DuckDB has a read-only mechanism: https://duckdb.org/docs/api/python

import duckdb
con = duckdb.connect(database="/tmp/blah.db", read_only=True)

It's not possible yet, unfortunately. This came up on the forums recently: https://github.com/simonw/datasette/discussions/968

I'm leaning further towards making the database connection layer itself work via a plugin hook, which would open up the possibility of supporting DuckDB and other databases as well. I've not committed to doing this yet though.

One reason to prioritize this issue: Homebrew upgraded to SpatiaLite 5.0 recently https://formulae.brew.sh/formula/spatialite-tools and as a result SpatiaLite database created on my laptop don't appear to be compatible with Datasette when published using datasette publish.

I tried this patch against Dockerfile:

diff --git a/Dockerfile b/Dockerfile
index f4b1414..dd659e1 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -1,25 +1,26 @@
-FROM python:3.7.10-slim-stretch as build
+FROM python:3.9.2-slim-buster as build

 # Setup build dependencies
 RUN apt update \
-&& apt install -y python3-dev build-essential wget libxml2-dev libproj-dev libgeos-dev libsqlite3-dev zlib1g-dev pkg-config git \
- && apt clean
+  && apt install -y python3-dev build-essential wget libxml2-dev libproj-dev \
+  libminizip-dev libgeos-dev libsqlite3-dev zlib1g-dev pkg-config git \
+  && apt clean

-
-RUN wget "https://www.sqlite.org/2020/sqlite-autoconf-3310100.tar.gz" && tar xzf sqlite-autoconf-3310100.tar.gz \
-    && cd sqlite-autoconf-3310100 && ./configure --disable-static --enable-fts5 --enable-json1 CFLAGS="-g -O2 -DSQLITE_ENABLE_FTS3=1 -DSQLITE_ENABLE_FTS3_PARENTHESIS -DSQLITE_ENABLE_FTS4=1 -DSQLITE_ENABLE_RTREE=1 -DSQLITE_ENABLE_JSON1" \
+RUN wget "https://www.sqlite.org/2021/sqlite-autoconf-3340100.tar.gz" && tar xzf sqlite-autoconf-3340100.tar.gz \
+    && cd sqlite-autoconf-3340100 && ./configure --disable-static --enable-fts5 --enable-json1 \
+    CFLAGS="-g -O2 -DSQLITE_ENABLE_FTS3=1 -DSQLITE_ENABLE_FTS3_PARENTHESIS -DSQLITE_ENABLE_FTS4=1 -DSQLITE_ENABLE_RTREE=1 -DSQLITE_ENABLE_JSON1" \
     && make && make install

-RUN wget "http://www.gaia-gis.it/gaia-sins/freexl-sources/freexl-1.0.5.tar.gz" && tar zxf freexl-1.0.5.tar.gz \
-    && cd freexl-1.0.5 && ./configure && make && make install
+RUN wget "http://www.gaia-gis.it/gaia-sins/freexl-1.0.6.tar.gz" && tar zxf freexl-1.0.6.tar.gz \
+    && cd freexl-1.0.6 && ./configure && make && make install

-RUN wget "http://www.gaia-gis.it/gaia-sins/libspatialite-sources/libspatialite-4.4.0-RC0.tar.gz" && tar zxf libspatialite-4.4.0-RC0.tar.gz \
-    && cd libspatialite-4.4.0-RC0 && ./configure && make && make install
+RUN wget "http://www.gaia-gis.it/gaia-sins/libspatialite-5.0.1.tar.gz" && tar zxf libspatialite-5.0.1.tar.gz \
+    && cd libspatialite-5.0.1 && ./configure --disable-rttopo && make && make install

 RUN wget "http://www.gaia-gis.it/gaia-sins/readosm-sources/readosm-1.1.0.tar.gz" && tar zxf readosm-1.1.0.tar.gz && cd readosm-1.1.0 && ./configure && make && make install

-RUN wget "http://www.gaia-gis.it/gaia-sins/spatialite-tools-sources/spatialite-tools-4.4.0-RC0.tar.gz" && tar zxf spatialite-tools-4.4.0-RC0.tar.gz \
-    && cd spatialite-tools-4.4.0-RC0 && ./configure && make && make install
+RUN wget "http://www.gaia-gis.it/gaia-sins/spatialite-tools-5.0.0.tar.gz" && tar zxf spatialite-tools-5.0.0.tar.gz \
+    && cd spatialite-tools-5.0.0 && ./configure --disable-rttopo && make && make install


 # Add local code to the image instead of fetching from pypi.
@@ -27,7 +28,7 @@ COPY . /datasette

 RUN pip install /datasette

-FROM python:3.7.10-slim-stretch
+FROM python:3.9.2-slim-buster

 # Copy python dependencies and spatialite libraries
 COPY --from=build /usr/local/lib/ /usr/local/lib/

I had to use --disable-rttopo from the tip in https://github.com/OSGeo/gdal/pull/3443 and also needed to install libminizip-dev.

