scuffed-code/icu4c/source/tools/gentz/tzparse.pm

246 lines
10 KiB
Perl

######################################################################
# Copyright (C) 1999, International Business Machines
# Corporation and others. All Rights Reserved.
######################################################################
# See: ftp://elsie.nci.nih.gov/pub/tzdata<year>
# where <year> is "1999b" or a similar string.
######################################################################
# This package handles the parsing of time zone files.
# Author: Alan Liu
######################################################################
# Usage:
# Call ParseFile for each file to be imported. Then call Postprocess
# to remove unused rules and links.
package TZ;
use strict;
use Carp;
use vars qw(@ISA @EXPORT $VERSION $YEAR $STANDARD);
require 'dumpvar.pl';
@ISA = qw(Exporter);
@EXPORT = qw(ParseFile
Postprocess
);
$VERSION = '0.1';
$STANDARD = '-'; # Name of the Standard Time rule
######################################################################
# Param: File name
# Param: Ref to hash of zones
# Param: Ref to hash of rules
# Param: Current year
sub ParseFile {
my ($FILE, $ZONES, $RULES, $YEAR) = @_;
local(*FILE);
open(FILE,"<$FILE") or confess "Can't open $FILE: $!";
my $zone; # Current zone
my $badLineCount = 0;
while (<FILE>) {
s/\#.*//;
next if (!/\S/);
#|# Zone NAME GMTOFF RULES FORMAT [UNTIL]
#|Zone America/Montreal -4:54:16 - LMT 1884
#| -5:00 Mont E%sT
#|Zone America/Thunder_Bay -5:57:00 - LMT 1895
#| -5:00 Canada E%sT 1970
#| -5:00 Mont E%sT 1973
#| -5:00 - EST 1974
#| -5:00 Canada E%sT
my ($zoneGmtoff, $zoneRule, $zoneFormat, $zoneUntil);
if (/^zone/i) {
# Zone block start
if (/^zone\s+(\S+)\s+(\S+)\s+(\S+)\s+(\S+)\s+(\S+)/i
|| /^zone\s+(\S+)\s+(\S+)\s+(\S+)\s+(\S+)()/i) {
$zone = $1;
($zoneGmtoff, $zoneRule, $zoneFormat, $zoneUntil) =
($2, $3, $4, $5);
} else {
print STDERR "Can't parse in $FILE: $_";
++$badLineCount;
}
} elsif (/^\s/ && $zone) {
# Zone continuation
if (/^\s+(\S+)\s+(\S+)\s+(\S+)\s+(\S+)/
|| /^\s+(\S+)\s+(\S+)\s+(\S+)()/) {
($zoneGmtoff, $zoneRule, $zoneFormat, $zoneUntil) =
($1, $2, $3, $4);
} else {
print STDERR "Can't parse in $FILE: $_";
++$badLineCount;
}
} elsif (/^rule/i) {
# Here is where we parse a single line of the rule table.
# Our goal is to accept only rules applying to the current
# year. This is normally a matter of accepting rules
# that match the current year. However, in some cases this
# is more complicated. For example:
#|# Tonga
#|# Rule NAME FROM TO TYPE IN ON AT SAVE LETTER/S
#|Rule Tonga 1999 max - Oct Sat>=1 2:00s 1:00 S
#|Rule Tonga 2000 max - Apr Sun>=16 2:00s 0 -
# To handle this properly, we save every rule we encounter
# (thus overwriting older ones with newer ones, since rules
# are listed in order), and also use slot [2] to mark when
# we see a current year rule. When that happens, we stop
# saving rules. Thus we match the latest rule we see, or
# a matching rule if we find one. The format of slot [2]
# is just a 2 bit flag ([2]&1 means slot [0] matched,
# [2]&2 means slot [1] matched).
# Note that later, when the rules are post processed
# (see Postprocess), the slot [2] will be overwritten
# with the compressed rule string used to implement
# equality testing.
$zone = undef;
# Rule
#|# Rule NAME FROM TO TYPE IN ON AT SAVE LETTER/S
#|Rule US 1918 1919 - Mar lastSun 2:00 1:00 W # War
#|Rule US 1918 1919 - Oct lastSun 2:00 0 S
#|Rule US 1942 only - Feb 9 2:00 1:00 W # War
#|Rule US 1945 only - Sep 30 2:00 0 S
#|Rule US 1967 max - Oct lastSun 2:00 0 S
#|Rule US 1967 1973 - Apr lastSun 2:00 1:00 D
#|Rule US 1974 only - Jan 6 2:00 1:00 D
#|Rule US 1975 only - Feb 23 2:00 1:00 D
#|Rule US 1976 1986 - Apr lastSun 2:00 1:00 D
#|Rule US 1987 max - Apr Sun>=1 2:00 1:00 D
if (/^rule\s+(\S+)\s+(\S+)\s+(\S+)\s+(\S+)\s+
(\S+)\s+(\S+)\s+(\S+)\s+(\S+)\s+(\S+)/xi) {
my ($name, $from, $to, $type, $in, $on, $at, $save, $letter) =
($1, $2, $3, $4, $5, $6, $7, $8, $9);
my $i = $save ? 0:1;
if (!exists $RULES->{$name}) {
$RULES->{$name} = [];
}
my $ruleArray = $RULES->{$name};
