mirror of
https://gitlab.com/flectra-community/devops/odoo-2-flectra-converter.git
synced 2024-11-24 14:22:03 +00:00
455 lines
16 KiB
Python
455 lines
16 KiB
Python
# -*- coding: utf-8 -*-
|
|
# !/usr/bin/python
|
|
import ast
|
|
import logging
|
|
import os
|
|
import re
|
|
import shutil
|
|
from os.path import join as oj
|
|
from shutil import copytree
|
|
|
|
from jinja2 import FileSystemLoader, Environment
|
|
|
|
_logger = logging.getLogger(__name__)
|
|
|
|
|
|
class RepoMigrator(object):
|
|
|
|
def __init__(self, name, src_path, dst_path, company, contributor):
|
|
self._name = name
|
|
self.src_path = src_path
|
|
self.dst_path = dst_path
|
|
self._company = company
|
|
self._contributor = contributor
|
|
template_loader = FileSystemLoader(searchpath=oj(os.path.dirname(__file__), 'templates'))
|
|
self._template_env = Environment(loader=template_loader)
|
|
self._migrated_modules = False
|
|
|
|
def migrate(self):
|
|
if not os.path.isdir(self.src_path):
|
|
return
|
|
|
|
_logger.info('Starting migration of repository %s', self._name)
|
|
self._migrated_modules = self._migrate_modules()
|
|
bug_path = oj(self.dst_path, '.gitlab', 'issue_templates')
|
|
bug_file_path = oj(bug_path, 'Bug.md')
|
|
os.makedirs(bug_path, exist_ok=True)
|
|
shutil.copy(oj(os.path.dirname(__file__), 'templates', 'Bug.md'), bug_file_path)
|
|
|
|
# Generate README.md based on modules
|
|
template_file = 'README.md'
|
|
template = self._template_env.get_template(template_file)
|
|
name = self._name
|
|
description = ''
|
|
content = template.render(
|
|
project_title=name,
|
|
project_intro=description,
|
|
modules=self._migrated_modules
|
|
)
|
|
out = open(oj(self.dst_path, 'README.md'), 'w')
|
|
out.write(content)
|
|
out.close()
|
|
|
|
# copy requirements.txt if exists
|
|
self._has_requirements = False
|
|
if os.path.exists(oj(self.src_path, 'requirements.txt')):
|
|
shutil.copy(oj(self.src_path, 'requirements.txt'), oj(self.dst_path, 'requirements.txt'))
|
|
self._has_requirements = True
|
|
|
|
return self._migrated_modules
|
|
|
|
def generate_gitlab_ci(self):
|
|
template_file = 'gitlab-ci.yml'
|
|
template = self._template_env.get_template(template_file)
|
|
|
|
content = template.render(
|
|
modules=self._migrated_modules,
|
|
repository=self._name,
|
|
pip=self._has_requirements)
|
|
out = open(oj(self.dst_path, '.gitlab-ci.yml'), 'w')
|
|
out.write(content)
|
|
out.close()
|
|
|
|
def _migrate_modules(self):
|
|
migrated_modules = {}
|
|
for file_name in os.listdir(self.src_path):
|
|
if file_name[:1] == '.':
|
|
continue
|
|
src_path = os.path.join(os.path.abspath(self.src_path), file_name)
|
|
if os.path.isdir(src_path) and file_name not in ['setup', 'docs', 'doc', 'client_tools']:
|
|
dst_path = os.path.join(os.path.abspath(self.dst_path), file_name)
|
|
if os.path.exists(dst_path):
|
|
shutil.rmtree(dst_path)
|
|
module = ModuleMigrator(src_path, dst_path, '1.0', self._company, self._contributor).migrate()
|
|
migrated_modules = {**migrated_modules, **module}
|
|
return migrated_modules
|
|
|
|
def cleanup(self):
|
|
shutil.rmtree(self.src_path)
|
|
shutil.rmtree(self.dst_path)
|
|
|
|
|
|
class ModuleMigrator(object):
