zsh/Functions/Calendar/calendar_scandate

# Scan a line for various common date and time formats.
# Set REPLY to the number of seconds since the epoch at which that
# time occurs.  The time does not need to be matched; this will
# produce midnight at the start of the date.
#
# Absolute times
#
# The rules below are fairly complicated, to allow any natural (and
# some highly unnatural but nonetheless common) combination of
# time and date used by English speakers.  It is recommended that,
# rather than exploring the intricacies of the system, users find
# a date format that is natural to them and stick to it.  This
# will avoid unexpected effects.  Various key facts should be noted,
# explained in more detail below:
#
# - In particular, note the confusion between month/day/year and
#   day/month/year when the month is numeric; this format should be
#   avoided if at all possible.  Many alternatives are available.
# - However, there is currently no localization support, so month
#   names must be English (though only the first three letters are required).
#   The same applies to days of the week if they occur (they are not useful).
# - The year must be given in full to avoid confusion, and only years
#   from 1900 to 2099 inclusive are matched.
# - Although timezones are parsed (complicated formats may not be recognized),
#   they are then ignored; no time adjustment is made.
#
# The following give some obvious examples; users finding here
# a format they like and not subject to vagaries of style may skip
# the full description.  As dates and times are matched separately
# (even though the time may be embedded in the date), any date format
# may be mixed with any format for the time of day provide the
# separators are clear (whitespace, colons, commas).
#   2007/04/03 13:13
#   2007/04/03:13:13
#   2007/04/03 1:13 pm
#   3rd April 2007, 13:13
#   April 3rd 2007 1:13 p.m.
#   Apr 3, 2007 13:13
#   Tue Apr 03 13:13:00 2007
#   13:13 2007/apr/3
#
# Times are parsed and extracted before dates.  They must use colons
# to separate hours and minutes, though a dot is allowed before seconds
# if they are present.  This limits time formats to
#   HH:MM[:SS[.FFFFF]] [am|pm|a.m.|p.m.]
#   HH:MM.SS[.FFFFF] [am|pm|a.m.|p.m.]
# in which square brackets indicate optional elements, possibly with
# alternatives.  Fractions of a second are recognised but ignored.
# Unless -r is given (see below), a date is mandatory but a time of day is
# not; the time returned is at the start of the date.
#
# Time zones are not handled, though if one is matched following a time
# specification it will be removed to allow a surrounding date to be
# parsed.  This only happens if the format of the timezone is not too
# wacky:
#   +0100
#   GMT
#   GMT-7
#   CET+1CDT
# etc. are all understood, but any part of the timezone that is not numeric
# must have exactly three capital letters in the name.
#
# Dates suffer from the ambiguity between DD/MM/YYYY and MM/DD/YYYY.  It is
# recommended this form is avoided with purely numeric dates, but use of
# ordinals, eg. 3rd/04/2007, will resolve the ambiguity as the ordinal is
# always parsed as the day of the month.  Years must be four digits (and
# the first two must be 19 or 20); 03/04/08 is not recognised.  Other
# numbers may have leading zeroes, but they are not required.  The
# following are handled:
#   YYYY/MM/DD
#   YYYY-MM-DD
#   YYYY/MNM/DD
#   YYYY-MNM-DD
#   DD[th|st|rd] MNM[,] YYYY
#   DD[th|st|rd] MNM[,]            current year assumed
#   MNM DD[th|st|rd][,] YYYY
#   MNM DD[th|st|rd][,]            current year assumed
#   DD[th|st|rd]/MM[,] YYYY
#   DD[th|st|rd]/MM/YYYY
#   MM/DD[th|st|rd][,] YYYY
#   MM/DD[th|st|rd]/YYYY
# Here, MNM is at least the first three letters of a month name,
# matched case-insensitively.  The remainder of the month name may appear but
# its contents are irrelevant, so janissary, febrile, martial, apricot,
# etc. are happily handled.
#
# Note there are only two cases that assume the current year, the
# form "Jun 20" or "14 September" (the only two commonly occurring
# forms, apart from a "the" in some forms of English, which isn't
# currently supported).  Such dates will of course become ambiguous
# in the future, so should ideally be avoided.
#
# Times may follow dates with a colon, e.g. 1965/07/12:09:45; this
# is in order to provide a format with no whitespace.  A comma
# and whitespace are allowed, e.g. "1965/07/12, 09:45".
# Currently the order of these separators is not checked, so
# illogical formats such as "1965/07/12, : ,09:45" will also
# be matched.  Otherwise, a time is only recognised as being associated
# with a date if there is only whitespace in between, or if the time
# was embedded in the date.
#
# Days of the week are not scanned, but will be ignored if they occur
# at the start of the date pattern only.
#
# For example, the standard date format:
#   Fri Aug 18 17:00:48 BST 2006
# is handled by matching HH:MM:SS and removing it together with the
# matched (but unused) time zone.  This leaves the following:
#   Fri Aug 18 2006
# "Fri" is ignored and the rest is matched according to the sixth of
# the standard rules.
#
# Relative times
# ==============
#
# The option -r allows a relative time.  Years (or ys, yrs, or without s),
# months (or mths, mons, mnths, months, or without s --- "m", "ms" and
# "mns" are ambiguous and are not handled), weeks (or ws, wks, or without
# s) and days (or ds, dys, days, or without s), hours (or hs, hrs, with or
# without s), minutes (or mins, with or without s) and seconds (or ss,
# secs, with or without s) are understood.  Spaces between the numbers
# are optional, but are required between items, although a comma
# may be used (with or without spaces).
#
# Note that a year here is 365.25 days and a month is 30 days.  TODO:
# improve this by passing down base time and adjusting.  (This will
# be crucial for events repeating monthly.)  TODO: it then makes
# sense to make PERIODly = 1 PERIOD (also for PERIOD = dai!)
#
# This allows forms like:
#   30 years 3 months 4 days 3:42:41
#   14 days 5 hours
#   4d,10hr
# In this case absolute dates are ignored.

