1
mirror of https://github.com/rclone/rclone synced 2024-11-21 22:50:16 +01:00
rclone/bin/make_manual.py
iotmaestro 4b5c10f72e
Add a new backend for uloz.to
Note that this temporarily skips uploads of files over 2.5 GB.

See https://github.com/rclone/rclone/pull/7552#issuecomment-1956316492
for details.
2024-03-26 09:46:47 +00:00

207 lines
5.4 KiB
Python
Executable File

#!/usr/bin/env python3
"""
Make single page versions of the documentation for release and
conversion into man pages etc.
"""
import os
import re
import time
from datetime import datetime
docpath = "docs/content"
outfile = "MANUAL.md"
# Order to add docs segments to make outfile
docs = [
"_index.md",
"install.md",
"docs.md",
"remote_setup.md",
"filtering.md",
"gui.md",
"rc.md",
"overview.md",
"flags.md",
"docker.md",
"bisync.md",
"release_signing.md",
# Keep these alphabetical by full name
"fichier.md",
"alias.md",
"s3.md",
"b2.md",
"box.md",
"cache.md",
"chunker.md",
"sharefile.md",
"crypt.md",
"compress.md",
"combine.md",
"dropbox.md",
"filefabric.md",
"ftp.md",
"googlecloudstorage.md",
"drive.md",
"googlephotos.md",
"hasher.md",
"hdfs.md",
"hidrive.md",
"http.md",
"imagekit.md",
"internetarchive.md",
"jottacloud.md",
"koofr.md",
"linkbox.md",
"mailru.md",
"mega.md",
"memory.md",
"netstorage.md",
"azureblob.md",
"azurefiles.md",
"onedrive.md",
"opendrive.md",
"oracleobjectstorage.md",
"qingstor.md",
"quatrix.md",
"sia.md",
"swift.md",
"pcloud.md",
"pikpak.md",
"premiumizeme.md",
"protondrive.md",
"putio.md",
"protondrive.md",
"seafile.md",
"sftp.md",
"smb.md",
"storj.md",
"sugarsync.md",
"tardigrade.md", # stub only to redirect to storj.md
"ulozto.md",
"uptobox.md",
"union.md",
"webdav.md",
"yandex.md",
"zoho.md",
"local.md",
"changelog.md",
"bugs.md",
"faq.md",
"licence.md",
"authors.md",
"contact.md",
]
# Order to put the commands in - any not on here will be in sorted order
commands_order = [
"rclone_config.md",
"rclone_copy.md",
"rclone_sync.md",
"rclone_move.md",
"rclone_delete.md",
"rclone_purge.md",
"rclone_mkdir.md",
"rclone_rmdir.md",
"rclone_check.md",
"rclone_ls.md",
"rclone_lsd.md",
"rclone_lsl.md",
"rclone_md5sum.md",
"rclone_sha1sum.md",
"rclone_size.md",
"rclone_version.md",
"rclone_cleanup.md",
"rclone_dedupe.md",
]
# Docs which aren't made into outfile
ignore_docs = [
"downloads.md",
"privacy.md",
"sponsor.md",
"amazonclouddrive.md",
]
def read_doc(doc):
"""Read file as a string"""
path = os.path.join(docpath, doc)
with open(path) as fd:
contents = fd.read()
parts = contents.split("---\n", 2)
if len(parts) != 3:
raise ValueError("Couldn't find --- markers: found %d parts" % len(parts))
contents = parts[2].strip()+"\n\n"
# Remove icons
contents = re.sub(r'<i class="fa.*?</i>\s*', "", contents)
# Interpret img shortcodes
# {{< img ... >}}
contents = re.sub(r'\{\{<\s*img\s+(.*?)>\}\}', r"<img \1>", contents)
# Make any img tags absolute
contents = re.sub(r'(<img.*?src=")/', r"\1https://rclone.org/", contents)
# Make [...](/links/) absolute
contents = re.sub(r'\]\((\/.*?\/(#.*)?)\)', r"](https://rclone.org\1)", contents)
# Add additional links on the front page
contents = re.sub(r'\{\{< rem MAINPAGELINK >\}\}', "- [Donate.](https://rclone.org/donate/)", contents)
# Interpret provider shortcode
# {{< provider name="Amazon S3" home="https://aws.amazon.com/s3/" config="/s3/" >}}
contents = re.sub(r'\{\{<\s*provider.*?name="(.*?)".*?>\}\}', r"- \1", contents)
# Remove remaining shortcodes
contents = re.sub(r'\{\{<.*?>\}\}', r"", contents)
contents = re.sub(r'\{\{%.*?%\}\}', r"", contents)
return contents
def check_docs(docpath):
"""Check all the docs are in docpath"""
files = set(f for f in os.listdir(docpath) if f.endswith(".md"))
files -= set(ignore_docs)
docs_set = set(docs)
if files == docs_set:
return
print("Files on disk but not in docs variable: %s" % ", ".join(files - docs_set))
print("Files in docs variable but not on disk: %s" % ", ".join(docs_set - files))
raise ValueError("Missing files")
def read_command(command):
doc = read_doc("commands/"+command)
doc = re.sub(r"### Options inherited from parent commands.*$", "", doc, 0, re.S)
doc = doc.strip()+"\n"
return doc
def read_commands(docpath):
"""Reads the commands an makes them into a single page"""
files = set(f for f in os.listdir(docpath + "/commands") if f.endswith(".md"))
docs = []
for command in commands_order:
docs.append(read_command(command))
files.remove(command)
for command in sorted(files):
if command != "rclone.md":
docs.append(read_command(command))
return "\n".join(docs)
def main():
check_docs(docpath)
command_docs = read_commands(docpath).replace("\\", "\\\\") # escape \ so we can use command_docs in re.sub
build_date = datetime.utcfromtimestamp(
int(os.environ.get('SOURCE_DATE_EPOCH', time.time())))
with open(outfile, "w") as out:
out.write("""\
%% rclone(1) User Manual
%% Nick Craig-Wood
%% %s
""" % build_date.strftime("%b %d, %Y"))
for doc in docs:
contents = read_doc(doc)
# Substitute the commands into doc.md
if doc == "docs.md":
contents = re.sub(r"The main rclone commands.*?for the full list.", command_docs, contents, 0, re.S)
out.write(contents)
print("Written '%s'" % outfile)
if __name__ == "__main__":
main()