issue_comments

Prototype so far: diff diff --git a/sqlite_utils/cli.py b/sqlite_utils/cli.py index 767b170..d96c507 100644 --- a/sqlite_utils/cli.py +++ b/sqlite_utils/cli.py @@ -1762,6 +1762,17 @@ def query( is_flag=True, help="Analyze resulting tables and output results", ) +@click.option("--key", help="read data from this key of the root object") +@click.option( + "--auto-key", + is_flag=True, + help="Find a key in the root object that is a list of objects", +) +@click.option( + "--analyze", + is_flag=True, + help="Analyze resulting tables and output results", +) @load_extension_option def memory( paths, @@ -1784,6 +1795,8 @@ def memory( schema, dump, save, + key, + auto_key, analyze, load_extension, ): @@ -1838,7 +1851,9 @@ def memory( csv_table = stem stem_counts[stem] = stem_counts.get(stem, 1) + 1 csv_fp = csv_path.open("rb") - rows, format_used = rows_from_file(csv_fp, format=format, encoding=encoding) + rows, format_used = rows_from_file( + csv_fp, format=format, encoding=encoding, key=key, auto_key=auto_key + ) tracker = None if format_used in (Format.CSV, Format.TSV) and not no_detect_types: tracker = TypeTracker() diff --git a/sqlite_utils/utils.py b/sqlite_utils/utils.py index 8754554..2e69c26 100644 --- a/sqlite_utils/utils.py +++ b/sqlite_utils/utils.py @@ -231,6 +231,8 @@ def rows_from_file( encoding: Optional[str] = None, ignore_extras: Optional[bool] = False, extras_key: Optional[str] = None, + key: Optional[str] = None, + auto_key: Optional[bool] = False, ) -> Tuple[Iterable[dict], Format]: """ Load a sequence of dictionaries from a file-like object containing one of four different formats. @@ -271,13 +273,31 @@ def rows_from_file( :param encoding: the character encoding to use when reading CSV/TSV data :param ignore_extras: ignore any extra fields on rows :param extras_key: put any extra fields in a list with this key + :param key: read data from this key of the root object + :param auto_key: find a key in the root object that is a list of objects """ if ignore_extras and extras_key: raise ValueError("Cannot use ignore_extras= and extras_key= together") + if key and auto_key: + raise ValueError("Cannot use key= and auto_key= together") if format == Format.JSON: decoded = json.load(fp) if isinstance(decoded, dict): - decoded = [decoded] + if auto_key: + list_keys = [ + k + for k in decoded + if isinstance(decoded[k], list) + and decoded[k] + and all(isinstance(o, dict) for o in decoded[k]) + ] + if len(list_keys) == 1: + decoded = decoded[list_keys[0]] + elif key: + # Raises KeyError, I think that's OK + decoded = decoded[key] + if not isinstance(decoded, list): + decoded = [decoded] if not isinstance(decoded, list): raise RowsFromFileBadJSON("JSON must be a list or a dictionary") return decoded, Format.JSON @@ -305,7 +325,9 @@ def rows_from_file( first_bytes = buffered.peek(2048).strip() if first_bytes.startswith(b"[") or first_bytes.startswith(b"{"): # TODO: Detect newline-JSON - return rows_from_file(buffered, format=Format.JSON) + return rows_from_file( + buffered, format=Format.JSON, key=key, auto_key=auto_key + ) else: dialect = csv.Sniffer().sniff( first_bytes.decode(encoding or "utf-8-sig", "ignore")

I had a big CSV file lying around, I converted it to other formats like this:

sqlite-utils insert /tmp/t.db t /tmp/en.openfoodfacts.org.products.csv --csv
sqlite-utils rows /tmp/t.db t --nl > /tmp/big.nl
sqlite-utils rows /tmp/t.db t > /tmp/big.json

Then tested the progress bar like this:

sqlite-utils insert /tmp/t2.db t /tmp/big.nl --nl

Output:

sqlite-utils insert /tmp/t2.db t /tmp/big.nl --nl [------------------------------------] 0% [#######-----------------------------] 20% 00:00:20 With --silent it is silent.

And for regular JSON:

sqlite-utils insert /tmp/t3.db t /tmp/big.json [####################################] 100% This is actually not doing the right thing. The problem is that sqlite-utils doesn't include a streaming JSON parser, so it instead reads that entire JSON file into memory first (exhausting the progress bar to 100% instantly) and then does the rest of the work in-memory while the bar sticks at 100%.

A note from PR #486: https://github.com/simonw/sqlite-utils/issues/486#issuecomment-1248591268_

sqlite-utils insert /tmp/t3.db t /tmp/big.json [####################################] 100% This is actually not doing the right thing. The problem is that sqlite-utils doesn't include a streaming JSON parser, so it instead reads that entire JSON file into memory first (exhausting the progress bar to 100% instantly) and then does the rest of the work in-memory while the bar sticks at 100%.

