view Documentations/setGPLHeader.py @ 483:90d1f793dcf2 FixLayout_Header_LogView

Development option only read bottledata: If the header divisor is set (in my case because of switching between development and main versions) the samples need to be read considering the bottle values, too. Divelogs written by a SW not supporting bottle pressure have the divisor set to 0 => no impact to main SW Bugfix header memory layout: The batterygaugeregisters were not used and will be reused for other information. To keep the OSTC3 format up and running the datafield is not filled with "0x00"
author ideenmodellierer
date Tue, 26 May 2020 21:12:02 +0200
parents 6237372f76a4
children
line wrap: on
line source

#! /usr/bin/env python3

from __future__ import print_function
import os, sys, glob,re, codecs

##############################################################################

OSTC4=os.getcwd()
if not os.path.exists( os.path.join(OSTC4, 'Discovery') ):
    raise 'Wrong start directory...'

##############################################################################

def walk(OSTC4):
    for dir in ['BootLoader', 'Common', 'Discovery', 'FontPack', 'Small_CPU']:
        # Make sure we have the top directory...
        if not os.path.exists( os.path.join(OSTC4, dir) ):
            raise 'Missing ' + dir

        # Then walk in all its existing source sub-directories...
        for sub in ['.', 'Inc', 'Src']:
            path = os.path.join(OSTC4, dir, sub)
            if not os.path.exists(path):
                continue
            print(path + ':')
            for file in sorted( glob.iglob( os.path.join(path, '*.[chs]') ) ):
                try:
                    work(file)
                except Exception as e:
                    raise RuntimeError('Cannot process ' + file + ':\n\t' + str(e))

##############################################################################

with open(os.path.join(OSTC4, 'Documentations', 'GPL_template.txt')) as f:
    template = f.read()

def work(file):

    # Unclear what is done if encoding is not yet UTF-8...
    with open(file, 'rt', encoding='cp1252') as f:
        lines = f.read().splitlines()

    # Set defaults
    kw = {}
    kw['file']   = file.replace(OSTC4+'/', '')
    kw['brief']  = ''
    kw['date']   = '2018'

    # Try to gather info from existing header, if any:
    for line in lines:
        # Skip files that already have a GNU license:
        if 'GNU General Public License' in line:
            print('(done)\t' + file)
            return

        get(line, 'brief', kw)
        get(line, 'date', kw)

    # Replace kw in header
    header = template
    for k,v in kw.items():
        header = header.replace('$'+k, v)

    # Overwrite resulting file, normalizing EOL
    with open(file, 'wt', encoding='utf-8') as f:
        for line in header.splitlines():
            f.write(line.rstrip() + '\n')
        for line in lines:
            f.write(line.rstrip() + '\n')
    print('+\t' + file)

##############################################################################

def get(line, word, kw):
    m = re.search('(@|\\\\)' + word + '\s*(.*)', line)
    if m:
        kw[word] = m.group(2)

##############################################################################

walk(OSTC4)