|
|
_readme_intro = u''
|
|
|
|
_replacements = {
|
|
'odoo': 'flectra',
|
|
'Odoo': 'Flectra',
|
|
'ODOO': 'FLECTRA',
|
|
'8069': '7073',
|
|
'Part of Flectra.': 'Part of Odoo, Flectra.',
|
|
'openerp': 'flectra',
|
|
'Openerp': 'Flectra',
|
|
'OpenERP': 'Flectra',
|
|
'OpenErp': 'Flectra',
|
|
'OPENERP': 'FLECTRA',
|
|
'Part of Flectra.': 'Part of Openerp, Flectra.',
|
|
'Part of Flectra.': 'Part of OpenERP, Flectra.',
|
|
}
|
|
|
|
_xml_replacements = {
|
|
'odoo': 'flectra',
|
|
'Odoo': 'Flectra',
|
|
'ODOO': 'FLECTRA',
|
|
'Part of Odoo.': 'Part of Odoo, Flectra.',
|
|
'openerp': 'flectra',
|
|
'Openerp': 'Flectra',
|
|
'OpenERP': 'Flectra',
|
|
'OpenErp': 'Flectra',
|
|
'OPENERP': 'FLECTRA',
|
|
'Part of Flectra.': 'Part of Openerp, Flectra.',
|
|
'Part of Flectra.': 'Part of OpenERP, Flectra.',
|
|
}
|
|
|
|
_init_replacements = {
|
|
'Odoo.': 'Odoo, Flectra.',
|
|
'odoo': 'flectra',
|
|
'openerp': 'flectra',
|
|
'Openerp': 'Flectra',
|
|
'OpenERP': 'Flectra',
|
|
'OpenErp': 'Flectra',
|
|
}
|
|
|
|
_manifest_replacements = {
|
|
'Odoo.': 'Odoo, Flectra.',
|
|
'odoo': 'flectra',
|
|
'openerp': 'flectra',
|
|
'Openerp': 'Flectra',
|
|
'OpenERP': 'Flectra',
|
|
'OpenErp': 'Flectra',
|
|
}
|
|
|
|
_readme_regex_patterns = [
|
|
'(.. image:: https://odoo-community.org.*?)^(Bug)',
|
|
'(Bug Tracker.*)^(Credits)',
|
|
'(Maintainer.*)',
|
|
'(.. \|badge.*\|)',
|
|
]
|
|
|
|
_obsolete_line_patterns = [
|
|
'-*- coding: utf-8 -*-',
|
|
]
|
|
_copyright_line_patterns = [
|
|
'copyright',
|
|
'Copyright',
|
|
'©',
|
|
]
|
|
|
|
_ingnore_dir = [
|
|
'cla',
|
|
'doc',
|
|
]
|
|
|
|
_delete_dir = [
|
|
'readme',
|
|
]
|
|
|
|
_ingnore_files = [
|
|
'LICENSE',
|
|
'COPYRIGHT',
|
|
'README.md',
|
|
'CONTRIBUTING.md',
|
|
'Makefile',
|
|
'MANIFEST.in'
|
|
]
|
|
|
|
_ingnore_py_words = [
|
|
'OpenERPSession'
|
|
]
|
|
|
|
_ignore_xml_words = [
|
|
'provider_openerp'
|
|
]
|
|
|
|
_website_replacements = {
|
|
'https://www.odoo.com': 'https://flectrahq.com',
|
|
'www.odoo.com': 'https://flectrahq.com',
|
|
'https://www.openerp.com': 'https://flectrahq.com',
|
|
'www.opernerp.com': 'https://flectrahq.com',
|
|
}
|
|
|
|
_replace_email = {
|
|
'info@odoo.com': 'info@flectrahq.com',
|
|
'info@openerp.com': 'info@flectrahq.com',
|
|
}
|
|
_copyright_list = []
|
|
|
|
def __init__(self, src, destination, flectra_release, company, contributor):
|
|
self._src = src
|
|
self._destination = destination
|
|
self._module_name = os.path.basename(destination)
|
|
self._repo_name = os.path.basename(os.path.dirname(self._destination))
|
|
self._release = flectra_release.split('.')[:2]
|
|
self._company = company
|
|
self._contributor = contributor
|
|
self._copyright_list = []
|
|
self._manifest = False
|
|
|
|
self._contributor_part = u'''Contributors
|
|
------------
|
|
|
|
* %s''' % self._contributor
|
|
|
|
def migrate(self):
|
|
_logger.info('Starting migration of module %s', self._module_name)
|
|
copytree(self._src, self._destination, symlinks=True, ignore_dangling_symlinks=True)
|
|
for base, odoo_dirs, odoo_files in os.walk(self._destination, topdown=True):
|
|
odoo_files = [f for f in odoo_files if not f[0] == '.']