emulate -L zsh
setopt extendedglob

zmodload -i zsh/datetime || return 1

# separator characters before time or between time and date
# allow , - or : before the time: this allows spaceless but still
# relatively logical dates like 2006/09/19:14:27
# don't allow / before time !  the above
# is not 19 hours 14 mins and 27 seconds after anything.
local tschars="[-,:[:space:]]"
# start pattern for time when anchored
local tspat_anchor="(${tschars}#)"
# ... when not anchored
local tspat_noanchor="(|*${tschars})"
# separator characters between elements.  comma is fairly
# natural punctuation; otherwise only allow whitespace.
local schars="[.,[:space:]]"
local daypat="${schars}#(sun|mon|tue|wed|thu|fri|sat)[a-z]#${schars}#"
# Start pattern for date: treat , as space for simplicity.  This
# is illogical at the start but saves lots of minor fiddling later.
# Date start pattern when anchored at the start.
# We need to be able to ignore the day here, although (for consistency
# with the unanchored case) we don't remove it until later.
# (The problem in the other case is that matching anything before
# the day of the week is greedy, so the day of the week gets ignored
# if it's optional.)
local dspat_anchor="(|(#B)${daypat}(#b)${schars}#)"
# Date start pattern when not anchored at the start.
local dspat_noanchor="(|*${schars})"
# end pattern for relative times: similar remark about use of $schars.
local repat="(|s)(|${schars}*)"
# not locale-dependent!  I don't know how to get the months out
# of the system for the purpose of finding out where they occur.
# We may need some completely different heuristic.
local monthpat="(jan|feb|mar|apr|may|jun|jul|aug|sep|oct|nov|dec)[a-z]#"
# days, not handled but we need to ignore them. also not localized.

integer year month day hour minute second
local opt line orig_line mname MATCH MBEGIN MEND tz
local -a match mbegin mend
# Flags that we found a date or a time (maybe a relative time)
integer date_found time_found
# Indices of positions of start and end of time and dates found.
# These are actual character indices as zsh would normally use, i.e.
# line[time_start,time_end] is the string for the time.
integer time_start time_end date_start date_end
integer anchor anchor_end debug relative reladd setvar

while getopts "aAdrs" opt; do
  case $opt in
    (a)
    # anchor
    (( anchor = 1 ))
    ;;

    (A)
    # anchor at end, too
    (( anchor = 1, anchor_end = 1 ))
    ;;

    (d)
    # enable debug output
    (( debug = 1 ))
    ;;

    (r)
    (( relative = 1 ))
    ;;

