Enable `streamlit run` to serve a list of apps.

Problem

You don’t just have one app file. But many. Right now streamlit will only serve one app.

Solution

Enable streamlit run without a file name to serve a folder of apps. Just like Voila does

Additional context

Consider whether this requires some configuration options in the configuration file

• Serve subfolders as well?
• Serve all python files or just some satisfying a filter like *app.py?

[isConic]

Questions:

• Is this a feature for developers to navigate their projects?
• Or is it for navigation of a set of tools once they’ve been written?

For users and tools: If I recall correctly, around 2016 airbnb open-sourced something called knowledge repos for organizing jupyter notebooks. My understanding is that it never took off, however it had some really nice UI features for navigating projects.

Here is an example of what the nb navigation gui looked like.

I’m throwing this sort of thinking in as an idea. I am bad at follow-through since I’m not well versed in web.

Heres a link to the knowledge repo: https://github.com/airbnb/knowledge-repo

[SimonBiggs]

My thought with this would be to have an index.py app which runs at the root path, and then every other example.py test.py etc would run at http://hostname.com/example and http://hostname.com/test.

The idea is that the CLI would point to a directory, if that directory does not contain an index.py a template one would be created, and it would do a little bit of a sanity check to make sure that the built in names are not being overwritten (healthz, stream, etc):

https://github.com/streamlit/streamlit/blob/a20aba7fbc0380b8ed05664f33275d81435a2df7/lib/streamlit/server/Server.py#L296-L330

So CLI usage would be:

streamlit run path/to/a/directory

and the directory tree could look something like:

path/to/a/directory
 | index.py
 | example.py
 | boo.py
 | foo.py
 | etc.py

Importantly, if any of the scripts imports a shared function that is cached, this cache should be shared amongst them so that the server scales appropriately (without doubling up on caching).

An example index.py that could be created and placed within that directory if an index.py doesn’t exist already might look a bit like the following:

import pathlib

import streamlit as st


HERE = pathlib.Path(__file__).parent

base = st.config.get_option("server.baseUrlPath")

all_links = []

if len(base) != 0:
    base += '/'

for path in HERE.rglob('*.py'):
    relative_path = path.relative_to(HERE)
    suffix_removed = relative_path.with_suffix("")
    
    url = str(suffix_removed).replace('\\', '/')
    url = f'/{base}{url}'
    
    if suffix_removed.name == "index":
        url = '/'.join(url.split('/')[:-1])
        
    all_links.append(url)
        
markdown_urls = "\n\n".join([f"[{link}]({link})" for link in all_links])
markdown_urls

This index.py template could be further improved by also importing watchdog, and when the directory tree changes (such as a new python file added), then the following function could be run:

def rerun():
    raise st.ScriptRunner.RerunException(st.ScriptRequestQueue.RerunData(None))

Which would rerun the script, and update the URLs 😃.

---

[EDIT]

Turns out running the following within a watchdog thread doesn’t actually cause the parent thread to rerun, instead it just causes the watchdog thread to error out and stop:

def rerun():
    raise st.ScriptRunner.RerunException(st.ScriptRequestQueue.RerunData(None))

However the following PR addresses a means to rerun a streamlit session from within a watchdog thread:

https://github.com/streamlit/streamlit/pull/2060

[/EDIT]