This works, sort of... I'm getting a weird issue where the /dbname page is hanging some of the time instead of loading correctly. Other than that it seems to work, but a hanging page is bad!

Worth noting that the Docker image used by datasette publish cloudrun doesn't actually use that Datasette docker image - it does this:

https://github.com/simonw/datasette/blob/d0fd833b8cdd97e1b91d0f97a69b494895d82bee/datasette/utils/__init__.py#L349-L353

Where the apt extras for SpatiaLite are: https://github.com/simonw/datasette/blob/d0fd833b8cdd97e1b91d0f97a69b494895d82bee/datasette/utils/__init__.py#L344-L345

libsqlite3-mod-spatialite against that official python:3.8 image doesn't appear to install SpatiaLite 5.0.

Apologies if I sound dense but I don't see where you would pass
'shell=True'. I'm using the CLI installed via pip.

On Sun., Mar. 7, 2021, 2:15 a.m. David Smith, notifications@github.com
wrote:

To get it to work I had to:

-

add shell=true to the various commands in datasette
-

use the name argument of the publish command. (
https://docs.datasette.io/en/stable/publish.html)

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
https://github.com/simonw/datasette/issues/858#issuecomment-792230560,
or unsubscribe
https://github.com/notifications/unsubscribe-auth/AAJJTOMZMGYSCGUU4J3AVSDTCMRX5ANCNFSM4ODNEDYA
.

This is fantastic, thanks so much for tracking this down.

To get it to work I had to:

• add shell=true to the various commands in datasette

• use the name argument of the publish command. (https://docs.datasette.io/en/stable/publish.html)

@smithdc1 Can you tell us what you did to get it to publish in Windows? What commands did you pass?

Hello,
I have the same wildcards search problems with an instance of Datasette. http://crbc-dataset.huma-num.fr/inventaires/fonds_auguste_dupouy_1872_1967?_search=gwerz&_sort=rowid is OK but http://crbc-dataset.huma-num.fr/inventaires/fonds_auguste_dupouy_1872_1967?_search=gwe* is not (FTS is activated on "Reference" "IntituleAnalyse" "NomDuProducteur" "PresentationDuContenu" "Notes").

Notice that a SQL query as below launched directly from SQLite in the server's shell, retrieves results.

select * from fonds_auguste_dupouy_1872_1967_fts where IntituleAnalyse MATCH "gwe*";

Thanks,

I just tried to run this on a small VPS instance with 2GB of memory and it crashed out of memory while processing a 12GB mbox from Takeout.

Is it possible to stream the emails to sqlite instead of loading it all into memory and upserting at once?

@maxhawkins a limitation of the python mbox module is it loads the entire mbox into memory. I did find another approach to this problem that didn't use the builtin python mbox module and created a generator so that it didn't have to load the whole mbox into memory. I was hoping to use standard library modules, but this might be a good reason to investigate that approach a bit more. My worry is making sure a custom processor handles all the ins and outs of the mbox format correctly.

Hm. As I'm writing this, I thought of something. I think I can parse each message one at a time, and then use an mbox function to load each message using the python mbox module. That way the mbox module can still deal with the specifics of the mbox format, but I can use a generator.

I'll give that a try. Thanks for the feedback @maxhawkins and @simonw. I'll give that a try.

@simonw can we hold off on merging this until I can test this new approach?

I just tried to run this on a small VPS instance with 2GB of memory and it crashed out of memory while processing a 12GB mbox from Takeout.

Is it possible to stream the emails to sqlite instead of loading it all into memory and upserting at once?

I am getting the same thing for US West (N. California) us-west-1

Sorry for the delay, I got sidetracked after class last night. I am getting the following error:

/content# google-takeout-to-sqlite mbox takeout.db Takeout/Mail/gmail.mbox 
Usage: google-takeout-to-sqlite [OPTIONS] COMMAND [ARGS]...Try 'google-takeout-to-sqlite --help' for help.

Error: No such command 'mbox'.

On the box, I installed with pip after cloning: https://github.com/UtahDave/google-takeout-to-sqlite.git

@rgieseke Ah, that's super helpful. Thank you for the workaround for now!

