP_LITERAL (begin.c_str (), PREFIX))
+	return Xapian::BAD_VALUENO;
+
+    /* ...and remove it. */
+    begin.erase (0, sizeof (PREFIX) - 1);
+
+    /* Use the same 'now' for begin and end. */
+    if (time (&now) == (time_t) -1)
+	return Xapian::BAD_VALUENO;
+
+    if (!begin.empty ()) {
+	if (parse_time_string (begin.c_str (), &t, &now, PARSE_TIME_ROUND_DOWN))
+	    return Xapian::BAD_VALUENO;
+
+	begin.assign (Xapian::sortable_serialise ((double) t));
+    }
+
+    if (!end.empty ()) {
+	if (parse_time_string (end.c_str (), &t, &now, PARSE_TIME_ROUND_UP_INCLUSIVE))
+	    return Xapian::BAD_VALUENO;
+
+	end.assign (Xapian::sortable_serialise ((double) t));
+    }
+
+    return valno;
+}
diff --git a/lib/parse-time-vrp.h b/lib/parse-time-vrp.h
new file mode 100644
index 0000000..526c217
--- /dev/null
+++ b/lib/parse-time-vrp.h
@@ -0,0 +1,19 @@
+
+#ifndef NOTMUCH_PARSE_TIME_VRP_H
+#define NOTMUCH_PARSE_TIME_VRP_H
+
+#include <xapian.h>
+
+/* see *ValueRangeProcessor in xapian-core/include/xapian/queryparser.h */
+class ParseTimeValueRangeProcessor : public Xapian::ValueRangeProcessor {
+protected:
+    Xapian::valueno valno;
+
+public:
+    ParseTimeValueRangeProcessor (Xapian::valueno slot_)
+	: valno(slot_) { }
+
+    Xapian::valueno operator() (std::string &begin, std::string &end);
+};
+
+#endif /* NOTMUCH_PARSE_TIME_VRP_H */
-- 
1.7.9.5

[PATCH v4 7/9] test: add tests for date:since..until range queries

[Jani Nikula]

A brief initial test set.
---
 test/notmuch-test |    1 +
 test/search-date  |   21 +++++++++++++++++++++
 2 files changed, 22 insertions(+)
 create mode 100755 test/search-date

diff --git a/test/notmuch-test b/test/notmuch-test
index 7eadfdf..9a1b375 100755
--- a/test/notmuch-test
+++ b/test/notmuch-test
@@ -61,6 +61,7 @@ TESTS="
   emacs-show
   missing-headers
   parse-time-string
+  search-date
 "
 TESTS=${NOTMUCH_TESTS:=$TESTS}
 
diff --git a/test/search-date b/test/search-date
new file mode 100755
index 0000000..70bcf34
--- /dev/null
+++ b/test/search-date
@@ -0,0 +1,21 @@
+#!/usr/bin/env bash
+test_description="date:since..until queries"
+. ./test-lib.sh
+
+add_email_corpus
+
+test_begin_subtest "Absolute date range"
+output=$(notmuch search date:2010-12-16..12/16/2010 | notmuch_search_sanitize)
+test_expect_equal "$output" "thread:XXX   2010-12-16 [1/1] Olivier Berger; Essai accentué (inbox unread)"
+
+test_begin_subtest "Absolute time range with TZ"
+notmuch search date:18-Nov-2009_02:19:26-0800..2009-11-18_04:49:52-06:00 | notmuch_search_sanitize > OUTPUT
+cat <<EOF >EXPECTED
+thread:XXX   2009-11-18 [1/3] Carl Worth| Jan Janak; [notmuch] What a great idea! (inbox unread)
+thread:XXX   2009-11-18 [1/2] Carl Worth| Jan Janak; [notmuch] [PATCH] Older versions of install do not support -C. (inbox unread)
+thread:XXX   2009-11-18 [1/3] Carl Worth| Aron Griffis, Keith Packard; [notmuch] archive (inbox unread)
+thread:XXX   2009-11-18 [1/2] Carl Worth| Keith Packard; [notmuch] [PATCH] Make notmuch-show 'X' (and 'x') commands remove inbox (and unread) tags (inbox unread)
+EOF
+test_expect_equal_file OUTPUT EXPECTED
+
+test_done
-- 
1.7.9.5

