issues

Context

We ran into this bug randomly, noticing that deleted ROWID would get reused after migrating the DB. Using transform to change any column in the table will also unexpectedly strip away the AUTOINCREMENT keyword from the primary key definition, even if it was not the transformation target.

Reproducible example

Original database

```sql $ sqlite3 test.db << EOF CREATE TABLE mytable ( col1 INTEGER PRIMARY KEY AUTOINCREMENT, col2 TEXT NOT NULL ) EOF

$ sqlite3 test.db ".schema mytable" CREATE TABLE mytable ( col1 INTEGER PRIMARY KEY AUTOINCREMENT, col2 TEXT NOT NULL ); ```

Modified database after sqlite-utils

```sql $ sqlite-utils transform test.db mytable --rename col2 renamedcol2

$ sqlite3 test.db "SELECT sql FROM sqlite_master WHERE name = 'mytable';" CREATE TABLE IF NOT EXISTS "mytable" ( [col1] INTEGER PRIMARY KEY, [renamedcol2] TEXT NOT NULL ); ```

https://github.com/simonw/sqlite-utils-plugins should be in the official documentation.

I can use the same pattern as https://llm.datasette.io/en/stable/plugins/directory.html

https://til.simonwillison.net/readthedocs/stable-docs

Currently using both insert-files and convert is needed in order to create sqlar files, it would be more convenient if it could be done with just one command.

```shell ~ ❯ cat test.py import os

class Example: def init(self, arg1, arg2): self.arg1 = arg1

~ ❯ sqlite-utils insert-files test.sqlar sqlar test.py -c name:name -c data:content -c mode:mode -c mtime:mtime -c sz:size --pk=name [####################################] 100%

~ ❯ sqlite-utils convert test.sqlar sqlar data "zlib.compress(value)" --import=zlib --where "name = 'test.py'" [####################################] 100%

~ ❯ cat test.py | sqlite-utils convert test.sqlar sqlar data "zlib.compress(sys.stdin.buffer.read())" --import=zlib --import=sys --where "name = 'test.py'" # Alternative way [####################################] 100%

~ ❯ sqlite3 test.sqlar "SELECT hex(data) FROM sqlar WHERE name = 'test.py';" | python3 -c "import sys, zlib; sys.stdout.buffer.write(zlib.decompress(bytes.fromhex(sys.stdin.read())))" import os

class Example: def init(self, arg1, arg2): self.arg1 = arg1

~ ❯ rm test.py

~ ❯ sqlar -l test.sqlar test.py

~ ❯ sqlar -x test.sqlar

~ ❯ cat test.py import os

class Example: def init(self, arg1, arg2): self.arg1 = arg1