I'm not sure why but my most recent import, when displayed in Datasette, looks like this:

https://user-images.githubusercontent.com/9599/109985836-0ab00080-7cba-11eb-97d5-0631a0835b61.png">

Sorting by id in the opposite order gives me the data I would expect - so it looks like a bunch of null/blank messages are being imported at some point and showing up first due to ID ordering.

I imported my 10GB mbox with 750,000 emails in it, ran this tool (with a hacked fix for the blob column problem) - and now a search that returns 92 results takes 25.37ms! This is fantastic.

Solution could be to pre-process that string by splitting on ( and dropping everything afterwards, assuming that the (...) bit isn't necessary for correctly parsing the date.

I added this code to output a message ID on errors:

             print("Errors: {}".format(num_errors))
             print(traceback.format_exc())
+            print("Message-Id: {}".format(email.get("Message-Id", "None")))
             continue

Having found a message ID that had an error, I ran this command to see the context:

rg --text --context 20 '44F289B0.000001.02100@SCHWARZE-DWFXMI' ~/gmail.mbox

This was for the following error:

  File "/Users/simon/Dropbox/Development/google-takeout-to-sqlite/google_takeout_to_sqlite/utils.py", line 102, in get_mbox
    message["date"] = get_message_date(email.get("Date"), email.get_from())
  File "/Users/simon/Dropbox/Development/google-takeout-to-sqlite/google_takeout_to_sqlite/utils.py", line 178, in get_message_date
    datetime_tuple = email.utils.parsedate_tz(mail_date)
  File "/Library/Developer/CommandLineTools/Library/Frameworks/Python3.framework/Versions/3.7/lib/python3.7/email/_parseaddr.py", line 50, in parsedate_tz
    res = _parsedate_tz(data)
  File "/Library/Developer/CommandLineTools/Library/Frameworks/Python3.framework/Versions/3.7/lib/python3.7/email/_parseaddr.py", line 69, in _parsedate_tz
    data = data.split()
AttributeError: 'Header' object has no attribute 'split'

Here's what I spotted in the ripgrep output:

177133570:Message-Id: <44F289B0.000001.02100@SCHWARZE-DWFXMI>
177133571-Date: Mon, 28 Aug 2006 08:14:08 +0200 (Westeurop�ische Sommerzeit)
177133572-X-Mailer: IncrediMail (5002253)

So it could it be that _parsedate_tz is having trouble with that Mon, 28 Aug 2006 08:14:08 +0200 (Westeurop�ische Sommerzeit) string.

Looks like you're doing this:

python elif message.get_content_type() == "text/plain": body = message.get_payload(decode=True)

So presumably that decodes to a unicode string?

I imagine the reason the column is a BLOB for me is that sqlite-utils determines the column type based on the first batch of items - https://github.com/simonw/sqlite-utils/blob/09c3386f55f766b135b6a1c00295646c4ae29bec/sqlite_utils/db.py#L1927-L1928 - and I got unlucky and had something in my first batch that wasn't a unicode string.

Ah, that's good to know. I think explicitly creating the tables will be a great improvement. I'll add that.

Also, I noticed after I opened this PR that the message.get_payload() is being deprecated in favor of message.get_content() or something like that. I'll see if that handles the decoding better, too.

Thanks for the feedback. I should have time tomorrow to put together some improvements.

The command takes quite a while to start running, presumably because this line causes it to have to scan the WHOLE file in order to generate a count:

https://github.com/dogsheep/google-takeout-to-sqlite/blob/a3de045eba0fae4b309da21aa3119102b0efc576/google_takeout_to_sqlite/utils.py#L66-L67

I'm fine with waiting though. It's not like this is a command people run every day - and without that count we can't show a progress bar, which seems pretty important for a process that takes this long.

The wait is from python loading the mbox file. This happens regardless if you're getting the length of the mbox. The mbox module is on the slow side. It is possible to do one's own parsing of the mbox, but I kind of wanted to avoid doing that.

3 also mentions the conflicting version with other tools.

Looks like you're doing this:

    elif message.get_content_type() == "text/plain":
        body = message.get_payload(decode=True)

So presumably that decodes to a unicode string?

I imagine the reason the column is a BLOB for me is that sqlite-utils determines the column type based on the first batch of items - https://github.com/simonw/sqlite-utils/blob/09c3386f55f766b135b6a1c00295646c4ae29bec/sqlite_utils/db.py#L1927-L1928 - and I got unlucky and had something in my first batch that wasn't a unicode string.

Confirmed: removing the len() call does not speed things up, so it's reading through the entire file for some other purpose too.

