freebsd-skq/contrib/tzdata/zishrink.awk
2020-10-08 00:35:36 +00:00

319 lines
8.2 KiB
Awk

# Convert tzdata source into a smaller version of itself.
# Contributed by Paul Eggert. This file is in the public domain.
# This is not a general-purpose converter; it is designed for current tzdata.
# 'zic' should treat this script's output as if it were identical to
# this script's input.
# Record a hash N for the new name NAME, checking for collisions.
function record_hash(n, name)
{
if (used_hashes[n]) {
printf "# ! collision: %s %s\n", used_hashes[n], name
exit 1
}
used_hashes[n] = name
}
# Return a shortened rule name representing NAME,
# and record this relationship to the hash table.
function gen_rule_name(name, \
n)
{
# Use a simple memonic: the first two letters.
n = substr(name, 1, 2)
record_hash(n, name)
# printf "# %s = %s\n", n, name
return n
}
function prehash_rule_names( \
name)
{
# Rule names are not part of the tzdb API, so substitute shorter
# ones. Shortening them consistently from one release to the next
# simplifies comparison of the output. That being said, the
# 1-letter names below are not standardized in any way, and can
# change arbitrarily from one release to the next, as the main goal
# here is compression not comparison.
# Abbreviating these rules names to one letter saved the most space
# circa 2018e.
rule["Arg"] = "A"
rule["Brazil"] = "B"
rule["Canada"] = "C"
rule["Denmark"] = "D"
rule["EU"] = "E"
rule["France"] = "F"
rule["GB-Eire"] = "G"
rule["Halifax"] = "H"
rule["Italy"] = "I"
rule["Jordan"] = "J"
rule["Egypt"] = "K" # "Kemet" in ancient Egyptian
rule["Libya"] = "L"
rule["Morocco"] = "M"
rule["Neth"] = "N"
rule["Poland"] = "O" # arbitrary
rule["Palestine"] = "P"
rule["Cuba"] = "Q" # Its start sounds like "Q".
rule["Russia"] = "R"
rule["Syria"] = "S"
rule["Turkey"] = "T"
rule["Uruguay"] = "U"
rule["Vincennes"] = "V"
rule["Winn"] = "W"
rule["Mongol"] = "X" # arbitrary
rule["NT_YK"] = "Y"
rule["Zion"] = "Z"
rule["Austria"] = "a"
rule["Belgium"] = "b"
rule["C-Eur"] = "c"
rule["Algeria"] = "d" # country code DZ
rule["E-Eur"] = "e"
rule["Taiwan"] = "f" # Formosa
rule["Greece"] = "g"
rule["Hungary"] = "h"
rule["Iran"] = "i"
rule["StJohns"] = "j"
rule["Chatham"] = "k" # arbitrary
rule["Lebanon"] = "l"
rule["Mexico"] = "m"
rule["Tunisia"] = "n" # country code TN
rule["Moncton"] = "o" # arbitrary
rule["Port"] = "p"
rule["Albania"] = "q" # arbitrary
rule["Regina"] = "r"
rule["Spain"] = "s"
rule["Toronto"] = "t"
rule["US"] = "u"
rule["Louisville"] = "v" # ville
rule["Iceland"] = "w" # arbitrary
rule["Chile"] = "x" # arbitrary
rule["Para"] = "y" # country code PY
rule["Romania"] = "z" # arbitrary
rule["Macau"] = "_" # arbitrary
# Use ISO 3166 alpha-2 country codes for remaining names that are countries.
# This is more systematic, and avoids collisions (e.g., Malta and Moldova).
rule["Armenia"] = "AM"
rule["Aus"] = "AU"
rule["Azer"] = "AZ"
rule["Barb"] = "BB"
rule["Dhaka"] = "BD"
rule["Bulg"] = "BG"
rule["Bahamas"] = "BS"
rule["Belize"] = "BZ"
rule["Swiss"] = "CH"
rule["Cook"] = "CK"
rule["PRC"] = "CN"
rule["Cyprus"] = "CY"
rule["Czech"] = "CZ"
rule["Germany"] = "DE"
rule["DR"] = "DO"
rule["Ecuador"] = "EC"
rule["Finland"] = "FI"
rule["Fiji"] = "FJ"
rule["Falk"] = "FK"
rule["Ghana"] = "GH"
rule["Guat"] = "GT"
rule["Hond"] = "HN"
rule["Haiti"] = "HT"
rule["Eire"] = "IE"
rule["Iraq"] = "IQ"
rule["Japan"] = "JP"
rule["Kyrgyz"] = "KG"
rule["ROK"] = "KR"
rule["Latvia"] = "LV"
rule["Lux"] = "LX"
rule["Moldova"] = "MD"
rule["Malta"] = "MT"
rule["Mauritius"] = "MU"
rule["Namibia"] = "NA"
rule["Nic"] = "NI"
rule["Norway"] = "NO"
rule["Peru"] = "PE"
rule["Phil"] = "PH"
rule["Pakistan"] = "PK"
rule["Sudan"] = "SD"
rule["Salv"] = "SV"
rule["Tonga"] = "TO"
rule["Vanuatu"] = "VU"