```

When creating a new table, it would be good to have an option to set unique columns similar to how not_null is set.

```python from sqlite_utils import Database

columns = {"mRID": str, "name": str} db = Database("example.db") db["ExampleTable"].create(columns, pk="mRID", not_null=["mRID"], if_not_exists=True) db["ExampleTable"].create_index(["mRID"], unique=True, if_not_exists=True) ```

So something like this would add the UNIQUE flag to the table definition.

python db["ExampleTable"].create(columns, pk="mRID", not_null=["mRID"], unique=["mRID"], if_not_exists=True)

sql CREATE TABLE ExampleTable ( mRID TEXT PRIMARY KEY NOT NULL UNIQUE, name TEXT );

I was using sqlite-utils to create a DB from a CSV and it turns out the CSV contains a NUL byte.

When the processing reaches the line that contains the NUL an exception is raised.

I'm wondering if there is something that can be done in sqlite-utils to say "skip lines with encoding errors" or some such. I think it isn't super straightforward though as the exception comes from inside the csv module that does all the parsing.

Concretely the file is the KernelVersions.csv from https://www.kaggle.com/datasets/kaggle/meta-kaggle

This is the command and output: $ sqlite-utils insert --csv kaggle.db kaggle KernelVersions.csv [------------------------------------] 0% [#####################---------------] 60% 00:04:24Traceback (most recent call last): File "/home/foobar/miniconda/envs/meta-kaggle/bin/sqlite-utils", line 10, in <module> sys.exit(cli()) File "/home/foobar/miniconda/envs/meta-kaggle/lib/python3.10/site-packages/click/core.py", line 1128, in __call__ return self.main(*args, **kwargs) File "/home/foobar/miniconda/envs/meta-kaggle/lib/python3.10/site-packages/click/core.py", line 1053, in main rv = self.invoke(ctx) File "/home/foobar/miniconda/envs/meta-kaggle/lib/python3.10/site-packages/click/core.py", line 1659, in invoke return _process_result(sub_ctx.command.invoke(sub_ctx)) File "/home/foobar/miniconda/envs/meta-kaggle/lib/python3.10/site-packages/click/core.py", line 1395, in invoke return ctx.invoke(self.callback, **ctx.params) File "/home/foobar/miniconda/envs/meta-kaggle/lib/python3.10/site-packages/click/core.py", line 754, in invoke return __callback(*args, **kwargs) File "/home/foobar/miniconda/envs/meta-kaggle/lib/python3.10/site-packages/sqlite_utils/cli.py", line 1223, in insert insert_upsert_implementation( File "/home/foobar/miniconda/envs/meta-kaggle/lib/python3.10/site-packages/sqlite_utils/cli.py", line 1085, in insert_upsert_implementation db[table].insert_all( File "/home/foobar/miniconda/envs/meta-kaggle/lib/python3.10/site-packages/sqlite_utils/db.py", line 3198, in insert_all chunk = list(chunk) File "/home/foobar/miniconda/envs/meta-kaggle/lib/python3.10/site-packages/sqlite_utils/db.py", line 3742, in fix_square_braces for record in records: File "/home/foobar/miniconda/envs/meta-kaggle/lib/python3.10/site-packages/sqlite_utils/cli.py", line 1071, in <genexpr> docs = (decode_base64_values(doc) for doc in docs) File "/home/foobar/miniconda/envs/meta-kaggle/lib/python3.10/site-packages/sqlite_utils/cli.py", line 1068, in <genexpr> docs = (verify_is_dict(doc) for doc in docs) File "/home/foobar/miniconda/envs/meta-kaggle/lib/python3.10/site-packages/sqlite_utils/cli.py", line 1003, in <genexpr> docs = (dict(zip(headers, row)) for row in reader) _csv.Error: line contains NUL

According to the documentation, we can make a csv output using the CLI tool, but not the Python library. Could we have the latter?

sqlite-utils should detect and have specially handling for column with a JSON column. For example:

sql CREATE TABLE "dogs" ( id INTEGER PRIMARY KEY, name TEXT, friends JSON );

Automatic Nesting

According to "Nested JSON Values", sqlite-utils will only expand JSON if the --json-cols flag is passed. It looks like it'll try to json.load all text column to test if its JSON, which can get expensive on non-json columns.

Instead, sqlite-utils should be default (ie without the --json-cols flags) do the maybe_json() operation on columns with a declared JSON type. So the above table would expand the "friends" column as expected, withoutthe --json-cols flag:

bash sqlite-utils dogs.db "select * from dogs" | python -mjson.tool

[ { "id": 1, "name": "Cleo", "friends": [ { "name": "Pancakes" }, { "name": "Bailey" } ] } ]

I'm sure there's other ways sqlite-utils can specially handle JSON columns, so keeping this open while I think of more

It would be nice if we could specify a custom encoder (and decoder) for types that will need extra deserialisation – e.g., sets, enums or sparse matrices – or even project-specific types

I'm working on some small front-end enhancements to the laion-aesthetic-datasette project, and I wanted to partially populate a database directly using exports from the existing Datasette instance instead of downloading the parquet files and creating my own multi-GB database.

There have been a number of small issues that are certainly related to my relative lack of familiarity with the toolkit, but that are still surprising.

For example: a CSV export of the images table (http://laion-aesthetic.datasette.io/laion-aesthetic-6pls.csv?sql=select+rowid%2C+url%2C+text%2C+domain_id%2C+width%2C+height%2C+similarity%2C+punsafe%2C+pwatermark%2C+aesthetic%2C+hash%2C+index_level_0+from+images+order+by+random%28%29+limit+100) has nested single quotes, double quotes, and commas that aren't handled by rows_from_file. Similarly, the json output has to be manually transformed to add the column names and remove extraneous information before sqlite_utils can import it.

I was able to work through these issues, but as an enhancement it would be really helpful to create or document a clear workflow that avoids the friction of this data transformation.

Could offer syntactic sugar for:

sql create table foo as select * from bar

sqlite-utils create-table data.db foo --sql "select * from bar" https://sqlite-utils.datasette.io/en/stable/cli-reference.html#create-table

Maybe when calling .vacuum() and other DB-level write-lock operations sqlite_utils could guard against this error message by automatically committing first?