|
|
odoo_dirs[:] = [d for d in odoo_dirs if not d[0] == '.']
|
|
self._rename_files(base, odoo_files)
|
|
self._rename_dir(base, odoo_dirs)
|
|
self._render_templates()
|
|
return {self._module_name: self._manifest}
|
|
|
|
def _render_templates(self):
|
|
# COPYRIGHT file
|
|
cr_file = oj(self._destination, 'COPYRIGHT')
|
|
if os.path.exists(cr_file):
|
|
os.remove(cr_file)
|
|
shutil.copy(oj(os.path.dirname(__file__), 'templates', 'COPYRIGHT'), self._destination)
|
|
for cr in self._copyright_list:
|
|
if not re.match('(copyright[ ]*[0-9-]+)', cr, re.IGNORECASE):
|
|
self._copyright_list.remove(cr)
|
|
copyright_list = list(set(self._copyright_list))
|
|
copyright_list = sorted(copyright_list)
|
|
existing_copyright = '\n '.join(copyright_list)
|
|
infile = self._open_read(self._destination, 'COPYRIGHT')
|
|
out = self._open_write(self._destination, 'COPYRIGHT')
|
|
infile = infile.replace('{% existing %}', existing_copyright)
|
|
out.write(infile)
|
|
out.close()
|
|
|
|
# LICENSE file
|
|
license_exists = False
|
|
if os.path.exists(oj(self._destination, 'LICENSE')):
|
|
if os.path.isdir(oj(self._destination, 'LICENSE')):
|
|
shutil.rmtree(oj(self._destination, 'LICENSE'))
|
|
else:
|
|
os.remove(oj(self._destination, 'LICENSE'))
|
|
|
|
shutil.copy(oj(os.path.dirname(__file__), 'templates', 'LICENSE'), self._destination)
|
|
|
|
# Module icon
|
|
desc_path = oj(self._destination, 'static', 'description')
|
|
os.makedirs(desc_path, exist_ok=True)
|
|
if not os.path.exists(oj(desc_path, 'icon.png')) and \
|
|
not os.path.exists(oj(desc_path, 'icon.jpg')):
|
|
shutil.copy(oj(os.path.dirname(__file__), 'templates', 'icon.jpg'), desc_path)
|
|
|
|
def _rename_files(self, base, items):
|
|
for name in items:
|
|
if name in self._ingnore_files:
|
|
continue
|
|
|
|
if name == '__openerp__.py':
|
|
os.rename(oj(base, name), oj(base, '__manifest__.py'))
|
|
name = '__manifest__.py'
|
|
if name == '__init__.py':
|
|
self._init_files(base)
|
|
self._remove_obsolete_lines(base, name)
|
|
elif name == '__manifest__.py':
|
|
self._manifest_files(base)
|
|
self._remove_obsolete_lines(base, name)
|
|
elif name.lower() == 'readme.rst':
|
|
self._readme_files(base, name)
|
|
else:
|
|
sp_name = name.split('.')
