mirror of
https://github.com/LCPQ/quantum_package
synced 2024-11-04 05:03:54 +01:00
214 lines
5.6 KiB
Python
Executable File
214 lines
5.6 KiB
Python
Executable File
#!/usr/bin/env python
|
|
"""Updates the README.rst file as the include directive is disabled on GitHub."""
|
|
__date__ = "Thu Apr 3 23:06:18 CEST 2014"
|
|
__author__ = "Anthony Scemama<scemama@irsamc.ups-tlse.fr> & TApplencourt "
|
|
|
|
README = "README.rst"
|
|
Assum_key = "Assumptions\n===========\n"
|
|
Needed_key = "Needed Modules\n==============\n"
|
|
Doc_key = "Documentation\n=============\n"
|
|
Sentinel = "@@$%&@@"
|
|
URL = "http://github.com/LCPQ/quantum_package/tree/master/src/"
|
|
|
|
import os
|
|
import subprocess
|
|
|
|
header = """
|
|
.. Do not edit this section. It was auto-generated from the
|
|
.. NEEDED_MODULES_CHILDREN file by the `update_README.py` script.
|
|
|
|
"""
|
|
|
|
|
|
def fetch_splitted_data():
|
|
"""Read the README.rst file and split it in strings:
|
|
* The description
|
|
* The assumptions
|
|
* The documentation
|
|
* The needed modules
|
|
The result is given as a list of strings
|
|
"""
|
|
|
|
file = open(README, 'r')
|
|
data = file.read()
|
|
file.close()
|
|
|
|
# Place sentinels
|
|
data = data.replace(Assum_key, Sentinel + Assum_key)
|
|
data = data.replace(Doc_key, Sentinel + Doc_key)
|
|
data = data.replace(Needed_key, Sentinel + Needed_key)
|
|
|
|
# Now Split data using the sentinels
|
|
result = data.split(Sentinel)
|
|
|
|
return result
|
|
|
|
|
|
def update_assumptions(data):
|
|
"""Read the ASSUMPTIONS.rst file, and replace the data with it."""
|
|
|
|
try:
|
|
file = open('ASSUMPTIONS.rst', 'r')
|
|
except IOError:
|
|
file = open('ASSUMPTIONS.rst', 'w')
|
|
assumptions = ""
|
|
else:
|
|
assumptions = file.read()
|
|
file.close()
|
|
|
|
if assumptions.strip() != "":
|
|
assumptions = Assum_key + header + assumptions + '\n\n'
|
|
|
|
has_assumptions = False
|
|
for i in range(len(data)):
|
|
if data[i].startswith(Assum_key):
|
|
has_assumptions = True
|
|
data[i] = assumptions
|
|
|
|
if not has_assumptions:
|
|
data.insert(1, assumptions)
|
|
|
|
return data
|
|
|
|
|
|
def update_needed(data):
|
|
"""Read the NEEDED_CHILDREN_MODULES file, and replace the data with it.
|
|
Create the links to the GitHub pages."""
|
|
|
|
file = open('NEEDED_CHILDREN_MODULES', 'r')
|
|
modules = file.read()
|
|
file.close()
|
|
|
|
header_image = ".. image:: tree_dependency.png\n\n"
|
|
|
|
if modules.strip() != "":
|
|
modules = ['* `%s <%s%s>`_' % (x, URL, x) for x in modules.split()]
|
|
modules = "\n".join(modules)
|
|
modules = Needed_key + header + header_image + modules + '\n\n'
|
|
|
|
has_modules = False
|
|
for i in range(len(data)):
|
|
if data[i].startswith(Needed_key):
|
|
has_modules = True
|
|
data[i] = modules
|
|
|
|
if not has_modules:
|
|
data.append(modules)
|
|
|
|
return data
|
|
|
|
|
|
def update_documentation(data):
|
|
"""Reads the BEGIN_DOC ... END_DOC blocks and builds the documentation"""
|
|
|
|
# If the file does not exist, don't do anything
|
|
try:
|
|
file = open('tags', 'r')
|
|
except:
|
|
return
|
|
tags = file.readlines()
|
|
file.close()
|
|
|
|
def extract_doc(item):
|
|
"""Extracts the documentation contained in IRPF90_man file"""
|
|
file = open("IRPF90_man/%s.l" % (item), 'r')
|
|
lines = file.readlines()
|
|
file.close()
|
|
result = []
|
|
inside = False
|
|
for line in lines:
|
|
if not inside:
|
|
inside = line.startswith(".SH Description")
|
|
else:
|
|
if line.startswith(".SH"):
|
|
break
|
|
result.append(" " + line.strip())
|
|
|
|
if result == []:
|
|
result = [" Undocumented"]
|
|
return "\n".join(result) + '\n'
|
|
|
|
items = []
|
|
dirname = os.path.basename(os.getcwd())
|
|
command = "git ls-tree --full-tree --name-only HEAD:src/%s"
|
|
command = command % (dirname)
|
|
try:
|
|
if dirname != 'src':
|
|
p = subprocess.Popen(command.split(), stdout=subprocess.PIPE)
|
|
tracked_files = p.stdout.read()
|
|
else:
|
|
tracked_files = ""
|
|
tracked_files = tracked_files.splitlines()
|
|
except:
|
|
tracked_files = []
|
|
for filename in tracked_files:
|
|
if filename.endswith('.irp.f'):
|
|
# Search for providers, subroutines and functions in each file using
|
|
# the tags file
|
|
search = "\t" + filename + "\t"
|
|
tmp = filter(lambda line: search in line, tags)
|
|
|
|
# Search for the documentation in the IRPF90_man directory
|
|
for item in tmp:
|
|
item, _, line = item.strip().split('\t')
|
|
doc = extract_doc(item)
|
|
items.append((item, filename, doc, line))
|
|
|
|
dirname = os.path.basename(os.getcwd())
|
|
# Write the documentation in the README
|
|
template = "`%(item)s <%(url)s%(dirname)s/%(filename)s#L%(line)s>`_\n%(doc)s\n"
|
|
|
|
documentation = Doc_key + header
|
|
url = URL
|
|
for item, filename, doc, line in items:
|
|
documentation += template % locals()
|
|
documentation += '\n\n'
|
|
|
|
has_doc = False
|
|
for i in range(len(data)):
|
|
if data[i].startswith(Doc_key):
|
|
has_doc = True
|
|
data[i] = documentation
|
|
|
|
if not has_doc:
|
|
data.append(documentation)
|
|
|
|
return data
|
|
|
|
|
|
def git_add():
|
|
"""Executes:
|
|
git add README.rst
|
|
throw an error if git is not precent"""
|
|
|
|
import subprocess
|
|
|
|
try:
|
|
# pipe output to /dev/null for silence
|
|
null = open("/dev/null", "w")
|
|
subprocess.Popen("git add README.rst", stdout=null, stderr=null)
|
|
null.close()
|
|
|
|
except OSError:
|
|
raise
|
|
|
|
|
|
def main():
|
|
data = fetch_splitted_data()
|
|
data = update_assumptions(data)
|
|
data = update_documentation(data)
|
|
data = update_needed(data)
|
|
output = ''.join(data)
|
|
|
|
with open(README, 'w') as f:
|
|
f.write(output)
|
|
|
|
try:
|
|
git_add()
|
|
except OSError:
|
|
pass
|
|
|
|
|
|
if __name__ == '__main__':
|
|
main()
|