# Avoid collisions.
rule["Detroit"] = "Dt" # De = Denver
for (name in rule) {
record_hash(rule[name], name)
}
}
# Process the input line LINE and save it for later output.
function process_input_line(line, \
field, end, i, n, startdef, \
linkline, ruleline, zoneline)
{
# Remove comments, normalize spaces, and append a space to each line.
sub(/#.*/, "", line)
line = line " "
gsub(/[\t ]+/, " ", line)
# Abbreviate keywords and determine line type.
linkline = sub(/^Link /, "L ", line)
ruleline = sub(/^Rule /, "R ", line)
zoneline = sub(/^Zone /, "Z ", line)
# Replace FooAsia rules with the same rules without "Asia", as they
# are duplicates.
if (match(line, /[^ ]Asia /)) {
if (ruleline) return
line = substr(line, 1, RSTART) substr(line, RSTART + 5)
}
# Abbreviate times.
while (match(line, /[: ]0+[0-9]/))
line = substr(line, 1, RSTART) substr(line, RSTART + RLENGTH - 1)
while (match(line, /:0[^:]/))
line = substr(line, 1, RSTART - 1) substr(line, RSTART + 2)
# Abbreviate weekday names.
while (match(line, / (last)?(Mon|Wed|Fri)[ <>]/)) {
end = RSTART + RLENGTH
line = substr(line, 1, end - 4) substr(line, end - 1)
}
while (match(line, / (last)?(Sun|Tue|Thu|Sat)[ <>]/)) {
end = RSTART + RLENGTH
line = substr(line, 1, end - 3) substr(line, end - 1)
}
# Abbreviate "max", "min", "only" and month names.
gsub(/ max /, " ma ", line)
gsub(/ min /, " mi ", line)
gsub(/ only /, " o ", line)
gsub(/ Jan /, " Ja ", line)
gsub(/ Feb /, " F ", line)
gsub(/ Apr /, " Ap ", line)
gsub(/ Aug /, " Au ", line)
gsub(/ Sep /, " S ", line)
gsub(/ Oct /, " O ", line)
gsub(/ Nov /, " N ", line)
gsub(/ Dec /, " D ", line)
# Strip leading and trailing space.
sub(/^ /, "", line)
sub(/ $/, "", line)
# Remove unnecessary trailing zero fields.
sub(/ 0+$/, "", line)
# Remove unnecessary trailing days-of-month "1".
if (match(line, /[A-Za-z] 1$/))
line = substr(line, 1, RSTART)
# Remove unnecessary trailing " Ja" (for January).
sub(/ Ja$/, "", line)
n = split(line, field)
# Abbreviate rule names.
i = zoneline ? 4 : linkline ? 0 : 2
if (i && field[i] ~ /^[^-+0-9]/) {
if (!rule[field[i]])
rule[field[i]] = gen_rule_name(field[i])
field[i] = rule[field[i]]
}
# If this zone supersedes an earlier one, delete the earlier one
# from the saved output lines.
startdef = ""
if (zoneline)
zonename = startdef = field[2]
else if (linkline)
zonename = startdef = field[3]
else if (ruleline)
zonename = ""
if (startdef) {
i = zonedef[startdef]
if (i) {
do
output_line[i - 1] = ""
while (output_line[i++] ~ /^[-+0-9]/);
}
}
zonedef[zonename] = nout + 1
# Save the line for later output.
line = field[1]
for (i = 2; i <= n; i++)
line = line " " field[i]
output_line[nout++] = line
}
function output_saved_lines( \
i)
{
for (i = 0; i < nout; i++)
if (output_line[i])
print output_line[i]
}
BEGIN {
# Files that the output normally depends on.
default_dep["africa"] = 1
default_dep["antarctica"] = 1
default_dep["asia"] = 1
default_dep["australasia"] = 1
default_dep["backward"] = 1
default_dep["etcetera"] = 1
default_dep["europe"] = 1
default_dep["factory"] = 1
default_dep["northamerica"] = 1
default_dep["southamerica"] = 1
default_dep["ziguard.awk"] = 1
default_dep["zishrink.awk"] = 1
# Output a version string from 'version' and related configuration variables
# supported by tzdb's Makefile. If you change the makefile or any other files
# that affect the output of this script, you should append '-SOMETHING'
# to the contents of 'version', where SOMETHING identifies what was changed.
ndeps = split(deps, dep)
ddeps = ""
for (i = 1; i <= ndeps; i++) {
if (default_dep[dep[i]]) {
default_dep[dep[i]]++
} else {
ddeps = ddeps " " dep[i]
}
}
for (d in default_dep) {
if (default_dep[d] == 1) {
ddeps = ddeps " !" d
}
}
print "# version", version
if (dataform != "main") {
print "# dataform", dataform
}
if (redo != "posix_right") {
print "# redo " redo
}
if (ddeps) {
print "# ddeps" ddeps
}
print "# This zic input file is in the public domain."
prehash_rule_names()
}
/^[\t ]*[^#\t ]/ {
process_input_line($0)
}
END {
output_saved_lines()
}