issue_comments

New docs section on the need to call await ds.invoke_startup(): https://github.com/simonw/datasette/blob/ddc999ad1296e8c69cffede3e367dda059b8adad/docs/testing_plugins.rst#setting-up-a-datasette-test-instance

In looking at the documentation on writing tests, there are a lot of examples like this: python def test_that_opens_the_debugger_or_errors(): ds = Datasette([db_path], pdb=True) response = await ds.client.get("/") I really don't like having to tell people to add await ds.invoke_startup() to every test that might look like this.

Since it's safe to call that function multiple times, I'm going to have ds.client.get() and friends call it for you too - so if you forget in a plugin test it won't matter.

Datasette's own tests started to break because calls to the TestClient were performed without awaiting that method. I fixed that by adding this to _request() inside that class:

python async def _request( self, path, follow_redirects=True, redirect_count=0, method="GET", cookies=None, headers=None, post_body=None, content_type=None, if_none_match=None, ): if not self.ds._startup_invoked: await self.ds.invoke_startup()

This is likely going to cause some tests in plugins to break, but I'm OK with that - I'll fix them as I find them once this release is out.

I'm going to throw an error in ds.render_template() if you haven't previously called await ds.invoke_startup().

It looks like Datasette Lite does NOT invoke that method, which is likely a bug: https://github.com/simonw/datasette-lite/blob/e7ccaf621b3cdf613ebaf544304d387f2af32edf/webworker.js#L103-L110

It might be good to have Datasette LOUDLY fail if you attempt to use it without calling await ds.invoke_startup().

It looks like this is the only place that calls invoke_startup(): https://github.com/simonw/datasette/blob/1d64c9a8dac45b9a3452acf8e76dfadea2b0bc49/datasette/cli.py#L590-L591

datasette-publish-vercel is the one deployment mechanism that skips running Uvicorn, and it calls that method separately here: https://github.com/simonw/datasette-publish-vercel/blob/1559d979b4e3b1f2f83c51c3c0c10192ff9a6d0c/datasette_publish_vercel/init.py#L42-L52

python ds = Datasette( [], {database_files}, static_mounts=static_mounts, metadata=metadata{extras}, secret=secret, cors=True, settings={settings}{crossdb} ) asyncio.run(ds.invoke_startup()) app = ds.app() So preparing the Jinja environment inside that function would work fine.

The reason to support await_me_maybe is in case a hook wants to execute a SQL query as part of configuring the Jinja template loader.

That's exactly what datasette-edit-templates needs to do, though it's currently achieving that in a startup() hook instead: https://github.com/simonw/datasette-edit-templates/blob/087f6a6cabc20020f2b0524f11aa3a7836320848/datasette_edit_templates/init.py#L32-L48

```python @hookimpl def startup(datasette): datasette._edit_templates = {}

async def inner():
    db = get_database(datasette)
    # Does the table exist?
    if not await db.table_exists(TABLE):
        for sql in CREATE_TABLE:
            await db.execute_write(sql, block=True)
    else:
        # Load all templates from that table
        rows = await db.execute("select template, body FROM {}".format(TABLE))
        for name, content in rows:
            datasette._edit_templates[name] = content

return inner

```

await_me_maybe might be tricky though, because the only place that hook is executed is here:

https://github.com/simonw/datasette/blob/8430c3bc7dd22b173c1a8c6cd7180e3b31240cd1/datasette/app.py#L348

Which is inside the Datasette.__init__ method.

To implement an await on that would need to move it somewhere else - probably here: https://github.com/simonw/datasette/blob/8430c3bc7dd22b173c1a8c6cd7180e3b31240cd1/datasette/app.py#L391-L393

But I'm not 100% confident that is always executed at the right time to ensure the Jinja environment is properly configured? I think it is, but would need to make sure.

It should also implement the await_me_maybe pattern so you can return an async function from it.