As seen in https://til.simonwillison.net/sqlite/import-csv - .mode csv and then .import school.csv schools is hugely faster than importing via sqlite-utils insert and doing the work in Python - but it can only be implemented by shelling out to the sqlite3 CLI tool, it's not functionality that is exposed to the Python sqlite3 module.

An option to use this would be useful - maybe something like this:

sqlite-utils insert blah.db blah blah.csv --fast

... which raises the challenge that datasette publish doesn't yet know what to do with a config file!

Originally posted by @simonw in https://github.com/simonw/datasette/issues/2194#issuecomment-1730259871

I've started to encounter a bug where queries to tables inside named in-memory databases sometimes trigger server hangs.

I'm still trying to figure out what's going on here - on one occasion I managed to Ctrl+C the server and saw an exception that mentioned a thread lock, but usually hitting Ctrl+C does nothing and I have to kill -9 the PID instead.

This is all running on my M2 Mac.

I've seen the bug in the Datasette 1.0 alphas and in Datasette 0.64.3 - but reverting to 0.61 appeared to fix it.

Got this running tests against Python 3.11.

../../../.local/share/virtualenvs/datasette-big-local-6Yn-280V/lib/python3.11/site-packages/datasette/app.py:14: in <module> import pkg_resources ../../../.local/share/virtualenvs/datasette-big-local-6Yn-280V/lib/python3.11/site-packages/pkg_resources/__init__.py:121: in <module> warnings.warn("pkg_resources is deprecated as an API", DeprecationWarning) E DeprecationWarning: pkg_resources is deprecated as an API

I ran with pytest -Werror --pdb -x to get the debugger for that warning, but it turned out searching the code worked better. It's used in these two places:

https://github.com/simonw/datasette/blob/5890a20c374fb0812d88c9b0ef26a838bfa06c76/datasette/plugins.py#L43-L50

https://github.com/simonw/datasette/blob/5890a20c374fb0812d88c9b0ef26a838bfa06c76/datasette/app.py#L1037

https://github.com/simonw/datasette/actions/runs/6266449018/job/17017460074

This is a bit of a mystery, I don't think I've changed anything recently that could have broken this.

That passed on 3.8 but should have failed: https://github.com/simonw/datasette/actions/runs/6266341481/job/17017099801 - the "Test DATASETTE_LOAD_PLUGINS" test shows errors but did not fail the CI run.

Originally posted by @simonw in https://github.com/simonw/datasette/issues/2057#issuecomment-1730201226

Refs: - #2189

:books: Documentation preview :books:: https://datasette--2192.org.readthedocs.build/en/2192/

Sometimes a plugin may want to add content to an existing default template - for example datasette-search-all adds a new search box at the top of index.html. I also want datasette-upload-csvs to add a CTA on the database.html page: https://github.com/simonw/datasette-upload-csvs/issues/18

Currently plugins can do this by providing a new version of the index.html template - but if multiple plugins try to do that only one of them will succeed.

It would be better if there were known areas of those templates which plugins could add additional content to, such that multiple plugins can use the same spot.

The cause of this bug: - https://github.com/simonw/datasette-export-notebook/issues/17

Is that datasette-export-notebook was consulting data["filtered_table_rows_count"] in the render output plugin function in order to show the total number of rows that would be exported.

That field is no longer available by default - the "count" field is only available if ?_extra=count was passed.

It would be useful if plugins like this could access the total count on demand, should they need to.

This made me think. That {"$env": "ENV_VAR"} hack was introduced back here:

• https://github.com/simonw/datasette/issues/538

The problem it was solving was that metadata was visible to everyone with access to the instance at /-/metadata but plugins clearly needed a way to set secret settings.

Now that this stuff is moving to config, we have some decisions to make:

1. Add /-/config to let people see the configuration of their instance, and keep the $env trick for secret settings.
2. Say all configuration aside from metadata is secret and make $env optional or ditch it entirely.
3. Allow plugins to announce which of their configuration options are secret so we can automatically redact them from /-/config

I've found /-/metadata extraordinarily useful as a user of Datasette - it really helps me understand exactly what's going on if I run into any problems with a plugin, if I can quickly check what the settings look like.

So I'm leaning towards option 1 or 3.