I decided to land this anyway. If a streaming JSON parser is added later it will start to work.

I'm going to land this anyway. The lack of a streaming JSON parser is a separate issue, I don't think it should block landing this improvement.

This looks good to me. I need to run some manual tests before merging (it's a good sign that the automated tests pass though).

https://github.com/actions/setup-python/blob/main/docs/advanced-usage.md#using-the-python-version-input says can set the full version:

- uses: actions/setup-python@v4 with: python-version: "3.10.6" I'll try that.

There's a fix for mypy that has landed but isn't out in a release yet: - https://github.com/python/mypy/issues/13385

For the moment looks like pinning to Python 3.10.6 could help. Need to figure out how to do that in GitHub Actions though.

Annoying mypy test failure:

/Users/runner/hostedtoolcache/Python/3.10.7/x64/lib/python3.10/site-packages/numpy/__init__.pyi:636: error: Positional-only parameters are only supported in Python 3.8 and greater Looks like this: - https://github.com/python/mypy/issues/13627

I think refactoring sqlite-utils insert to use rows_from_file needs to happen as part of this work.

Why doesn't sqlite-utils insert use the rows_from_file function I wonder?

https://github.com/simonw/sqlite-utils/issues/279#issuecomment-864207841 says:

I can refactor sqlite-utils insert to use this new code too.

Maybe I forgot to do that?

This feels pretty good: % sqlite-utils memory ~/Downloads/CVR_Export_20220908084311/*.json --schema --auto-key CREATE TABLE [BallotTypeContestManifest] ( [BallotTypeId] INTEGER, [ContestId] INTEGER ); CREATE VIEW t1 AS select * from [BallotTypeContestManifest]; CREATE VIEW t AS select * from [BallotTypeContestManifest]; CREATE TABLE [BallotTypeManifest] ( [Description] TEXT, [Id] INTEGER, [ExternalId] TEXT );

Actually I quite like --key X - it could work for single nested objects too. You could insert a single record like this:

json { "record" { "id": 1 } } sqlite-utils insert db.db records record.json --key record

Maybe this would make more sense as a mechanism where you can say "Use the data in the key called X" - but there's a special option for "figure out that key automatically".

The syntax then could be:

--list-key List

Or for automatic detection:

--list-key-auto

Could also go with --key List and --key-auto - but would that be as obvious as --list-key?

The most similar option I have at the moment is probably --flatten. What would good names for this option be?

• --auto-list
• --auto-key
• --inner-key
• --auto-json
• --find-list
• --find-key

Those are all bad.

Another option: introduce a new explicit format for it. Right now the explicit formats you can use are:

https://github.com/simonw/sqlite-utils/blob/d9b9e075f07a20f1137cd2e34ed5d3f1a3db4ad8/docs/cli-reference.rst#L153-L158

So I could add a :autojson format.

Here's how I used my prototype to build that Gist:

sqlite-utils memory ~/Downloads/CVR_Export_20220908084311/*.json --schema > database.sql

Built a prototype of this that works really well: diff diff --git a/sqlite_utils/utils.py b/sqlite_utils/utils.py index c0b7bf1..f9a482c 100644 --- a/sqlite_utils/utils.py +++ b/sqlite_utils/utils.py @@ -272,7 +272,19 @@ def rows_from_file( if format == Format.JSON: decoded = json.load(fp) if isinstance(decoded, dict): - decoded = [decoded] + # TODO: Solve for if this isn't what people want + # Does it have just one key that is a list of dicts? + list_keys = [ + k + for k in decoded + if isinstance(decoded[k], list) + and decoded[k] + and all(isinstance(o, dict) for o in decoded[k]) + ] + if len(list_keys) == 1: + decoded = decoded[list_keys[0]] + else: + decoded = [decoded] if not isinstance(decoded, list): raise RowsFromFileBadJSON("JSON must be a list or a dictionary") return decoded, Format.JSON I used that to build this: https://gist.github.com/simonw/0e6901974a14ab7d56c2746a04d72c8c

One problem though: right now, if you do this sqlite-utils treats it as a single object and adds a tags column with JSON in it: echo '{"title": "Hi", "tags": [{"t": "one"}]}` | sqlite-utils insert db.db t - If I implement this new mechanism the above line would behave differently - which would be a backwards incompatible change.

So I probably need some kind of opt-in mechanism for this. And I need a good name for it.

I could prototype this idea as a datasette-featured-tables plugin that delivers its own custom index.html template.

Idea: allow the user to specify one or more featured tables. Each table is then shown as a summary on the homepage - with the total number of rows and the first 5 rows. If the table has search configured there's a search box too.

If the instance has only one database with only one table (excluding hidden tables) it gets featured automatically perhaps (maybe with a way to opt-out of that if you want to).

Twitter conversation that inspired this issue: https://twitter.com/psychemedia/status/1570410108785684481