|
|
if len(sp_name) >= 2:
|
|
if sp_name[-1] in ['xml', 'csv', 'json', 'html']:
|
|
self._remove_obsolete_lines(base, name)
|
|
self._xml_csv_json_files(base, name)
|
|
elif sp_name[-1] in ['py', 'css', 'less', 'js', 'yml']:
|
|
self._python_files(base, name)
|
|
self._remove_obsolete_lines(base, name)
|
|
try:
|
|
for old, new in self._replacements.items():
|
|
if name != (name.replace(old, new)):
|
|
os.rename(oj(base, name), oj(base, name.replace(old, new)))
|
|
except OSError:
|
|
pass
|
|
|
|
def _rename_dir(self, base, items):
|
|
for folder in items:
|
|
if folder in self._delete_dir:
|
|
shutil.rmtree(oj(base, folder))
|
|
continue
|
|
if folder in self._ingnore_dir:
|
|
continue
|
|
for new_base, odoo_dirs, odoo_files in os.walk(oj(base, folder), topdown=True):
|
|
if odoo_files:
|
|
self._rename_files(new_base, odoo_files)
|
|
if odoo_dirs:
|
|
self._rename_dir(new_base, odoo_dirs)
|
|
if 'odoo' in folder:
|
|
os.rename(oj(base, folder), oj(base, folder.replace('odoo', 'flectra')))
|
|
|
|
def _init_files(self, base):
|
|
infile = self._open_read(base, '__init__.py')
|
|
out = self._open_write(base, '__init__.py')
|
|
for old, new in self._init_replacements.items():
|
|
infile = infile.replace(old, new)
|
|
out.write(infile)
|
|
out.close()
|
|
|
|
def _manifest_files(self, base):
|
|
temp = {**self._replace_email, **self._website_replacements}
|
|
infile = self._open_read(base, '__manifest__.py')
|
|
out = self._open_write(base, '__manifest__.py')
|
|
for old, new in temp.items():
|
|
infile = infile.replace(old, new)
|
|
out.write(infile)
|
|
out.close()
|
|
self._content_replacements(base, '__manifest__.py', self._manifest_replacements)
|
|
infile = self._open_read(base, '__manifest__.py')
|
|
out = self._open_write(base, '__manifest__.py')
|
|
manifest = ast.literal_eval(infile)
|
|
manifest['author'] += ', %s' % self._company
|
|
version_parts = manifest['version'].split('.')
|
|
if len(version_parts) <= 3:
|
|
version_parts = ['0', '0'] + version_parts
|
|
if len(version_parts) < 5:
|
|
version_parts += ['0'] * (5 - len(version_parts))
|
|
version_parts[0] = self._release[0]
|
|
version_parts[1] = self._release[1]
|
|
|
|
manifest['version'] = '.'.join(version_parts)
|
|
manifest['website'] = 'https://gitlab.com/flectra-community/%s' % self._repo_name
|
|
self._manifest = manifest
|
|
m = re.search('''["']version["']:[ '"]*([0-9a-zA-Z.]+)["']''', infile)
|
|
if m and m.group(1):
|
|
infile = infile.replace(m.group(1), manifest['version'])
|
|
m = re.search('''["']website["']:[ '"]*(.*?)["']''', infile)
|
|
if m and m.group(1):
|
|
infile = infile.replace(m.group(1), manifest['website'])
|
|
m = re.search('''["']author["']:[\n '"]*(.*?)["']''', infile, re.MULTILINE | re.DOTALL)
|
|
if m and m.group(1):
|
|
infile = infile.replace(m.group(1), m.group(1) + ', %s' % self._company)
|
|
out.write(infile)
|
|
out.close()
|
|
|
|
def _xml_csv_json_files(self, base, name):
|
|
infile = self._open_read(base, name)
|
|
out = self._open_write(base, name)
|
|
for old, new in self._replace_email.items():
|
|
infile = infile.replace(old, new)
|
|
for old, new in self._xml_replacements.items():