Originally posted by @simonw in https://github.com/simonw/datasette/pull/2183#discussion_r1325076924

Also refs: - #2093

Given this schema: sql CREATE TABLE departments ( campus_name TEXT NOT NULL, dept_code TEXT NOT NULL, dept_name TEXT, PRIMARY KEY (campus_name, dept_code) ); CREATE TABLE courses ( course_code TEXT PRIMARY KEY, course_name TEXT, campus_name TEXT NOT NULL, dept_code TEXT NOT NULL, FOREIGN KEY (campus_name, dept_code) REFERENCES departments(campus_name, dept_code) ); The output of db["courses"].foreign_keys right now is: [ForeignKey(table='courses', column='campus_name', other_table='departments', other_column='campus_name'), ForeignKey(table='courses', column='dept_code', other_table='departments', other_column='dept_code')] Which suggests two normal foreign keys, not one compound foreign key.

There's a long justification for why this might be needed here: - https://github.com/simonw/datasette-auth-tokens/issues/10#issuecomment-1701820001

Short version: it would be neat if it was possible to stash some data on the request object such that a later plugin/middleware-type-thing could use that to influence the final returned response - similar to the kinds of things you can do with Django middleware.

The asgi_wrapper() mechanism doesn't have access to the request or response objects - it gets scope and can mess around with receive and send, but those are pretty low-level primitives.

Since Datasette has well-defined request and response objects now it might be nice to have a middleware layer that can manipulate those directly.

I just spotted a bug when using https://datasette.io/plugins/datasette-configure-fts and https://datasette.io/plugins/datasette-edit-schema at the same time.

Steps to reproduce:

• Configure FTS for a table, then run a test search
• Edit the schema for that table and change the order of columns
• Run the test search again

I got the wrong search results, which I think is because the _fts table pointed to the first table by rowid but those rowid values were entirely rewritten as a consequence of running table.transform() on the table.

Reconfiguring FTS on the table fixed the problem.

I think table.transform() should be able to preserve rowid values.

Refs: - #592

• [x] Tests against weird shaped tables

I need to test that this works against:

• rowid tables
• Tables that have a column called rowid even though they are not rowid tables

:books: Documentation preview :books:: https://sqlite-utils--593.org.readthedocs.build/en/593/

In building Datasette Cloud we realized that a bunch of the features we are building need a way of resolving an actor ID to the actual actor, in order to display something more interesting than just an integer ID.

Social plugins in particular need this - comments by X, CSV uploaded by X, that kind of thing.

I think the solution is a new plugin hook: actors_from_ids(datasette, ids) which can return a list of actor dictionaries.

The default implementation can return [{"id": "..."}] for the IDs passed to it.

Pluggy has a firstresult=True option which is relevant here, since this is the first plugin hook we will have implemented where only one plugin should provide an answer.

In: - https://github.com/simonw/datasette.io/issues/152#issuecomment-1712323347

This turned out to be the fix:

bash dogsheep-beta index dogsheep-index.db templates/dogsheep-beta.yml sqlite-utils rebuild-fts dogsheep-index.db

I'm going to try and get this working through an environment variable, so that you can start Datasette and it will only load a subset of plugins including those that use the register_commands() hook.

Initial research on this: - https://github.com/pytest-dev/pluggy/issues/422

Refs: - #2180

This plugin hook is feature complete - including documentation and tests.

I'm not going to land it in Datasette main until we've used it at least once though, which should happen promptly in development for Datasette Cloud.

:books: Documentation preview :books:: https://datasette--2181.org.readthedocs.build/en/2181/

Spotted this problem while working on this plugin: - https://github.com/simonw/datasette-public

It's possible to make a table public to any users - but then you may end up with situations like this:

That table is public, but the foreign key links go to tables that are NOT public.

We're also leaking the names of the values in those private tables here, which we shouldn't do. So this is a tiny bit of an information leak.

Since this only affects people who have configured a table to be public that has foreign keys to a table that is private I don't think this is worth issuing a vulnerability report about - I very much doubt anyone is running Datasette configured in a way that could result in problems because of this.

