Mock CLI Framework
Description
This is a framework that can be called from a short Python console script to simulate the responses of an actual command-line tool.
mock-cli-framework
provides API for two purposes:
- Playing back canned responses based on a provided list of command-line arguments in order to simulate an actual command-line tool's behavior
- Generating a directory of canned responses from invocations of the actual tool
A command "invocation" is a unique list of arguments to that command. A "response" is the output the command wrote to stdout
, stderr
, as well as the numerical exit status of the command.
Why?
This is useful in cases where the real command-line tool can't be used. For example you might have a script or other program that shells out to a command. You want to verify that your code processes that command's output and handles its errors properly, such as in automated tests. But the tests can't provide things the real command requires like human interaction or online account access that would be needed by the real tool. You can use mock-cli-work
to "record" the responses of the real command-line tool as well as use mock-cli-framework
to create a fake version of the command. You can then have your code shell out to the "fake" command (using symlinks, PATH variable, etc.) to play back those reponses in your testing environment.
Usage
Response Directory
To use mock-cli-framework
to simulate a real command, it requires a few things:
- A JSON-encoded "response directory" listing all the command-line argument variations and a response dictionary for each
- An on-disk repository of "responses" that can be read and then written to
stdout
& stderr
The response directory is designed to be easily hand generated (or easily scripted), but there is also API to automate creation of it.
Before discussing the Response Generator API, let's look at the response directory's anatomy.
As an example, the JSON response directory for the md5sum
command might look like:
{
"meta": {
"response_dir": "./responses"
},
"commands": {
"--binary|big-file.bin": {
"exit_status": 0,
"stdout": "output",
"stderr": "error_output",
"name": "binary-big-file-bin"
},
"--text|missing-file.txt": {
"exit_status": 1,
"stdout": "output",
"stderr": "error_output",
"name": "text-missing-file-txt"
},
"--check|file-list.txt": {
"exit_status": 0,
"stdout": "output",
"stderr": "error_output",
"name": "check-file-list-txt"
}
}
}
In the example above, each "command" dictionary is keyed by an encoded argument string, and contains:
- Numerical exit status
- Filename of
stdout
written to disk - Filename of
stderr
written to disk - An arbitary, unique name for this invocation which serves as the parent directory for the output files.
Then the "responses" directory looks like:
responses
├── binary-big-file-bin
│ ├── error_output
│ └── output
├── check-file-list-txt
│ ├── error_output
│ └── output
└── text-missing-file-txt
├── error_output
└── output
The names of the response output files and their containing directories don't especially matter, but it is recommended to use names that relate to what the command is intended to simulate.
Command Mock-up
Given the above response directory, you can programmatically use mock_cli
API to make a short script that pretends to be the real command:
from mock_cli import MockCommand
def build_arg_parser():
parser = argparse.ArgumentParser()
return parser
def main():
parser = build_arg_parser()
parser.parse_args()
responsedir_json_file = "./response-directory.json"
cmd = MockCommand(responsedir_json_file)
args = sys.argv[1:]
exit_status = cmd.respond(args)
return exit_status
if __name__ == "__main__":
exit(main())
Now you can run your mock-md5sum.py
, and it should play back responses from disk without having to actually md5 anything, or requiring the original input files:
$ mock-md5sum.py --binary big-file.bin
f4c014ae60f420d90c2b52f4969f8d99 *big-file.bin
If your code shells out to md5sum
to hash really large files that you don't want to have in your testing harness, you can substitute mock-md5sum.py
which will behave the same way as the real thing (or similarly enough). Presumably you can trust that md5sum
hashes big-file.bin
properly, so there's no need to replicate that part.
Response Generation API
To generate responses, there are two classes to know about;
CommandInvocation
ResponseDirectory
The ResponseDirectory
class takes one required arguments and two optional ones:
responsedir_json_file
is the path and filename of the response JSON dictionary that either exists or is to be created.create
Is an optional boolean flag to create the JSON dictionary if it doesn't already exist, defaulting to False
response_dir
Is an optional path to a directory on disk that will contain recorded response output files.
- It is required if
create
is true, so that it can be stored in the response dictionary for later use
ResponseDirectory
provides a method to record a command invocation: add_command_invocation()
. It takes two arguments:
cmd
is a CommandInvocation
object (discussed next)save
is an optional boolean flag to write the response dictionary to disk for later use, defaulting to False.
- If the response directory is not saved to disk it will be held in memory and discarded when not in use.
- If the response directory is to be used later or by a different program, pass
save=True
The CommandInvocation
class serves to bundle up a set of command-line arguments as well as command response context, including normal and error output and exit status. It takes several required arguments:
cmd_args
is a list of strings represengting command-line argumentsoutput
is a bytes
object read from the command's standard output.error_output
is a bytes
object read from the command's standard error.returncode
is the command's numerical exit status when executed with the provided command-line argumentsinvocation_name
is a unique, arbitrary name given to this particular invocation
- It is recommended that the name be related to the command's arguments and intended action
- The name should be filesystem-safe as it will be used as the directory name on disk to hold the output files
changes_state
a boolean flag indicating if this command should trigger a state iterationinput
is an optional bytes-like object that will be hashed if provided.
- The command invocation will be added under "commands_with_input" using its hash as a key
- Since more than one command may work with the same input, the invocations will be further keyed by their command-line arguments
Here's an example:
import subprocess
from mock_cli import ResponseDirectory, CommandInvocation
def main():
directory = ResponseDirectory("./resopnse-directory.json", create=True, response_dir="./responses")
argv = ["md5sum", "--binary", "big-file.bin"]
p = subprocess.run(argv, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
stdout = p.stdout
stderr = p.stderr
returncode = p.returncode
invocation_name = "md5sum-[binary]-[big-file-bin]"
cmd_args = argv[1:]
invocation = CommandInvocation(cmd_args, stdout, stderr, returncode, invocation_name)
directory.add_command_invocation(invocation, save=True)
)
Limitations
There are a number of limitations to be aware of that prevent mock-cli-framework
from fully simulating some commands:
- Environment variables aren't processed, so behavior that is affected by them isn't simulated
- Normal and error output can't be interleaved on the console
- Standard output, if any, is written first
- Standard error, if any, is written next
- Timing/performance can't be simulated, and will usually be virtually instaneous
- A command's typical run-time for a given input/workload can't be simulated
- No assumptions can be made about one workload being faster/slower than another
- Commands that run interactively can't be simulated
- No side effects such as data written to disk can be simulated
- If the output contains any time-sensitive details such as time-stamped logs, those details will reflect whatever was recorded and may not be current