[PATCH v4 8/9] man: document the date:since..until range queries

[Jani Nikula]

---
 man/man7/notmuch-search-terms.7 |  147 +++++++++++++++++++++++++++++++++++----
 1 file changed, 135 insertions(+), 12 deletions(-)

diff --git a/man/man7/notmuch-search-terms.7 b/man/man7/notmuch-search-terms.7
index 17a109e..fbd3ee7 100644
--- a/man/man7/notmuch-search-terms.7
+++ b/man/man7/notmuch-search-terms.7
@@ -54,6 +54,8 @@ terms to match against specific portions of an email, (where
 
 	folder:<directory-path>
 
+	date:<since>..<until>
+
 The
 .B from:
 prefix is used to match the name or address of the sender of an email
@@ -104,6 +106,26 @@ contained within particular directories within the mail store. Only
 the directory components below the top-level mail database path are
 available to be searched.
 
+The
+.B date:
+prefix can be used to restrict the results to only messages within a
+particular time range (based on the Date: header) with a range syntax
+of:
+
+	date:<since>..<until>
+
+See \fBDATE AND TIME SEARCH\fR below for details on the range
+expression, and supported syntax for <since> and <until> date and time
+expressions.
+
+The time range can also be specified using timestamps with a syntax
+of:
+
+	<initial-timestamp>..<final-timestamp>
+
+Each timestamp is a number representing the number of seconds since
+1970\-01\-01 00:00:00 UTC.
+
 In addition to individual terms, multiple terms can be
 combined with Boolean operators (
 .BR and ", " or ", " not
@@ -117,20 +139,121 @@ operators, but will have to be protected from interpretation by the
 shell, (such as by putting quotation marks around any parenthesized
 expression).
 
-Finally, results can be restricted to only messages within a
-particular time range, (based on the Date: header) with a syntax of:
+.SH DATE AND TIME SEARCH
 
-	<initial-timestamp>..<final-timestamp>
+This is a non-exhaustive description of the date and time search with
+some pseudo notation. Most of the constructs can be mixed freely, and
+in any order, but the same absolute date or time can't be expressed
+twice.
 
-Each timestamp is a number representing the number of seconds since
-1970\-01\-01 00:00:00 UTC. This is not the most convenient means of
-expressing date ranges, but until notmuch is fixed to accept a more
-convenient form, one can use the date program to construct
-timestamps. For example, with the bash shell the following syntax would
-specify a date range to return messages from 2009\-10\-01 until the
-current time:
-
-	$(date +%s \-d 2009\-10\-01)..$(date +%s)
+.RS 4
+.TP 4
+.B The range expression
+
+date:<since>..<until>
+
+The above expression restricts the results to only messages from
+<since> to <until>, based on the Date: header.
+
+If <since> or <until> describes time at an accuracy of days or less,
+the date/time is rounded, towards past for <since> and towards future
+for <until>, to be inclusive. For example, date:january..february
+matches from the beginning of January until the end of
+February. Similarly, date:yesterday..yesterday matches from the
+beginning of yesterday until the end of yesterday.
+
+Open-ended ranges are supported (since Xapian 1.2.1), i.e. it's
+possible to specify date:..<until> or date:<since>.. to not limit the
+start or end time, respectively. Unfortunately, pre-1.2.1 Xapian does
+not report an error on open ended ranges, but it does not work as
+expected either.
+
+Xapian does not support spaces in range expressions. You can replace
+the spaces with '_', or (in most cases) '-', or (in some cases) leave
+the spaces out altogether.
+
+Entering date:expr without ".." (for example date:yesterday) won't
+work, as it's not interpreted as a range expression at all. You can
+achieve the expected result by duplicating the expr both sides of ".."
+(for example date:yesterday..yesterday).
+.RE
+
+.RS 4
+.TP 4
+.B Relative date and time
+[N|number] (years|months|weeks|days|hours|hrs|minutes|mins|seconds|secs) [...]
+
+All refer to past, can be repeated and will be accumulated.
+
+Units can be abbreviated to any length, with the otherwise ambiguous
+single m being m for minutes and M for months.
+
+Number multiplier can also be written out one, two, ..., ten, dozen,
+hundred. As special cases last means one ("last week") and this means
+zero ("this month").
+
+When combined with absolute date and time, the relative date and time
+specification will be relative from the specified absolute date and
+time.
+
+Examples: 5M2d, two weeks
+.RE
+
+.RS 4
+.TP 4
+.B Supported time formats
+H[H]:MM[:SS] [(am|a.m.|pm|p.m.)]
+
+H[H] (am|a.m.|pm|p.m.)
+
+HHMMSS
+
+now
+
+noon
+
+midnight
+
+Examples: 17:05, 5pm
+.RE
+
+.RS 4
+.TP 4
+.B Supported date formats
+YYYY-MM[-DD]
+
+DD-MM[-[YY]YY]
+
+MM-YYYY
+
+M[M]/D[D][/[YY]YY]
+
+M[M]/YYYY
+
+D[D].M[M][.[YY]YY]
+
+D[D][(st|nd|rd|th)] Mon[thname] [YYYY]
+
+Mon[thname] D[D][(st|nd|rd|th)] [YYYY]
+
+Wee[kday]
+
+Month names can be abbreviated at three or more characters.
+
+Weekday names can be abbreviated at three or more characters.
+
+Examples: 2012-07-31, 31-07-2012, 7/31/2012, August 3
+.RE
+
+.RS 4
+.TP 4
+.B Time zones
+(+|-)HH:MM
+
+(+|-)HH[MM]
+
+Some time zone codes, e.g. UTC, EET.
+.RE
 