This test here: https://github.com/simonw/datasette/blob/fbcb103c0cb6668018ace539a01a6a1f156e8d6a/tests/test_api.py#L1011-L1020

It failed for me like this:

E AssertionError: assert [('normal', False), ('sqlite_stat1', True), ('sqlite_stat4', True)] in ([('normal', False), ('sqlite_stat1', True)],)

Looks like some builds of SQLite include a sqlite_stat4 table.

This came up in discussion over: - https://github.com/simonw/datasette/pull/2174

https://dev.to/hugovk/help-test-python-312-beta-1508/

:books: Documentation preview :books:: https://sqlite-utils--591.org.readthedocs.build/en/591/

https://dev.to/hugovk/help-test-python-312-beta-1508/

:books: Documentation preview :books:: https://datasette--2175.org.readthedocs.build/en/2175/

Now available as a wheel: - https://github.com/click-contrib/click-default-group/issues/21

:books: Documentation preview :books:: https://datasette--2173.org.readthedocs.build/en/2173/

Currently the constructor accepts memory=True or memory_name=... and uses those to create a connection, but does not record what those values were:

https://github.com/simonw/sqlite-utils/blob/1260bdc7bfe31c36c272572c6389125f8de6ef71/sqlite_utils/db.py#L307-L349

This makes it hard to tell if a database object is to an in-memory or a file-based database, which is sometimes useful to know.

I used a custom SQL function in a migration script and then realized that it should be de-registered before the end of the script to avoid leaking into the calling code.

I've used this pattern twice now: https://til.simonwillison.net/datasette/register-new-plugin-hooks - in datasette-graphql and datasette-low-disk-space-hook. I should describe the pattern on https://docs.datasette.io/en/stable/writing_plugins.html

I noticed that a token with execute-sql permission alone did not work, because it was not allowed to view the instance of the database.

The return type isn't documented here: https://github.com/simonw/datasette/blob/4c3ef033110407f3b3dbce501659d523724985e0/docs/internals.rst#L327-L350

On inspecting the code I'm not 100% sure if it's possible for this. method to return None, or if it can only return True or False. Need to confirm that.

https://github.com/simonw/datasette/blob/4c3ef033110407f3b3dbce501659d523724985e0/datasette/app.py#L822C15-L853

The BaseView class has a method for cascading permission checks, but it's not easily accessible to plugins.

https://github.com/simonw/datasette/blob/5eb8e9bf250b26e30b017d39a392c33973997656/datasette/views/base.py#L75-L99

This leaves plugins like datasette-graphql having to implement their own versions of this logic, which is bad: https://github.com/simonw/datasette-graphql/issues/65

First check view-database - if that says False then disallow access, if it says True then allow access. If it says None check view-instance.

This should become a supported API that plugins are encouraged to use.

I just had to write a one-off custom plugin to redirect niche-musems.com to www.niche-museums.com (https://github.com/simonw/museums/issues/21) - it would be great if this kind of thing could be handled by a configurable plugin.

https://github.com/simonw/museums/blob/6b1faf00c463b2228860d4d62d104b11935e01b1/plugins/redirect_www.py

• 2164

TODO:

• [x] Automated tests
• [ ] Documentation
• [x] Make sure DATASETTE_LOAD_PLUGINS='' works for loading zero plugins

Discussed with Alex this morning. We think the American spelling is fine here (it's shorter than catalogue) and that it's a slightly less lazy name than core_.

Follows: - https://github.com/simonw/datasette/issues/2157

Spotted a problem while working on this: if you grant a token access to view table for a specific table but don't also grant view database and view instance permissions, that token is useless.

This was a deliberate design decision in Datasette - it's documented on https://docs.datasette.io/en/1.0a2/authentication.html#access-permissions-in-metadata

If a user cannot access a specific database, they will not be able to access tables, views or queries within that database. If a user cannot access the instance they will not be able to access any of the databases, tables, views or queries.

I'm now second-guessing if this was a good decision.