``` 46 db["media"].optimize() # type: ignore ---> 47 db.vacuum()

File ~/.local/lib/python3.10/site-packages/sqlite_utils/db.py:1047, in Database.vacuum(self) 1045 def vacuum(self): 1046 "Run a SQLite VACUUM against the database." -> 1047 self.execute("VACUUM;")

File ~/.local/lib/python3.10/site-packages/sqlite_utils/db.py:470, in Database.execute(self, sql, parameters) 468 return self.conn.execute(sql, parameters) 469 else: --> 470 return self.conn.execute(sql)

OperationalError: cannot VACUUM from within a transaction ```

It might also be nice to add a sentence or two about how transactions are committed on the docs page. When I was swapping out my sqlite3 code for this library it was nice that everything was pretty much drop-in but I was/am unsure what to do about the places I explicitly call .commit() in my code

Related to https://github.com/simonw/sqlite-utils/issues/121

Trying to follow the tutorial for sqlite-utils and datasette https://datasette.io/tutorials/clean-data on Windows 11 OS Microsoft Windows [Version 10.0.22622.440], with sqlite-utils and datasette installed using pipx.

pipx list package datasette 0.61.1, installed using Python 3.10.4 - datasette.exe package sqlite-utils 3.28, installed using Python 3.10.4 - sqlite-utils.exe

In the step to transform dates into ISO dates the quoted value 'r.parsedatetime(value)' is copied verbatim into the columns instead of applying the output of the Python recipe.

``` sqlite-utils convert manatees.db locations \ REPDATE created_date last_edited_date \ 'r.parsedatetime(value)' --dry-run

1975/01/31 00:00:00+00 --- becomes: r.parsedatetime(value)

Would affect 13568 rows ```

However, if I change the code from single quotes to double quotes, it works as expected.

``` sqlite-utils convert manatees.db locations \ REPDATE created_date last_edited_date \ "r.parsedatetime(value)" --dry-run

1975/01/31 00:00:00+00 --- becomes: 1975-01-31T00:00:00+00:00

Would affect 13568 rows ```

Specifying the transform code recipe should work with single quotes on Windows.

Inspired by this: https://news.ycombinator.com/item?id=29729283

Because SQLite is effectively serializing all the writes for us, we have zero locking in our code. We used to have to lock when inserting new items (to get the LastInsertRowId), but the newer version of SQLite supports the RETURNING keyword, so we don't even have to lock on inserts now.

Thank you to release this bravo project. When i use this project on multi table db, I want to implement convenient search on column name from different tables. I want to develop a meta table to save the meta data of different columns of different tables and search on this meta table to get rows from the data table (which the meta table describes) does this project provide some simple function on it ?

You can think a have a knowledge graph about the table in the db, and i save this knowledge graph into the db with fts enabled.

Refs #239. Still needs documentation and CLI implementation.

https://sqlite-utils.readthedocs.io/en/2.4.2/python-api.html#adding-columns says:

SQLite types you can specify are "TEXT", "INTEGER", "FLOAT" or "BLOB".

As discovered in #92 this isn't the right list of values. I should expand this to match https://www.sqlite.org/datatype3.html

The sqlar format supports this, it would be useful if insert-files could support this too.

https://www.sqlite.org/sqlar.html