 .SH SEE ALSO
 
-- 
1.7.9.5

[PATCH v4 9/9] NEWS: date range search support

[Jani Nikula]

---
 NEWS |   14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/NEWS b/NEWS
index 2b50ba3..5f5b726 100644
--- a/NEWS
+++ b/NEWS
@@ -1,3 +1,17 @@
+Notmuch 0.15 (YYYY-MM-DD)
+=========================
+
+Library changes
+---------------
+
+Date range search support
+
+  The `date:` prefix can now be used in queries to restrict the results to only
+  messages within a particular time range (based on the Date: header) with a
+  range syntax of `date:<since>..<until>`. Notmuch supports a wide variety of
+  expressions in `<since>` and `<until>`. Please refer to the
+  `notmuch-search-terms(7)` manual page for details.
+
 Notmuch 0.14 (2012-08-20)
 =========================
 
-- 
1.7.9.5

Re: [PATCH v4 2/9] parse-time-string: add a date/time parser to notmuch

[Ethan Glasser-Camp]

Jani Nikula <jani@nikula.org> writes:

Hi! I commend you for your work and persistence. This represents a lot
of work and I think it's good enough to be merged. I would certainly
love to see "last" and "ago" supported but this patch series, and this
patch especially, is cumbersome enough that I'd really rather it be
merged before it gets any worse.

If this patch series isn't accepted, I'd suggest making a patch series
that makes a "null" date parser, which just takes strings of date
timestamps the way notmuch does now: "12345" -> (time_t) 12345. We're
going to need a date parser anyhow, so a patch series that sets up the
scaffolding -- separate directory, tests -- could be merged without the
actual parser. Then you'd have fewer patches to juggle the next time
around.

I agree with Tomi Ollila that whether the parser is built in bison or in
straight C, as this one is, isn't important. The difficult part of
parsing dates isn't the translation from text to parse tree; it's
dealing with all the semantic difficulty that comes YYMMDD versus DDMMYY
and the difference between "last week" and "last Thursday".

I have a few minor quibbles that I would be happy to see addressed after
this was merged, or not at all.

> +/* Parse a previously postponed number if one exists. */
> +static int parse_postponed_number (struct state *state, int v, int n, char d);
> +static int
> +handle_postponed_number (struct state *state, enum field next_field)
> +{
> +    int v = state->postponed_value;
> +    int n = state->postponed_length;
> +    char d = state->postponed_delim;
> +    int r;
> +
> +    if (!n)
> +	return 0;
> +
> +    state->postponed_value = 0;
> +    state->postponed_length = 0;
> +    state->postponed_delim = 0;

This could be refactored to be a call to get_postponed_number. Also, I'd
prefer parse_postponed_number be up here, closer to its sole caller (handle_postponed_number).