It looks like the body is being loaded into a BLOB column - so in Datasette default it looks like this:

https://user-images.githubusercontent.com/9599/109924808-b4b96980-7c75-11eb-8c9e-307f2ae32d5a.png">

If I datasette install datasette-render-binary and then try again I get this:

https://user-images.githubusercontent.com/9599/109924944-ea5e5280-7c75-11eb-9a32-404f3d68455f.png">

It would be great if we could store the body as unicode text instead. May have to do something clever to decode it based on some kind of charset header?

I got 9 warnings that look like this:

Errors: 1
Traceback (most recent call last):
  File "/Users/simon/Dropbox/Development/google-takeout-to-sqlite/google_takeout_to_sqlite/utils.py", line 103, in get_mbox
    message["date"] = get_message_date(email.get("Date"), email.get_from())
  File "/Users/simon/Dropbox/Development/google-takeout-to-sqlite/google_takeout_to_sqlite/utils.py", line 167, in get_message_date
    datetime_tuple = email.utils.parsedate_tz(mail_date)
  File "/Library/Developer/CommandLineTools/Library/Frameworks/Python3.framework/Versions/3.7/lib/python3.7/email/_parseaddr.py", line 50, in parsedate_tz
    res = _parsedate_tz(data)
  File "/Library/Developer/CommandLineTools/Library/Frameworks/Python3.framework/Versions/3.7/lib/python3.7/email/_parseaddr.py", line 69, in _parsedate_tz
    data = data.split()
AttributeError: 'Header' object has no attribute 'split'

It would be useful if those warnings told me the message ID (or similar) of the affected message so I could grep for it in the mbox and see what was going on.

I'm not sure if it would work, but there is an alternative pattern for showing a progress bar against a really large file that I've used in healthkit-to-sqlite - you set the progress bar size to the size of the file in bytes, then update a counter as you read the file.

https://github.com/dogsheep/healthkit-to-sqlite/blob/3eb2b06bfe3b4faaf10e9cf9dfcb28e3d16c14ff/healthkit_to_sqlite/cli.py#L24-L57 and https://github.com/dogsheep/healthkit-to-sqlite/blob/3eb2b06bfe3b4faaf10e9cf9dfcb28e3d16c14ff/healthkit_to_sqlite/utils.py#L4-L19 (the progress_callback() bit) is where that happens.

It can be a bit of a convoluted pattern, and I'm not at all sure it would work for mbox files since it looks like that library has other reasons it needs to do a file scan rather than streaming it through one chunk of bytes at a time. So I imagine this would not work here.

The command takes quite a while to start running, presumably because this line causes it to have to scan the WHOLE file in order to generate a count:

https://github.com/dogsheep/google-takeout-to-sqlite/blob/a3de045eba0fae4b309da21aa3119102b0efc576/google_takeout_to_sqlite/utils.py#L66-L67

I'm fine with waiting though. It's not like this is a command people run every day - and without that count we can't show a progress bar, which seems pretty important for a process that takes this long.

The Rich-powered progress bar is pretty:

Wow, my mbox is a 10.35 GB download!

Thanks!

It's kind of an ugly hack, but you can try out what using the fts5 table as an actual datasette-accessible table looks like without changing any datasette code by creating yet another view on top of the fts5 table:

create view proxyview as select *, rank, table_fts as fts from table_fts;

That's now visible from datasette, just like any other view, but you can use fts match escape_fts(search_string) order by rank.

This is only good as a proof of concept because you're inefficiently going from view -> fts5 external content table -> view -> data table. However, it does show it works.

I am just seeing this sorry, yes! I will kick the tires later on tonight. My apologies for the delay.

One note on using this pragma I got an error on starting datasette no such table: pragma_database_list.

I diagnosed this to an older version of sqlite3 (3.14.2) and upgrading to a newer version (3.34.2) fixed the issue.

In FTS5, I think doing an FTS search is actually much easier than doing a join against the main table like datasette does now. In fact, FTS5 external content tables provide a transparent interface back to the original table or view.

Here's what I'm currently doing: * build a view that joins whatever tables I want and rename the columns to non-joiny names (e.g, chapter.name AS chapter_name in the view where needed) * Create an FTS5 table with content="viewname" * As described in the "external content tables" section (https://www.sqlite.org/fts5.html#external_content_tables), sql queries can be made directly to the FTS table, which behind the covers makes select calls to the content table when the content of the original columns are needed. * In addition, you get "rank" and "bm25()" available to you when you select on the _fts table.