Originally posted by @simonw in https://github.com/simonw/datasette-auth-tokens/issues/7#issuecomment-1636031702

Refs: - #2102

Also includes a prototype implementation of --actor option which I'm using for testing this, from: - #2153

:books: Documentation preview :books:: https://datasette--2154.org.readthedocs.build/en/2154/

I experimented with a prototype of this here: - https://github.com/simonw/datasette/issues/2102#issuecomment-1691037971_

Which lets me run requests as if they belonged to a specific actor like this:

bash datasette fixtures.db --get '/fixtures/facetable.json' --actor '{ "_r": { "r": { "fixtures": { "facetable": [ "vt" ] } } }, "a": "user" }'

Really useful for testing actors an _r options. Is this worth adding as a feature?

This change updates the -s/--setting option to datasette serve to allow it to be used to set arbitrarily complex nested settings in a way that is compatible with the new -c datasette.yml work happening in: - #2143

It will enable things like this: datasette data.db --setting plugins.datasette-ripgrep.path "/home/simon/code" For the moment though it just affects settings - so you can do this: datasette data.db --setting settings.sql_time_limit_ms 3500 I've also implemented a backwards compatibility mechanism, so if you use it this way (the old way): datasette data.db --setting sql_time_limit_ms 3500 It will notice that the setting you passed is one of Datasette's core settings, and will treat that as if you said settings.sql_time_limit_ms instead.

:books: Documentation preview :books:: https://datasette--2161.org.readthedocs.build/en/2161/

I've been thinking about what it might look like to allow command-line arguments to be used to define any of the configuration options in datasette.yml, as alternative and more convenient syntax.

Here's what I've come up with: datasette \ -s settings.sql_time_limit_ms 1000 \ -s plugins.datasette-auth-tokens.manage_tokens true \ -s plugins.datasette-auth-tokens.manage_tokens_database tokens \ mydatabase.db tokens.db Which would be equivalent to datasette.yml containing this: yaml plugins: datasette-auth-tokens: manage_tokens: true manage_tokens_database: tokens settings: sql_time_limit_ms: 1000 More details in https://github.com/simonw/datasette/issues/2143#issuecomment-1690792514

This came up working on this feature: - https://github.com/simonw/llm/pull/186

I have a table with this schema: sql CREATE TABLE [collections] ( [id] INTEGER PRIMARY KEY, [name] TEXT, [model] TEXT ); CREATE UNIQUE INDEX [idx_collections_name] ON [collections] ([name]); So the primary key is an integer (because it's going to have a huge number of rows foreign key related to it, and I don't want to store a larger text value thousands of times), but there is a unique constraint on the name - that would be the primary key column if not for all of those foreign keys.

Problem is, fetching the collection by name is actually pretty inconvenient.

Fetch by numeric ID:

python try: table["collections"].get(1) except NotFoundError: # It doesn't exist Fetching by name: python def get_collection(db, collection): rows = db["collections"].rows_where("name = ?", [collection]) try: return next(rows) except StopIteration: raise NotFoundError("Collection not found: {}".format(collection)) It would be neat if, for columns where we know that we should always get 0 or one result, we could do this instead: python try: collection = table["collections"].get(name="entries") except NotFoundError: # It doesn't exist The existing .get() method doesn't have any non-positional arguments, so using **kwargs like that should work:

https://github.com/simonw/sqlite-utils/blob/1260bdc7bfe31c36c272572c6389125f8de6ef71/sqlite_utils/db.py#L1495

In thinking about the best way to implement https://github.com/simonw/datasette-auth-passwords/issues/6 (SQL-backed user accounts for datasette-auth-passwords) I realized that there are a few different use-cases where a plugin might want to store data that isn't visible to regular Datasette users:

• Storing password hashes
• Storing API tokens
• Storing secrets that are used for data import integrations (secrets for talking to the Twitter API for example)

Idea: allow one or more private database files to be attached to Datasette, something like this:

datasette github.db linkedin.db -s secrets.db -m metadata.yml

The secrets.db file would not be visible using any of the Datasette's usual interface or API routes - but plugins would be able to run queries against it.

So datasette-auth-passwords might then be configured like this:

yaml plugins: datasette-auth-passwords: database: secrets sql: "select password_hash from passwords where username = :username" The plugin could even refuse to operate against a database that hadn't been loaded as a secret database.

See: - https://github.com/simonw/datasette-configure-fts/issues/14 - https://github.com/simonw/datasette-auth-tokens/issues/12

It's EOL now, as of 2023-06-27 (two days ago): https://devguide.python.org/versions/

I'm going to drop Python 3.7.