> +/*
> + * Postpone a number to be handled later. If one exists already,
> + * handle it first. n may be -1 to indicate a keyword that has no
> + * number length.
> + */
> +static int
> +set_postponed_number (struct state *state, int v, int n)
> +{
> +    int r;
> +    char d = state->delim;
> +
> +    /* Parse a previously postponed number, if any. */
> +    r = handle_postponed_number (state, TM_NONE);
> +    if (r)
> +	return r;

I would love a comment explaining under what circumstances this could
occur and what the caller is expected to do.

> +/*
> + * Accepted keywords.
> + */
> +static struct keyword keywords[] = {
> +    /* Weekdays. */
> +    { N_("sun|day"),	TM_ABS_WDAY,	0,	NULL },
> +    { N_("mon|day"),	TM_ABS_WDAY,	1,	NULL },

Maybe it's just my history with Python, but I'd prefer keywords, which
is a global and a constant, to be written in all caps (KEYWORDS).

> +/*
> + * Compare strings s and keyword. Return number of matching chars on
> + * match, 0 for no match. Match must be at least n chars, or all of
> + * keyword if n < 0, otherwise it's not a match. Use match_case for
> + * case sensitive matching.
> + */
> +static size_t
> +stringcmp (const char *s, const char *keyword, ssize_t n, bool match_case)
> +{

The name of this function makes it look uncomfortably like strcmp(3),
which has a very different calling semantics (specifically the -1, 0, 1
return value). I'd prefer a name like string_match_keyword.

Ethan

Re: [PATCH v4 2/9] parse-time-string: add a date/time parser to notmuch

[Jani Nikula]

On Mon, 15 Oct 2012, Ethan Glasser-Camp <ethan.glasser.camp@gmail.com> wrote:
>> +/* Parse a previously postponed number if one exists. */
>> +static int parse_postponed_number (struct state *state, int v, int n, char d);
>> +static int
>> +handle_postponed_number (struct state *state, enum field next_field)
>> +{
>> +    int v = state->postponed_value;
>> +    int n = state->postponed_length;
>> +    char d = state->postponed_delim;
>> +    int r;
>> +
>> +    if (!n)
>> +	return 0;
>> +
>> +    state->postponed_value = 0;
>> +    state->postponed_length = 0;
>> +    state->postponed_delim = 0;
>
> This could be refactored to be a call to get_postponed_number. Also,
> I'd prefer parse_postponed_number be up here, closer to its sole
> caller (handle_postponed_number).

I decided to nuke the intermediate handle_postponed_number altogether,
and fix parse_postponed_number to call get_postponed_number. Thanks for
pointing this out.

>> +/*
>> + * Postpone a number to be handled later. If one exists already,
>> + * handle it first. n may be -1 to indicate a keyword that has no
>> + * number length.
>> + */
>> +static int
>> +set_postponed_number (struct state *state, int v, int n)
>> +{
>> +    int r;
>> +    char d = state->delim;
>> +
>> +    /* Parse a previously postponed number, if any. */
>> +    r = handle_postponed_number (state, TM_NONE);
>> +    if (r)
>> +	return r;
>
> I would love a comment explaining under what circumstances this could
> occur and what the caller is expected to do.

Any errors anywhere the caller is expected to pop up all the way to the
main entry point. I did not verify, but, for example, I'd expect a
sequence of "2012 2012 2012" to fail right here.

>> +/*
>> + * Accepted keywords.
>> + */
>> +static struct keyword keywords[] = {
>> +    /* Weekdays. */
>> +    { N_("sun|day"),	TM_ABS_WDAY,	0,	NULL },
>> +    { N_("mon|day"),	TM_ABS_WDAY,	1,	NULL },
>
> Maybe it's just my history with Python, but I'd prefer keywords, which
> is a global and a constant, to be written in all caps (KEYWORDS).

It's just your history with Python. ;) IMO it's more in line with
notmuch coding style as it is. It could be made const though.

>> +/*
>> + * Compare strings s and keyword. Return number of matching chars on
>> + * match, 0 for no match. Match must be at least n chars, or all of
>> + * keyword if n < 0, otherwise it's not a match. Use match_case for
>> + * case sensitive matching.
>> + */
>> +static size_t
>> +stringcmp (const char *s, const char *keyword, ssize_t n, bool match_case)
>> +{
>
> The name of this function makes it look uncomfortably like strcmp(3),
> which has a very different calling semantics (specifically the -1, 0, 1
> return value). I'd prefer a name like string_match_keyword.

Agreed.


BR,
Jani.