Unfortunately, datasette doesn't currently seem happy being coerced into doing a real query on an fts5 table. This works:
select col1, col2, col3 from table_fts where coll1="value" and table_fts match escape_fts("search term") order by rank

But this doesn't work in the datasette SQL query interface:
select col1, col2, col3 from table_fts where coll1="value" and table_fts match escape_fts(:search) order by rank (the "search" input text field doesn't show up)

For what datasette is doing right now, I think you could just use contentless fts5 tables (content=""), since all you care about is the rowid since all you're doing a subselect to get the rowid anyway. In fts5, that's just a contentless table.

I guess if you want to follow this suggestion, you'd need a somewhat different code path for fts5.

A custom templates/index.html seems to work and custom pages as a workaround with moving them to pages/base_url_dir.

I like the .add_memory_database() option. I also like that it makes it more obvious that this is a capability of Datasette, since I'm excited to see more plugins, features and tests that take advantage of it.

A couple of options:

datasette.add_memory_database("test_json_array")
# or make that first argument to add_database() optional and support:
datasette.add_database(memory_name="test_json_array")

I'm going to try a new pattern for testing this, enabled by #1151 - the test will create a new named in-memory database, write some records to it and then run some test facets against that. This will save me from having to add yet another fixtures table for this.

It's out! https://github.com/encode/httpx/releases/tag/0.17.0

Microsoft's playwright Python library solves this problem by code generating both their sync AND their async libraries https://github.com/microsoft/playwright-python/tree/master/scripts

Hah or use this trick, which genuinely rewrites the code at runtime using a class decorator! https://github.com/python-happybase/aiohappybase/blob/0990ef45cfdb720dc987afdb4957a0fac591cb99/aiohappybase/sync/_util.py#L19-L32

Some other interesting background reading: https://docs.sqlalchemy.org/en/14/orm/extensions/asyncio.html - in particular see how SQLALchemy has a await conn.run_sync(meta.drop_all) mechanism for running methods that haven't themselves been provided in an async version

From the poster here: https://github.com/sethmlarson/pycon-async-sync-poster/blob/master/poster.pdf

https://user-images.githubusercontent.com/9599/109401634-9f0a1400-7904-11eb-8b3a-37df0678b8dc.png">

unasync is an implementation of the exact pattern I was talking about above - it uses the tokenize module from the Python standard library to apply some clever rules to transform an async codebase into a sync one. https://unasync.readthedocs.io/en/latest/ - implementation here: https://github.com/python-trio/unasync/blob/v0.5.0/src/unasync/__init__.py

"how to use some new tools to more easily maintain a codebase that supports both async and synchronous I/O and multiple async libraries" - yeah that's exactly what I need, thank you!

I had this resource by Seth Michael Larson saved https://github.com/sethmlarson/pycon-async-sync-poster I haven't had a look at it, but it may contain useful info.

On twitter, I mentioned passing an aiosqlite connection during the Database creation. I'm not 100% familiar with the sqlite-utils codebase, so I may be wrong here, but maybe decorating internal functions could be an option? Then they are awaited or not inside the decorator depending on how they are called.

Here's a really wild idea: I wonder if it would be possible to run a source transformation against either the sync or the async versions of the code to produce the equivalent for the other paradigm?

Could that even be as simple as a set of regular expressions against the await ... version that strips out or replaces the await and async def and async for statements?

If so... I could maintain just the async version, generate the sync version with a script and rely on robust unit testing to guarantee that this actually works.

I have a hunch this is actually going to be quite difficult, due to the internal complexity of some of the sqlite-utils API methods.

Consider db[table].extract(...) for example. It does a whole bunch of extra queries inside the method - each of those would need to be turned into an await call for the async version. Here's the method body today:

https://github.com/simonw/sqlite-utils/blob/09c3386f55f766b135b6a1c00295646c4ae29bec/sqlite_utils/db.py#L1060-L1152

Writing this method twice - looking similar but with await ... tucked in before every internal method it calls that needs to execute SQL - is going to be pretty messy.

One thing that would help a LOT is figuring out how to share the majority of the test code. If the exact same tests could run against both the sync and async versions with a bit of test trickery, maintaining parallel implementations would at least be a bit more feasible.

I think HTTPX gets it exactly right, with a clear separation between sync and async clients, each with a basically identical API. (I'm about to switch feed-to-sqlite over to it, from Requests, to eventually make way for async support.)

Another option here would be to add https://github.com/omnilib/aiosqlite/blob/main/aiosqlite/core.py as a dependency - it's four years old now and actively marinated, and the code is pretty small so it looks like a solid, stable, reliable dependency.

Datasette has its own implementation of a write queue for exactly this purpose - and there's no reason at all that should stay in Datasette rather than being extracted out and moved over here to sqlite-utils.

One small concern I have is around the API design. I'd want to keep supporting the existing synchronous API while also providing a similar API with await-based methods.