# Check our bit mask to see if we've already matched
# a current rule. If so, do nothing. If not, then
# save this rule line as the best one so far.
if (@{$ruleArray} < 3 ||
!($ruleArray->[2] & $i)) {
my $h = $ruleArray->[$i];
$ruleArray->[$i]->{from} = $from;
$ruleArray->[$i]->{to} = $to;
$ruleArray->[$i]->{type} = $type;
$ruleArray->[$i]->{in} = $in;
$ruleArray->[$i]->{on} = $on;
$ruleArray->[$i]->{at} = $at;
$ruleArray->[$i]->{save} = $save;
$ruleArray->[$i]->{letter} = $letter;
# Does this rule match the current year? If so,
# set the bit mask so we don't overwrite this rule.
# This makes us ingore rules for subsequent years
# that are already listed in the database -- as long
# as we have an overriding rule for the current year.
if (($from == $YEAR && $to =~ /only/i) ||
($from <= $YEAR &&
(($to =~ /^\d/ && $YEAR <= $to) || $to =~ /max/i))) {
$ruleArray->[2] |= $i;
}
}
} else {
print STDERR "Can't parse in $FILE: $_";
++$badLineCount;
}
} elsif (/^link/i) {
#|# Old names, for S5 users
#|
#|# Link LINK-FROM LINK-TO
#|Link America/New_York EST5EDT
#|Link America/Chicago CST6CDT
#|Link America/Denver MST7MDT
#|Link America/Los_Angeles PST8PDT
#|Link America/Indianapolis EST
#|Link America/Phoenix MST
#|Link Pacific/Honolulu HST
if (/^link\s+(\S+)\s+(\S+)/i) {
# We currently only record a single link -- if there
# are more than one, we should modify this.
my ($from, $to) = ($1, $2);
$ZONES->{$from}->{link} = $to;
} else {
print STDERR "Can't parse in $FILE: $_";
++$badLineCount;
}
} else {
# Unexpected line
print STDERR "Ignoring in $FILE: $_";
++$badLineCount;
}
if ($zoneRule &&
($zoneUntil !~ /\S/ || ($zoneUntil =~ /^\d/ &&
$zoneUntil >= $YEAR))) {
$ZONES->{$zone}->{gmtoff} = $zoneGmtoff;
$ZONES->{$zone}->{rule} = $zoneRule;
$ZONES->{$zone}->{format} = $zoneFormat;
$ZONES->{$zone}->{until} = $zoneUntil;
}
}
close(FILE);
}
######################################################################
# Param: Ref to hash of zones
# Param: Ref to hash of rules
sub Postprocess {
my ($ZONES, $RULES) = @_;
my %ruleInUse;
# Eliminate zone links that have no corresponding zone
foreach (keys %$ZONES) {
if (exists $ZONES->{$_}->{link} && !exists $ZONES->{$_}->{rule}) {
if (0) {
print STDERR
"Deleting link from historical/nonexistent zone: ",
$_, " -> ", $ZONES->{$_}->{link}, "\n";
}
delete $ZONES->{$_};
}
}
# Check that each zone has a corresponding rule
foreach (sort keys %$ZONES) {
my $ruleName = $ZONES->{$_}->{rule};
next if ($ruleName eq $STANDARD);
if (exists $RULES->{$ruleName}) {
$ruleInUse{$ruleName} = 1;
} else {
# This means the zone is using the standard rule now
$ZONES->{$_}->{rule} = $STANDARD;
}
}
# Check that both parts are there for rules
# Check for unused rules
# Make coded string for comparisons
foreach (keys %$RULES) {
if (!exists $ruleInUse{$_}) {
if (0) {
print STDERR "Deleting historical/unused rule: $_\n";
}
delete $RULES->{$_};
} elsif (!$RULES->{$_}->[0] || !$RULES->{$_}->[1]) {
print STDERR "Rule doesn't have both parts: $_\n";
} else {
# Generate code
# This has all the data about a rule; it can be used
# to see if two rules behave identically
$RULES->{$_}->[2] =
lc($RULES->{$_}->[0]->{in} . "," .
$RULES->{$_}->[0]->{on} . "," .
$RULES->{$_}->[0]->{at} . "," .
$RULES->{$_}->[0]->{save} . ";" .
$RULES->{$_}->[1]->{in} . "," .
$RULES->{$_}->[1]->{on} . "," .
$RULES->{$_}->[1]->{at}); # [1]->{save} is always zero
}
}
}