Originally posted by @simonw in https://github.com/simonw/datasette/issues/1153#issuecomment-1627455892

It's not supported any more: https://devguide.python.org/versions/

I want to use isort to order imports. A few steps here:

• [x] Add a .isort.cfg file (see below)
• [x] Use isort -rc to reformat existing code
• [ ] Commit this change
• [x] Add a unit test that ensures future changes remain isort compatible

Also added a lint unit test to ensure they stay sorted. #516

It is increasingly being useful configuration options, when it started out as purely metadata.

Could cause confusion with the --config mechanism though - maybe that should be called "settings" instead?

Split off from #274

Metadata should an optional "config" section at root, table or database level.

The TableView and RowView and DatabaseView and BaseView classes could all have a .config("key") method which knows how to resolve the hierarchy of configs.

This will allow individual tables (or databases) to set their own config settings for things like sql_time_limit_ms

Stumbled across this while experimenting with datasette-write-ui. The error I got was a 500 on the /db page:

'NoneType' object has no attribute 'encode'

Tracked it down to this code, which assembles the URL for a row page:

https://github.com/simonw/datasette/blob/943df09dcca93c3b9861b8c96277a01320db8662/datasette/utils/init.py#L120-L134

That's because tilde_encode can't handle None: https://github.com/simonw/datasette/blob/943df09dcca93c3b9861b8c96277a01320db8662/datasette/utils/init.py#L1175-L1178

Extremely detailed story of how I got to this point:

• https://github.com/simonw/llm/issues/162

Steps to reproduce (only if that pragma is on though): bash python -c ' import sqlite_utils db = sqlite_utils.Database(memory=True) db.execute(""" CREATE TABLE "logs" ( [id] INTEGER PRIMARY KEY, [model] TEXT, [prompt] TEXT, [system] TEXT, [prompt_json] TEXT, [options_json] TEXT, [response] TEXT, [response_json] TEXT, [reply_to_id] INTEGER, [chat_id] INTEGER REFERENCES [log]([id]), [duration_ms] INTEGER, [datetime_utc] TEXT ); """) db["logs"].add_foreign_key("reply_to_id", "logs", "id") ' This succeeds in some environments, fails in others.

https://github.com/simonw/sqlite-utils/blob/13ebcc575d2547c45e8d31288b71a3242c16b886/sqlite_utils/db.py#L1165-L1174

This is the only place in the code that attempts to modify sqlite_master directly, which fails on some Python installations.

Could this use the .transform() trick instead?

Or automatically switch to that trick if it hits an error?

I got this error trying to drop a column from a table that was part of a SQL view:

error in view plugins: no such table: main.pypi_releases

Upon further investigation I found that this pattern seemed to fix it: python def transform_the_table(conn): # Run this in a transaction: with conn: # We have to read all the views first, because we need to drop and recreate them db = sqlite_utils.Database(conn) views = {v.name: v.schema for v in db.views if table.lower() in v.schema.lower()} for view in views.keys(): db[view].drop() db[table].transform( types=types, rename=rename, drop=drop, column_order=[p[0] for p in order_pairs], ) # Now recreate the views for name, schema in views.items(): db.create_view(name, schema) So grab a copy of any view that might reference this table, start a transaction, drop those views, run the transform, recreate the views again.

I wonder if this should become an option in sqlite-utils? Maybe a recreate_views=True argument for table.tranform(...)? Should it be opt-in or opt-out?

Originally posted by @simonw in https://github.com/simonw/datasette-edit-schema/issues/35#issuecomment-1683370548

The new options added in: - #577 Deserve consideration in the CLI as well.

https://github.com/simonw/sqlite-utils/blob/d2bcdc00c6ecc01a6e8135e775ffdb87572b802b/sqlite_utils/db.py#L1706-L1708

I have a neater pattern for this now - instead of: https://github.com/simonw/sqlite-utils/blob/1dc6b5aa644a92d3654f7068110ed7930989ce71/tests/test_create.py#L472-L475

I now prefer:

https://github.com/simonw/sqlite-utils/blob/1dc6b5aa644a92d3654f7068110ed7930989ce71/tests/test_create.py#L1163-L1171