What are some good examples of libraries that do this? I like how https://www.python-httpx.org/ handles it, maybe that's a good example to imitate?

Thanks!

I requested my Gmail export from takeout - once that arrives I'll test it against this and then merge the PR.

Here's the test I wrote:

git diff tests/test_custom_pages.py
diff --git a/tests/test_custom_pages.py b/tests/test_custom_pages.py
index 6a23192..5a71f56 100644
--- a/tests/test_custom_pages.py
+++ b/tests/test_custom_pages.py
@@ -2,11 +2,19 @@ import pathlib
 import pytest
 from .fixtures import make_app_client

+TEST_TEMPLATE_DIRS = str(pathlib.Path(__file__).parent / "test_templates")
+

 @pytest.fixture(scope="session")
 def custom_pages_client():
+    with make_app_client(template_dir=TEST_TEMPLATE_DIRS) as client:
+        yield client
+
+
+@pytest.fixture(scope="session")
+def custom_pages_client_with_base_url():
     with make_app_client(
-        template_dir=str(pathlib.Path(__file__).parent / "test_templates")
+        template_dir=TEST_TEMPLATE_DIRS, config={"base_url": "/prefix/"}
     ) as client:
         yield client

@@ -23,6 +31,12 @@ def test_request_is_available(custom_pages_client):
     assert "path:/request" == response.text


+def test_custom_pages_with_base_url(custom_pages_client_with_base_url):
+    response = custom_pages_client_with_base_url.get("/prefix/request")
+    assert 200 == response.status
+    assert "path:/prefix/request" == response.text
+
+
 def test_custom_pages_nested(custom_pages_client):
     response = custom_pages_client.get("/nested/nest")
     assert 200 == response.status

I added a debug line just before for regex, wildcard_template here:

https://github.com/simonw/datasette/blob/afed51b1e36cf275c39e71c7cb262d6c5bdbaa31/datasette/app.py#L1148-L1155

And it showed that for some reason request.path is /prefix/prefix/request here - the prefix got doubled somehow.

Sounds like a bug - thanks for reporting this.

Could I take this part:

             suggested_facet_sql = """ 
                 select distinct json_type({column}) 
                 from ({sql}) 
             """.format( 
                 column=escape_sqlite(column), sql=self.sql 
             ) 

And add where {column} is not null and {column} != '' perhaps?

Relevant code: https://github.com/simonw/datasette/blob/afed51b1e36cf275c39e71c7cb262d6c5bdbaa31/datasette/facets.py#L271-L295

Could this handle lists of objects too? That would be pretty amazing - if the column has a [{...}, {...}] list in it could turn that into a many-to-many.

Linking to rows from custom queries is a lot harder - because given an arbitrary string of SQL it's difficult to analyze it and figure out which (if any) of the returned columns represent a primary key.

It's possible to manually write a SQL query that returns a column that will be treated as a link to another page using this plugin, but it's not particularly straight-forward: https://datasette.io/plugins/datasette-json-html

Agreed, this would be extremely useful. I'd love to be able to facet against custom queries. It's a fair bit of work to implement but it's not impossible. Closing this as a duplicate of #972.

If there's no primary key in the JSON could use the hash_id mechanism.

This came up in office hours!

New paragraph at the top of https://docs.datasette.io/en/latest/writing_plugins.html

Want to start by looking at an example? The Datasette plugins directory lists more than 50 open source plugins with code you can explore. The plugin hooks page includes links to example plugins for each of the documented hooks.

sqlite-utils create-view currently has a --ignore option, so adding that to sqlite-utils drop-view and sqlite-utils drop-table makes sense as well.

So:

    db["my_table"].drop(ignore=True)

Other methods (db.create_view() for example) have ignore=True to mean "don't throw an error if this causes a problem", so I'm good with adding that to .drop_view().

I don't like using it as the default partly because that would be a very minor breaking API change, but mainly because I don't want to hide mistakes people make - e.g. if you mistype the name of the table you are trying to drop.

Example from the docs:

>>> db = sqlite_utils.Database(memory=True)
>>> db["dogs"].insert({"name": "Cleo"})
>>> for pk, row in db["dogs"].pks_and_rows_where():
...     print(pk, row)
1 {'rowid': 1, 'name': 'Cleo'}

>>> db["dogs_with_pk"].insert({"id": 5, "name": "Cleo"}, pk="id")
>>> for pk, row in db["dogs_with_pk"].pks_and_rows_where():
...     print(pk, row)
5 {'id': 5, 'name': 'Cleo'}

>>> db["dogs_with_compound_pk"].insert(
...     {"species": "dog", "id": 3, "name": "Cleo"},
...     pk=("species", "id")
... )
>>> for pk, row in db["dogs_with_compound_pk"].pks_and_rows_where():
...     print(pk, row)
('dog', 3) {'species': 'dog', 'id': 3, 'name': 'Cleo'}

Documentation: https://sqlite-utils.datasette.io/en/latest/python-api.html#listing-rows-with-their-primary-keys

WIP in a pull request.

