23 11 | 2016

Generate man pages for awscli

Written by Tanguy

Classified in : Homepage, Debian, Command line, To remember

No man pages, but almost

The AWS Command Line Interface, which is available in Debian, provides no man page. Instead, that tool has an integrated help system, which allows you to run commands such as aws rds help, that, for what I have seen, generates some reStructuredText, then converts it to a man page in troff format, then calls troff to convert it to text with basic formatting, and eventually passes it to a pager. Since this is close to what man does, the result looks like a degraded man page, with some features missing such as the adaptation to the terminal width.

Well, this is better than nothing, and better than what many under-documented tools can offer, but for several reasons, it still sucks: most importantly, it does not respect administrators' habits and it does not integrate with the system man database. You it does not allow you to use commands such as apropos, and you will get no man page name auto-completion from your shell since there is no man page.

Generate the man pages

Now, since the integrated help system does generate a man page internally, we can hack it to output it, and save it to a file:

Description: Enable a mode to generate troff man pages
 The awscli help system internally uses man pages, but only to convert
 them to text and show them with the pager. This patch enables a mode
 that prints the troff code so the user can save the man page.
 To use that mode, run the help commands with an environment variable
 OUTPUT set to 'troff', for instance:
     OUTPUT='troff' aws rds help
Forwarded: no
Author: Tanguy Ortolo <tanguy+debian@ortolo.eu>
Last-Update: 2016-11-22

Index: /usr/lib/python3/dist-packages/awscli/help.py
--- /usr/lib/python3/dist-packages/awscli/help.py       2016-11-21 12:14:22.236254730 +0100
+++ /usr/lib/python3/dist-packages/awscli/help.py       2016-11-21 12:14:22.236254730 +0100
@@ -49,6 +49,8 @@
     Return the appropriate HelpRenderer implementation for the
     current platform.
+    if 'OUTPUT' in os.environ and os.environ['OUTPUT'] == 'troff':
+        return TroffHelpRenderer()
     if platform.system() == 'Windows':
         return WindowsHelpRenderer()
@@ -97,6 +99,15 @@
         return contents

+class TroffHelpRenderer(object):
+    """
+    Render help content as troff code.
+    """
+    def render(self, contents):
+        sys.stdout.buffer.write(publish_string(contents, writer=manpage.Writer()))
 class PosixHelpRenderer(PagingHelpRenderer):
     Render help content on a Posix-like system.  This includes

This patch must be applied from the root directory with patch -p0, otherwise GNU patch will not accept to work on files with absolute names.

With that patch, you can run help commands with an environment variable OUTPUT='troff' to get the man page to use it as you like, for instance:

% OUTPUT='troff' aws rds help > aws_rds.1
% man -lt aws_rds.1 | lp

Generate all the man pages

Now that we are able to generate the man page of any aws command, all we need to generate all of them is a list of all the available commands. This is not that easy, because the commands are somehow derived from functions provided by a Python library named botocore, which are derived from a bunch of configuration files, and some of them are added, removed or renamed. Anyway, I have been able to write a Python script that does that, but it includes a static list of these modifications:

#! /usr/bin/python3

import subprocess
import awscli.clidriver

def write_manpage(command):
    manpage = open('%s.1' % '_'.join(command), 'w')
    process = subprocess.Popen(command,
            env={'OUTPUT': 'troff'},

driver = awscli.clidriver.CLIDriver()
command_table = driver._get_command_table()

renamed_commands = \
        'config': 'configservice',
        'codedeploy': 'deploy',
        's3': 's3api'
added_commands = \
        's3': ['cp', 'ls', 'mb', 'mv', 'presign', 'rb', 'rm', 'sync',
removed_subcommands = \
        'ses': ['delete-verified-email-address',
        'ec2': ['import-instance', 'import-volume'],
        'emr': ['run-job-flow', 'describe-job-flows',
                'add-job-flow-steps', 'terminate-job-flows',
                'list-bootstrap-actions', 'list-instance-groups',
        'rds': ['modify-option-group']
added_subcommands = \
        'rds': ['add-option-to-option-group',

# Build a dictionary of real commands, including renames, additions and
# removals.
real_commands = {}
for command in command_table:
    subcommands = []
    subcommand_table = command_table[command]._get_command_table()
    for subcommand in subcommand_table:
        # Skip removed subcommands
        if command in removed_subcommands \
                and subcommand in removed_subcommands[command]:
    # Add added subcommands
    if command in added_subcommands:
        for subcommand in added_subcommands[command]:
    # Directly add non-renamed commands
    if command not in renamed_commands:
        real_commands[command] = subcommands
    # Add renamed commands
        real_commands[renamed_commands[command]] = subcommands
# Add added commands
for command in added_commands:
    real_commands[command] = added_commands[command]

# For each real command and subcommand, generate a manpage
for command in real_commands:
    write_manpage(['aws', command])
    for subcommand in real_commands[command]:
        write_manpage(['aws', command, subcommand])
                         'sync', 'website']}

This script will generate more than 2,000 man page files in the current directory; you will then be able to move them to /usr/local/share/man/man1.

Since this is a lot of man pages, it may be appropriate to concatenate them by major command, for instance all the aws rds together…


wednesday 23 november 2016 à 20:40 Colin Watson said : #1

If you used '-' as the separator in output file names rather than '_' then it would work a bit better with man(1): e.g. 'man aws foo' would do something sensible.

wednesday 23 november 2016 à 21:08 Tanguy said : #2

@Colin Watson : Both work! man-db used to recognize only dashes, for git Git subcommand man pages, e.g. man git commit, but a while back I found it would have been better to recognize underscores as well, for live-build subcommands e.g. man lb bootstrap, and I submitted a patch that was accepted, see Debian bug #574641. Current versions of man-db will try with dashes and underscores to find the relevant man page when you indicate several words.

Write a comment

What is the fourth letter of the word ratzc? :