    (s)
    (( setvar = 1 ))
    ;;

    (*)
    return 1
    ;;
  esac
done
shift $(( OPTIND - 1 ))

line=$1 orig_line=$1

local dspat tspat
if (( anchor )); then
  # Anchored at the start.
  dspat=$dspat_anchor
  if (( relative )); then
    tspat=$tspat_anchor
  else
    # We'll test later if the time is associated with the date.
    tspat=$tspat_noanchor
  fi
else
  dspat=$dspat_noanchor
  tspat=$tspat_noanchor
fi

# Look for a time separately; we need colons for this.
case $line in
  # with seconds, am/pm: don't match / in front.
  ((#ibm)${~tspat}(<0-12>):(<0-59>)[.:]((<0-59>)(.<->|))[[:space:]]#([ap])(|.)[[:space:]]#m(.|[[:space:]]|(#e))(*))
  hour=$match[2]
  minute=$match[3]
  second=$match[5]
  [[ $match[7] = (#i)p ]] && (( hour <= 12 )) && (( hour += 12 ))
  time_found=1
  ;;

  # no seconds, am/pm
  ((#ibm)${~tspat}(<0-12>):(<0-59>)[[:space:]]#([ap])(|.)[[:space:]]#m(.|[[:space:]]|(#e))(*))
  hour=$match[2]
  minute=$match[3]
  [[ $match[4] = (#i)p ]] && (( hour <= 12 )) && (( hour += 12 ))
  time_found=1
  ;;

  # no colon, even, but a.m./p.m. indicator
  ((#ibm)${~tspat}(<0-12>)[[:space:]]#([ap])(|.)[[:space:]]#m(.|[[:space:]]|(#e))(*))
  hour=$match[2]
  minute=0
  [[ $match[3] = (#i)p ]] && (( hour <= 12 )) && (( hour += 12 ))
  time_found=1
  ;;

  # 24 hour clock, with seconds
  ((#ibm)${~tspat}(<0-24>):(<0-59>)[.:]((<0-59>)(.<->|))(*))
  hour=$match[2]
  minute=$match[3]
  second=$match[5]
  time_found=1
  ;;

  # 24 hour clock, no seconds
  ((#ibm)${~tspat}(<0-24>):(<0-59>)(*))
  hour=$match[2]
  minute=$match[3]
  time_found=1
  ;;
esac

(( hour == 24 )) && hour=0

if (( time_found )); then
  # time was found
  time_start=$mbegin[2]
  time_end=$mend[-2]
  # Remove the timespec because it may be in the middle of
  # the date (as in the output of "date".
  # There may be a time zone, too, which we don't yet handle.
  # (It's not in POSIX strptime() and libraries don't support it well.)
  # This attempts to remove some of the weirder forms.
  if [[ $line[$time_end+1,-1] = (#b)[[:space:]]#([A-Z][A-Z][A-Z]|[-+][0-9][0-9][0-9][0-9])([[:space:]]|(#e))* || \
        $line[$time_end+1,-1] = (#b)[[:space:]]#([A-Z][A-Z][A-Z](|[-+])<0-12>)([[:space:]]|(#e))*  || \
        $line[$time_end+1,-1] = (#b)[[:space:]]#([A-Z][A-Z][A-Z](|[-+])<0-12>[A-Z][A-Z][A-Z])([[:space:]]|(#e))* ]]; then
     (( time_end += ${mend[-1]} ))
     tz=$match[1]
  fi
  line=$line[1,time_start-1]$line[time_end+1,-1]
  (( debug )) && print "line after time: $line"
fi

if (( relative == 0 )); then
  # Date.
  case $line in
  # Look for YEAR[-/.]MONTH[-/.]DAY
  ((#bi)${~dspat}((19|20)[0-9][0-9])[-/](<1-12>)[-/](<1-31>)*)
  year=$match[2]
  month=$match[4]
  day=$match[5]
  date_start=$mbegin[2] date_end=$mend[5]
  date_found=1
  ;;

  # Same with month name
  ((#bi)${~dspat}((19|20)[0-9][0-9])[-/]${~monthpat}[-/](<1-31>)*)
  year=$match[2]
  mname=$match[4]
  day=$match[5]
  date_start=$mbegin[2] date_end=$mend[5]
  date_found=1
  ;;