I prototyped this and I like it:

In [1]: import sqlite_utils
In [2]: db = sqlite_utils.Database("/Users/simon/Dropbox/Development/datasette/fixtures.db")
In [3]: list(db["compound_primary_key"].pks_and_rows_where())
Out[3]: [(('a', 'b'), {'pk1': 'a', 'pk2': 'b', 'content': 'c'})]

table.pks_and_rows_where(...) is explicit and I think less ambiguous than the other options.

I quite like pks_with_rows_where(...) - but grammatically it suggests it will return the primary keys that exist where their rows match the criteria - "pks with rows" can be interpreted as "pks for the rows that..." as opposed to "pks accompanied by rows"

Is pk_rows_where() a good name? It sounds like it returns "primary key rows" which isn't a thing. It actually returns rows along with their primary key.

Other options:

• table.rows_with_pk_where(...) - should this return (row, pk) rather than (pk, row)?
• table.rows_where_pk(...)
• table.pk_and_rows_where(...)
• table.pk_with_rows_where(...)
• table.pks_with_rows_where(...) - because rows is pluralized, so pks should be pluralized too?
• table.pks_rows_where(...)

One option: .rows_where() could grow a ensure_pk=True option which checks to see if the table is a rowid table and, if it is, includes that in the select.

Or... how about you can call .rows_where(..., pks=True) and it will yield (pk, rowdict) tuple pairs instead of just returning the sequence of dictionaries?

I'm always a little bit nervous of methods that vary their return type based on their arguments. Maybe this would be a separate method instead?

    for pk, row in table.pk_rows_where(...):
        # ...

Here's the current implementation of .extract(): https://github.com/simonw/sqlite-utils/blob/806c21044ac8d31da35f4c90600e98115aade7c6/sqlite_utils/db.py#L1049-L1074

Tricky detail here: I create the lookup table first, based on the types of the columns that are being extracted.

I need to do this because extraction currently uses unique tuples of values, so the table has to be created in advance.

But if I'm using these new expand functions to figure out what's going to be extracted, I don't know the names of the columns and their types in advance. I'm only going to find those out during the transformation.

This may turn out to be incompatible with how .extract() works at the moment. I may need a new method, .extract_expand() perhaps? It could be simpler - work only against a single column for example.

I can still use the existing sqlite-utils extract CLI command though, with a --json flag and a rule that you can't run it against multiple columns.

Problem with calling this argument transform= is that the term "transform" already means something else in this library.

I could use convert= instead.

... but that doesn't instantly make me think of turning a value into multiple columns.

How about expand=? I've not used that term anywhere yet.

db["Reports"].extract(["Reported by"], expand={"Reported by": json.loads})

I think that works. You're expanding a single value into several columns of information.

I'm going to go with last-wins - so if multiple transform functions return the same key the last one will over-write the others.

Maybe the Python version takes an optional dictionary mapping column names to transformation functions? It could then merge all of those results together - and maybe throw an error if the same key is produced by more than one column.

    db["Reports"].extract(["Reported by"], transform={"Reported by": json.loads})

Or it could have an option for different strategies if keys collide: first wins, last wins, throw exception, add a prefix to the new column name. That feels a bit too complex for an edge-case though.

It would be OK if the CLI version only allows you to specify a single column if you are using the --json option.

For the Python version I'd like to be able to provide a transformation callback function - which can be json.loads but could also be anything else which accepts the value of the current column and returns a Python dictionary of columns and their values to use in the new table.

Likewise the sqlite-utils extract command takes one or more columns:

Usage: sqlite-utils extract [OPTIONS] PATH TABLE COLUMNS...

  Extract one or more columns into a separate table

Options:
  --table TEXT             Name of the other table to extract columns to
  --fk-column TEXT         Name of the foreign key column to add to the table
  --rename <TEXT TEXT>...  Rename this column in extracted table

The Python .extract() method currently starts like this:

def extract(self, columns, table=None, fk_column=None, rename=None):
        rename = rename or {}
        if isinstance(columns, str):
            columns = [columns]
        if not set(columns).issubset(self.columns_dict.keys()):
            raise InvalidColumns(
                "Invalid columns {} for table with columns {}".format(
                    columns, list(self.columns_dict.keys())
                )
            )
        ...