Refs: - #577

:books: Documentation preview :books:: https://sqlite-utils--584.org.readthedocs.build/en/584/

For example this: https://github.com/simonw/datasette/blob/4535568f2ce907af646304d0ebce2500ebd55677/docs/authentication.rst?plain=1#L33-L35

The problem with that $ prefix is that it prevents users from copying and pasting the raw command.

https://docs.datasette.io/en/stable/authentication.html#using-the-root-actor

I think this has been a bug for a long time.

https://latest.datasette.io/ currently shows:

Those colors are based on a hash of the database name. But when you click through to https://latest.datasette.io/fixtures

It's red on all sub-pages too.

Spotted this on https://latest.datasette.io/extra_database

```html

This idea first came up here: - https://github.com/simonw/datasette/issues/2112#issuecomment-1652751140

If datasette.render_template(...) takes an optional Context subclass as an alternative to a context dictionary, I could then use dataclasses to define the context made available to specific templates - which then gives me something I can use to help document what they are.

Also refs: - https://github.com/simonw/datasette/issues/1510

This would be useful while working on: - #2114

Or... how about if the on_success_message option could define a SQL query to be executed to generate that message? Maybe on_success_message_sql.

• https://github.com/simonw/datasette/issues/2134

The default JSON just isn't right. I find myself using ?_shape=array for almost everything I build against the API.

118 commits! https://github.com/simonw/datasette/compare/1.0a2...26be9f0445b753fb84c802c356b0791a72269f25

I just noticed that https://latest.datasette.io/fixtures/roadside_attraction_characteristics.json?_labels=on&_size=1 returns: json { "ok": true, "next": "1", "rows": [ { "rowid": 1, "attraction_id": { "value": 1, "label": "The Mystery Spot" }, "characteristic_id": { "value": 2, "label": "Paranormal" } } ], "truncated": false } But https://latest.datasette.io/fixtures.json?sql=select+rowid%2C+attraction_id%2C+characteristic_id+from+roadside_attraction_characteristics+order+by+rowid+limit+1 returns: json { "rows": [ { "rowid": 1, "attraction_id": 1, "characteristic_id": 2 } ], "columns": [ "rowid", "attraction_id", "characteristic_id" ], "ok": true, "truncated": false } The columns key in the query response is inconsistent with the table response.

This URL here:

https://latest.datasette.io/fixtures/simple_primary_key/1.json?_extras=foreign_key_tables

json { "database": "fixtures", "table": "simple_primary_key", "rows": [ { "id": "1", "content": "hello" } ], "columns": [ "id", "content" ], "primary_keys": [ "id" ], "primary_key_values": [ "1" ], "units": {}, "foreign_key_tables": [ { "other_table": "foreign_key_references", "column": "id", "other_column": "foreign_key_with_blank_label", "count": 0, "link": "/fixtures/foreign_key_references?foreign_key_with_blank_label=1" }, { "other_table": "foreign_key_references", "column": "id", "other_column": "foreign_key_with_label", "count": 1, "link": "/fixtures/foreign_key_references?foreign_key_with_label=1" }, { "other_table": "complex_foreign_keys", "column": "id", "other_column": "f3", "count": 1, "link": "/fixtures/complex_foreign_keys?f3=1" }, { "other_table": "complex_foreign_keys", "column": "id", "other_column": "f2", "count": 0, "link": "/fixtures/complex_foreign_keys?f2=1" }, { "other_table": "complex_foreign_keys", "column": "id", "other_column": "f1", "count": 1, "link": "/fixtures/complex_foreign_keys?f1=1" } ], "query_ms": 4.226590999678592, "source": "tests/fixtures.py", "source_url": "https://github.com/simonw/datasette/blob/main/tests/fixtures.py", "license": "Apache License 2.0", "license_url": "https://github.com/simonw/datasette/blob/main/LICENSE", "ok": true, "truncated": false }

That ?_extras= should be ?_extra= - plus the row JSON should be redesigned to fit the new default JSON representation.

• 2109

From Discord: https://discordapp.com/channels/823971286308356157/996877076982415491/1068227071156965486

I ended up duplicating it here: https://github.com/simonw/datasette/blob/7532feb424b1dce614351e21b2265c04f9669fe2/datasette/views/database.py#L164-L167

I think it should move to datasette.render_template() - and maybe have a renamed template variable too.

• 2109

Caused by: - #2112

https://latest.datasette.io/fixtures?sql=select+pk1%2C+pk2%2C+pk3%2C+content+from+compound_three_primary_keys+where+%22pk1%22+%3D+%3Ap0+order+by+pk1%2C+pk2%2C+pk3+limit+101&p0=b