  # Look for DAY[th/st/rd] MNAME[,] YEAR
  ((#bi)${~dspat}(<1-31>)(|th|st|rd)[[:space:]]##${~monthpat}(|,)[[:space:]]##((19|20)[0-9][0-9])*)
  day=$match[2]
  mname=$match[4]
  year=$match[6]
  date_start=$mbegin[2] date_end=$mend[6]
  date_found=1
  ;;

  # Look for MNAME DAY[th/st/rd][,] YEAR
  ((#bi)${~dspat}${~monthpat}[[:space:]]##(<1-31>)(|th|st|rd)(|,)[[:space:]]##((19|20)[0-9][0-9])*)
  mname=$match[2]
  day=$match[3]
  year=$match[6]
  date_start=$mbegin[2] date_end=$mend[6]
  date_found=1
  ;;

  # Look for DAY[th/st/rd] MNAME; assume current year
  ((#bi)${~dspat}(<1-31>)(|th|st|rd)[[:space:]]##${~monthpat}(|,)([[:space:]]##*|))
  day=$match[2]
  mname=$match[4]
  strftime -s year "%Y" $EPOCHSECONDS
  date_start=$mbegin[2] date_end=$mend[5]
  date_found=1
  ;;

  # Look for MNAME DAY[th/st/rd]; assume current year
  ((#bi)${~dspat}${~monthpat}[[:space:]]##(<1-31>)(|th|st|rd)(|,)([[:space:]]##*|))
  mname=$match[2]
  day=$match[3]
  strftime -s year "%Y" $EPOCHSECONDS
  date_start=$mbegin[2] date_end=$mend[5]
  date_found=1
  ;;

  # Now it gets a bit ambiguous.
  # Look for DAY[th/st/rd][/]MONTH[/ ,]YEAR
  ((#bi)${~dspat}(<1-31>)(|th|st|rd)/(<1-12>)((|,)[[:space:]]##|/)((19|20)[0-9][0-9])*)
  day=$match[2]
  month=$match[4]
  year=$match[7]
  date_start=$mbegin[2] date_end=$mend[7]
  date_found=1
  ;;

  # Look for MONTH[/]DAY[th/st/rd][/ ,]YEAR
  ((#bi)${~dspat}(<1-12>)/(<1-31>)(|th|st|rd)((|,)[[:space:]]##|/)((19|20)[0-9][0-9])*)
  month=$match[2]
  day=$match[3]
  year=$match[7]
  date_start=$mbegin[2] date_end=$mend[7]
  date_found=1
  ;;
  esac
fi

if (( date_found )); then
  # date found
  # see if there's a day at the start
  if [[ ${line[1,$date_start-1]} = (#bi)${~daypat} ]]; then
    date_start=$mbegin[1]
  fi
  line=${line[1,$date_start-1]}${line[$date_end+1,-1]}
  if (( time_found )); then
    # If we found a time, it must be associated with the date,
    # or we can't use it.  Since we removed the time from the
    # string to find the date, however, it's complicated to
    # know where both were found.  Reconstruct the date indices of
    # the original string.
    if (( time_start <= date_start )); then
      # Time came before start of date; add length in.
      (( date_start += time_end - time_start + 1 ))
    fi
    if (( time_start <= date_end )); then
      (( date_end += time_end - time_start + 1 ))
    fi

    if (( time_end + 1 < date_start )); then
      # If time wholly before date, OK if only separator characters
      # in between.  (This allows some illogical stuff with commas
      # but that's probably not important.)
      if [[ ${orig_line[time_end+1,date_start-1]} != ${~schars}# ]]; then
	# Clearly this can't work if anchor is set.  In principle,
	# we could match the date and ignore the time if it wasn't.
	# However, that seems dodgy.
	return 1
      else
	# Form massaged line by removing the entire date/time chunk.
	line="${orig_line[1,time_start-1]}${orig_line[date_end+1,-1]}"
      fi
    elif (( date_end + 1 < time_start )); then
      # If date wholly before time, OK if only time separator characters
      # in between.  This allows 2006/10/12:13:43 etc.
      if [[ ${orig_line[date_end+1,time_start-1]} != ${~tschars}# ]]; then
	# Here, we assume the time is associated with something later
	# in the line.  This is pretty much inevitable for the sort
	# of use we are expecting.  For example,
	#   2006/10/24  Meeting from early, may go on till 12:00.
	# or with some uses of the calendar system,
	#   2006/10/24 MR 1 Another pointless meeting WARN 01:00
	# The 01:00 says warn an hour before, not that the meeting starts
	# at 1 am.  About the only safe way round would be to force
	# a time to be present, but that's not how the traditional
	# calendar programme works.
	#
	# Hence we need to reconstruct.
	(( time_found = 0, hour = 0, minute = 0, second = 0 ))
	line="${orig_line[1,date_start-1]}${orig_line[date_end+1,-1]}"
      else
	# As above.
	line="${orig_line[1,date_start-1]}${orig_line[time_end+1,-1]}"
      fi
    fi
    if (( debug )); then
      print "Time string: $time_start,$time_end:" \
	"'$orig_line[time_start,time_end]'"
      print "Date string: $date_start,$date_end:" \
	"'$orig_line[date_start,date_end]'"
      print "Remaining line: '$line'"
    fi
  fi
fi