Note that it takes a list of columns (and treats a string as a single item list). That's because it can be called with a list of columns and it will use them to populate another table of unique tuples of those column values.

So a new mechanism that can instead read JSON values from a single column needs to be compatible with that existing design.

I bet the bug is in here: https://github.com/simonw/sqlite-utils/blob/806c21044ac8d31da35f4c90600e98115aade7c6/sqlite_utils/db.py#L593-L602

Codecov Report

Merging #1243 (887bfd2) into main (726f781) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##             main    #1243   +/-   ##
=======================================
  Coverage   91.56%   91.56%           
=======================================
  Files          34       34           
  Lines        4242     4242           
=======================================
  Hits         3884     3884           
  Misses        358      358           

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 726f781...32652d9. Read the comment docs.

I noticed that @simonw is using black for formatting. I ran black on my additions in this PR.

I really like the way the Share feature on Stack Overflow works: https://stackoverflow.com/questions/18934149/how-can-i-use-postgresqls-text-column-type-in-django

I think it's possible that many users these days no longer assume they can paste a URL from the browser address bar (if they ever understood that at all) because to many apps are SPAs with broken URLs.

Absolutely, that's why I thought my corner case with iframe preventing access to the datasette URL could actually be relevant in more general situations.

A "Share link" button would only be needed on the table page and the arbitrary query page I think - and maybe on the row page, especially as that page starts to grow more features in the future.

This can definitely be done with a plugin.

Adding to Datasette itself is an interesting idea. I think it's possible that many users these days no longer assume they can paste a URL from the browser address bar (if they ever understood that at all) because to many apps are SPAs with broken URLs.

The shareable URLs are actually a key feature of Datasette - so maybe they should be highlighted in the default UI?

I built a "copy to clipboard" feature for datasette-copyable and wrote up how that works here: https://til.simonwillison.net/javascript/copy-button

Likewise, while answering to another issue regarding the Vega plugin, I realized that there is no such way of linking rows after a custom query, I only get this "Link" column with individual URLs for the default SQL view:

Or is it there and I am just missing the option in my custom queries?

Topline this fixed the same problem for me.

brew install python@3.7
ln -s /usr/local/opt/python@3.7/bin/python3.7 /usr/local/opt/python/bin/python3.7
pip3 uninstall -y numpy
pip3 uninstall -y setuptools
pip3 install setuptools
pip3 install numpy
pip3 install datasette-publish-fly

Also, @simonw I created a test based off the existing tests. I think it's working correctly

Bug is here: https://github.com/simonw/datasette/blob/42caabf7e9e6e4d69ef6dd7de16f2cd96bc79d5b/datasette/filters.py#L149-L165

Those json_each lines should be:

select {t}.rowid from {t}, json_each([{t}].[{c}]) j

@Btibert3 I've opened a PR with my initial attempt at this. Would you be willing to give this a try?

https://github.com/dogsheep/google-takeout-to-sqlite/pull/5

This is another change which is a little bit hard to figure out because I haven't solved #878 yet.

I'm not going to work on this for a while, but if anyone has needs or ideas around that they can add them to this issue.

It's good on mobile - iOS at least. Going to close this open new issues if anyone reports bugs.

Actually, coming back to this, I have a clearer use case for enabling fts generation for views: making it easier to bring in text from lookup tables and other joins.

The datasette documentation describes populating an fts table like so:

INSERT INTO "items_fts" (rowid, name, description, category_name)
    SELECT items. rowid,
    items.name,
    items.description,
    categories.name
    FROM items JOIN categories ON items.category_id=categories.id;

Alternatively if you have fts support in sqlite_utils for views (which sqlite and fts5 support), you can do the same thing just by creating a view that captures the above joins as columns, then creating an fts table from that view. Such an fts table can be created using sqlite_utils, where one created with your method can't.

The resulting fts table can then be used by a whole family of related tables and views in the manner you described earlier in this issue.

Hi! I don't think Prettier supports this syntax for globs: datasette/static/*[!.min].js Are you sure that works?
Prettier uses https://github.com/mrmlnc/fast-glob, which in turn uses https://github.com/micromatch/micromatch, and the docs for these packages don't mention this syntax. As per the docs, square brackets should work as in regexes (foo-[1-5].js).

Tested it. Apparently, it works as a negated character class in regexes (like [^.min]). I wonder where this syntax comes from. Micromatch doesn't support that:

micromatch(['static/table.js', 'static/n.js'], ['static/*[!.min].js']);
// result: ["static/n.js"] -- brackets are treated like [!.min] in regexes, without negation