The :p0 form field is missing. Submitting the form results in this error:

https://latest.datasette.io/_memory.csv?sql=select+blah is a blank page right now:

bash curl -I 'https://latest.datasette.io/_memory.csv?sql=select+blah' HTTP/2 200 access-control-allow-origin: * access-control-allow-headers: Authorization, Content-Type access-control-expose-headers: Link access-control-allow-methods: GET, POST, HEAD, OPTIONS access-control-max-age: 3600 content-type: text/plain; charset=utf-8 x-databases: _memory, _internal, fixtures, fixtures2, extra_database, ephemeral date: Mon, 07 Aug 2023 23:12:15 GMT server: Google Frontend

Originally posted by @simonw in https://github.com/simonw/datasette/issues/2118#issuecomment-1668688947

In the old code:

https://github.com/simonw/datasette/blob/08181823990a71ffa5a1b57b37259198eaa43e06/datasette/views/database.py#L34-L35

This name class attribute was later used by some of the plugin hooks, passed as view_name: https://github.com/simonw/datasette/blob/18dd88ee4d78fe9d760e9da96028ae06d938a85c/datasette/hookspecs.py#L50-L54

Figure out how that should work once I've refactored those classes to view functions instead.

Refs: - #2109

• 2109

This will help make the hook as robust as possible. - #2109

This will be the base that the remaining work builds on top of. Refs: - #2109

While working on: - https://github.com/simonw/datasette/pull/2118

It became clear that the render callback function documented here: https://docs.datasette.io/en/0.64.3/plugin_hooks.html#register-output-renderer-datasette

Needs to grow the ability to be told if an error occurred (an error string) and if the results were truncated (a truncated boolean).

WIP. Refs: - #2109

:books: Documentation preview :books:: https://datasette--2118.org.readthedocs.build/en/2118/

What would it take to add a format hook? I'm still thinking about my GIS workflow, and being able to do sqlite-utils query ... --geojson would be nice. It's the one place my Datasette workflow is messy, having to do datasette . --get /path/to/query.geojson --setting max_rows_returned 10000 --load-extension spatialite. I know the current pattern is --csv, but maybe --format geojson is more future-proof.

https://discord.com/channels/823971286308356157/997738192360964156/1133076679011602432

Currently looks like this: https://github.com/simonw/datasette/blob/08181823990a71ffa5a1b57b37259198eaa43e06/datasette/init.py#L1-L6

Adding __all__ = ["Permission", "Forbidden"...] would let me get rid of those # noqa comments.

I've been stuck on this for too long. I'm breaking it down into a full milestone:

https://github.com/simonw/datasette/milestone/29

While using sqlite-utils convert I realized it would be handy if you could pass --pdb to have it open the debugger at the first instance of a failed conversion.

Refs: - #2109

The idea here is that hitting /content without a ?sql= will show an empty result set AND default to including a bunch of extras about the list of tables in the database.

Then I won't have to think about /content and /content?sql= as separate pages any more.

A minor refactor, but it is a good starting point for this new branch. Refs: - #2109

Refs: - #2049

TODO:

• [x] Read queries JSON
• Implement error display with "ok": false and an errors key
• Read queries HTML
• Read queries other formats (plugins)
• Canned read queries (dispatched to from table)
• Write queries (a canned query thing)
• Implement different shapes, refactoring to share code with table
• Implement a sensible subset of extras, also refactoring to share code with table
• Get all tests passing

:books: Documentation preview :books:: https://datasette--2053.org.readthedocs.build/en/2053/

As seen in LLM and now in sqlite-utils too: - https://github.com/simonw/sqlite-utils/issues/570

Useful for developing plugins, see tutorial at https://llm.datasette.io/en/stable/plugins/tutorial-model-plugin.html

Currently the changelog starts at 0.4:

https://sqlite-utils.datasette.io/en/3.34/changelog.html#id115

I want the other releases - according to https://pypi.org/project/sqlite-utils/#history there are three missing:

Plugins affecting the CLI by default makes sense to me.

I'm less confident about them always affecting users of the Python API.

I'm going to have them apply by default, but I'm going to add a mechanism to opt-out on an individual database basis. Basically this:

```python from sqlite_utils import Database db = Database(memory=True, execute_plugins=False)

Anything using db from here on will not execute plugins

``` cc @asg017

Refs: - #567 - #574

Splitting off an issue for prepare_connection() since Alex got the PR in seconds before I shipped 3.34!

Originally posted by @simonw in https://github.com/simonw/sqlite-utils/issues/567#issuecomment-1646686424

Also need a design for an option for the .transform() method to indicate that the new table should be created with a new name without dropping the old one.

I think keep_table="name_of_table" is good for this.

Originally posted by @simonw in https://github.com/simonw/sqlite-utils/issues/565#issuecomment-1646657324

Spotted this in the GitHub Actions logs:

... and it turns out running .transform() with no arguments still fixes the format of the schema!

```pycon

db["log"].add_column("foo", str)

For every command/API method there should be a link to the equivalent in the other form factor.

Maybe also link to the API and CLI reference pages too.

I find myself wanting two new features in sqlite-utils: - The ability to have the new transformed table set to a specific name, while keeping the old table around - The ability to rename a table (sqlite-utils doesn't have a table rename function at all right now)

Originally posted by @simonw in https://github.com/simonw/llm/issues/65#issuecomment-1618375042

Found myself using this pattern to quickly prototype a schema:

```python import sqlite_utils db = sqlite_utils.Database(memory=True)

print(db["answers_chunks"].create({ "id": int, "content": str, "embedding_type_id": int, "embedding": bytes, "embedding_content_md5": str, "source": str, }, pk="id", transform=True).schema) ```

Using replace=True to drop and then recreate the table would be neat here, and would be consistent with other places that use replace=True.

I'm going to start by adding the register_command hook using the exact same pattern as Datasette and LLM.

Originally posted by @simonw in https://github.com/simonw/sqlite-utils/issues/567#issuecomment-1646643450

As seen in LLM.

Needed while working on: - #567

Plugins can then take advantage of this pattern, for example: - https://github.com/simonw/datasette-auth-tokens/pull/8

Maybe put it on the <tr> but probably better to go on the td.type-pk.

I thought this would never happen, but now that I'm deep in the weeds of running SQLite in production for Datasette Cloud I'm starting to reconsider my policy of only supporting SQLite.

Some of the factors making me think PostgreSQL support could be worth the effort: - Serverless. I'm getting increasingly excited about writable-database use-cases for Datasette. If it could talk to PostgreSQL then users could easily deploy it on Heroku or other serverless providers that can talk to a managed RDS-style PostgreSQL. - Existing databases. Plenty of organizations have PostgreSQL databases. They can export to SQLite using db-to-sqlite but that's a pretty big barrier to getting started - being able to run datasette postgresql://connection-string and start trying it out would be a massively better experience. - Data size. I keep running into use-cases where I want to run Datasette against many GBs of data. SQLite can do this but PostgreSQL is much more optimized for large data, especially given the existence of tools like Citus. - Marketing. Convincing people to trust their data to SQLite is potentially a big barrier to adoption. Even if I've convinced myself it's trustworthy I still have to convince everyone else. - It might not be that hard? If this required a ground-up rewrite it wouldn't be worth the effort, but I have a hunch that it may not be too hard - most of the SQL in Datasette should work on both databases since it's almost all portable SELECT statements. If Datasette did DML this would be a lot harder, but it doesn't. - Plugins! This feels like a natural surface for a plugin - at which point people could add MySQL support and suchlike in the future.

The above reasons feel strong enough to justify a prototype.

https://discord.com/channels/823971286308356157/823971286941302908/1120705940728066080

May I add a request to the whole metadata / settings ? Allow to pass --settings path/to/settings.json instead of having to rely exclusively on directory mode to centralize settings (this would reflect the behavior of providing metadata)