if (( relative )); then
  if [[ $line = (#bi)${~dspat}(<->)[[:blank:]]#(y|yr|year)${~repat} ]]; then
    (( reladd += ((365*4+1) * 24 * 60 * 60 * ${match[2]} + 1) / 4 ))
    line=${line[1,$mbegin[2]-1]}${line[$mend[4]+1,-1]}
    time_found=1
  fi
  if [[ $line = (#bi)${~dspat}(<->)[[:blank:]]#(mth|mon|mnth|month)${~repat} ]]; then
     (( reladd += 30 * 24 * 60 * 60 * ${match[2]} ))
     line=${line[1,$mbegin[2]-1]}${line[$mend[4]+1,-1]}
     time_found=1
  fi
  if [[ $line = (#bi)${~dspat}(<->)[[:blank:]]#(w|wk|week)${~repat} ]]; then
     (( reladd += 7 * 24 * 60 * 60 * ${match[2]} ))
     line=${line[1,$mbegin[2]-1]}${line[$mend[4]+1,-1]}
     time_found=1
  fi
  if [[ $line = (#bi)${~dspat}(<->)[[:blank:]]#(d|dy|day)${~repat} ]]; then
     (( reladd += 24 * 60 * 60 * ${match[2]} ))
     line=${line[1,$mbegin[2]-1]}${line[$mend[4]+1,-1]}
     time_found=1
  fi
  if [[ $line = (#bi)${~dspat}(<->)[[:blank:]]#(h|hr|hour)${~repat} ]]; then
     (( reladd += 60 * 60 * ${match[2]} ))
     line=${line[1,$mbegin[2]-1]}${line[$mend[4]+1,-1]}
     time_found=1
  fi
  if [[ $line = (#bi)${~dspat}(<->)[[:blank:]]#(min|minute)${~repat} ]]; then
     (( reladd += 60 * ${match[2]} ))
     line=${line[1,$mbegin[2]-1]}${line[$mend[4]+1,-1]}
     time_found=1
  fi
  if [[ $line = (#bi)${~dspat}(<->)[[:blank:]]#(s|sec|second)${~repat} ]]; then
     (( reladd += ${match[2]} ))
     line=${line[1,$mbegin[2]-1]}${line[$mend[4]+1,-1]}
     time_found=1
  fi
fi

if (( relative )); then
  # If no date was found, we're in trouble unless we found a time.
  if (( time_found )); then
    if (( anchor_end )); then
      # must be left with only separator characters
      if [[ $line != ${~schars}# ]]; then
	return 1
      fi
    fi
    (( REPLY = reladd + (hour * 60 + minute) * 60 + second ))
    [[ -n $setvar ]] && REPLY2=$line
    return 0
  fi
  return 1
elif (( ! date_found )); then
  return 1
fi

if (( anchor_end )); then
  # must be left with only separator characters
  if [[ $line != ${~schars}# ]]; then
    return 1
  fi
fi

local fmt nums
if [[ -n $mname ]]; then
  fmt="%Y %b %d %H %M %S"
  nums="$year $mname $day $hour $minute $second"
else
  fmt="%Y %m %d %H %M %S"
  nums="$year $month $day $hour $minute $second"
fi

strftime -s REPLY -r $fmt $nums

[[ -n $setvar ]] && REPLY2=$line

return 0