|
|
must_replace = True
|
|
for ignore_word in self._ignore_xml_words:
|
|
if ignore_word in old:
|
|
must_replace = False
|
|
break
|
|
if must_replace:
|
|
infile = infile.replace(old, new)
|
|
out.write(infile)
|
|
out.close()
|
|
|
|
def _python_files(self, base, name):
|
|
infile = self._open_read(base, name)
|
|
out = self._open_write(base, name)
|
|
for old, new in self._replace_email.items():
|
|
infile = infile.replace(old, new)
|
|
out.write(infile)
|
|
out.close()
|
|
self._content_replacements(base, name, self._replacements)
|
|
|
|
def _content_replacements(self, base, name, replace_dict):
|
|
infile = self._open_read_lines(base, name)
|
|
lines = []
|
|
if infile:
|
|
for line in infile:
|
|
words = line.split(' ')
|
|
line_parts = []
|
|
for word in words:
|
|
if word.startswith('info@') or word.startswith("'info@") or word.startswith('"info@'):
|
|
line_parts.append(word)
|
|
continue
|
|
for old, new in replace_dict.items():
|
|
must_replace = True
|
|
for ignore_word in self._ingnore_py_words:
|
|
if ignore_word in word:
|
|
must_replace = False
|
|
if must_replace:
|
|
word = word.replace(old, new)
|
|
line_parts.append(word)
|
|
lines.append(line_parts)
|
|
|
|
with open(oj(self._destination, 'temp'), 'a') as temp_file:
|
|
for line in lines:
|
|
for word in line:
|
|
word = word if word.endswith('\n') else word + ' ' if word else ' '
|
|
temp_file.write(word)
|
|
os.rename(oj(self._destination, 'temp'), oj(base, name))
|
|
|
|
def _open_read_lines(self, base, name):
|
|
return open(oj(base, name), 'r').readlines()
|
|
|
|
def _open_write(self, base, filename):
|
|
return open(oj(base, filename), 'w')
|
|
|
|
def _open_read(self, base, filename):
|
|
return open(oj(base, filename), 'r').read()
|
|
|
|
def _remove_obsolete_lines(self, base, name):
|
|
infile = self._open_read_lines(base, name)
|
|
lines = []
|
|
if not infile:
|
|
return
|
|
|
|
temp = self._obsolete_line_patterns + self._copyright_line_patterns
|
|
for line in infile:
|
|
remove_line = False
|
|
for pattern in temp:
|
|
if pattern in line:
|
|
if pattern in self._copyright_line_patterns:
|
|
cr_line = line.replace('#', '').strip()
|
|
self._copyright_list.append(cr_line)
|
|
else:
|
|
remove_line = True
|
|
if not remove_line:
|
|
lines.append(line)
|
|
|
|
with open(oj(self._destination, 'temp'), 'a') as temp_file:
|
|
for line in lines:
|
|
temp_file.write(line)
|
|
os.rename(oj(self._destination, 'temp'), oj(base, name))
|
|
|
|
def _readme_files(self, base, name):
|
|
infile = self._open_read(base, name)
|
|
out = self._open_write(base, name)
|
|
infile = self._readme_intro + infile
|
|
|
|
for pattern in self._readme_regex_patterns:
|
|
infile = self._regex_replace(infile, pattern)
|
|
infile = self._regex_replace(infile,
|
|
'(Contributors\n[-|~]*[\n|\n\n])',
|
|
self._contributor_part)
|
|
infile = infile.rstrip('\n')
|
|
out.write(infile)
|
|
out.close()
|
|
|
|
def _regex_replace(self, source, pattern, replace=''):
|
|
m = re.search(pattern, source, re.MULTILINE | re.DOTALL)
|
|
if not m or not m.group(1):
|
|
return source
|
|
return source.replace(m.group(1), replace)
|