mirror of
https://github.com/AsahiLinux/u-boot
synced 2024-11-17 02:08:38 +00:00
93e1459641
Signed-off-by: Wolfgang Denk <wd@denx.de> [trini: Drop changes for PEP 4 following python tools] Signed-off-by: Tom Rini <trini@ti.com>
132 lines
4.3 KiB
Python
Executable file
132 lines
4.3 KiB
Python
Executable file
#! /usr/bin/python
|
|
########################################################################
|
|
#
|
|
# reorder and reformat a file in columns
|
|
#
|
|
# this utility takes lines from its standard input and reproduces them,
|
|
# partially reordered and reformatted, on its standard output.
|
|
#
|
|
# It has the same effect as a 'sort | column -t', with the exception
|
|
# that empty lines, as well as lines which start with a '#' sign, are
|
|
# not affected, i.e. they keep their position and formatting, and act
|
|
# as separators, i.e. the parts before and after them are each sorted
|
|
# separately (but overall field widths are computed across the whole
|
|
# input).
|
|
#
|
|
# Options:
|
|
# -i:
|
|
# --ignore-case:
|
|
# Do not consider case when sorting.
|
|
# -d:
|
|
# --default:
|
|
# What to chage empty fields to.
|
|
# -s <N>:
|
|
# --split=<N>:
|
|
# Treat only the first N whitespace sequences as separators.
|
|
# line content after the Nth separator will count as only one
|
|
# field even if it contains whitespace.
|
|
# Example : '-s 2' causes input 'a b c d e' to be split into
|
|
# three fields, 'a', 'b', and 'c d e'.
|
|
#
|
|
# boards.cfg requires -ids 6.
|
|
#
|
|
########################################################################
|
|
|
|
import sys, getopt, locale
|
|
|
|
# ensure we sort using the C locale.
|
|
|
|
locale.setlocale(locale.LC_ALL, 'C')
|
|
|
|
# check options
|
|
|
|
maxsplit = 0
|
|
ignore_case = 0
|
|
default_field =''
|
|
|
|
try:
|
|
opts, args = getopt.getopt(sys.argv[1:], "id:s:",
|
|
["ignore-case","default","split="])
|
|
except getopt.GetoptError as err:
|
|
print str(err) # will print something like "option -a not recognized"
|
|
sys.exit(2)
|
|
|
|
for o, a in opts:
|
|
if o in ("-s", "--split"):
|
|
maxsplit = eval(a)
|
|
elif o in ("-i", "--ignore-case"):
|
|
ignore_case = 1
|
|
elif o in ("-d", "--default"):
|
|
default_field = a
|
|
else:
|
|
assert False, "unhandled option"
|
|
|
|
# collect all lines from standard input and, for the ones which must be
|
|
# reformatted and sorted, count their fields and compute each field's
|
|
# maximum size
|
|
|
|
input_lines = []
|
|
field_width = []
|
|
|
|
for line in sys.stdin:
|
|
# remove final end of line
|
|
input_line = line.strip('\n')
|
|
if (len(input_line)>0) and (input_line[0] != '#'):
|
|
# sortable line: split into fields
|
|
fields = input_line.split(None,maxsplit)
|
|
# if there are new fields, top up field_widths
|
|
for f in range(len(field_width), len(fields)):
|
|
field_width.append(0)
|
|
# compute the maximum witdh of each field
|
|
for f in range(len(fields)):
|
|
field_width[f] = max(field_width[f],len(fields[f]))
|
|
# collect the line for next stage
|
|
input_lines.append(input_line)
|
|
|
|
# run through collected input lines, collect the ones which must be
|
|
# reformatted and sorted, and whenever a non-reformattable, non-sortable
|
|
# line is met, sort the collected lines before it and append them to the
|
|
# output lines, then add the non-sortable line too.
|
|
|
|
output_lines = []
|
|
sortable_lines = []
|
|
for input_line in input_lines:
|
|
if (len(input_line)>0) and (input_line[0] != '#'):
|
|
# this line should be reformatted and sorted
|
|
input_fields = input_line.split(None,maxsplit)
|
|
output_fields = [];
|
|
# reformat each field to this field's column width
|
|
for f in range(len(input_fields)):
|
|
output_field = input_fields[f];
|
|
output_fields.append(output_field.ljust(field_width[f]))
|
|
# any missing field is set to default if it exists
|
|
if default_field != '':
|
|
for f in range(len(input_fields),len(field_width)):
|
|
output_fields.append(default_field.ljust(field_width[f]))
|
|
# join fields using two spaces, like column -t would
|
|
output_line = ' '.join(output_fields);
|
|
# collect line for later
|
|
sortable_lines.append(output_line)
|
|
else:
|
|
# this line is non-sortable
|
|
# sort collected sortable lines
|
|
if ignore_case!=0:
|
|
sortable_lines.sort(key=lambda x: str.lower(locale.strxfrm(x)))
|
|
else:
|
|
sortable_lines.sort(key=lambda x: locale.strxfrm(x))
|
|
# append sortable lines to the final output
|
|
output_lines.extend(sortable_lines)
|
|
sortable_lines = []
|
|
# append non-sortable line to the final output
|
|
output_lines.append(input_line)
|
|
# maybe we had sortable lines pending, so append them to the final output
|
|
if ignore_case!=0:
|
|
sortable_lines.sort(key=lambda x: str.lower(locale.strxfrm(x)))
|
|
else:
|
|
sortable_lines.sort(key=lambda x: locale.strxfrm(x))
|
|
output_lines.extend(sortable_lines)
|
|
|
|
# run through output lines and print them, except rightmost whitespace
|
|
|
|
for output_line in output_lines:
|
|
print output_line.rstrip()
|