Tools for analyzing Git history using SQLite
For background on this project see git-history: a tool for analyzing scraped data collected using Git and SQLite.
Measuring traffic during the Half Moon Bay Pumpkin Festival describes a project using this tool in detail.
Install this tool using pip
:
$ pip install git-history
git-history-demos.datasette.io hosts three example databases created using this tool:
- pge-outages shows a history of PG&E (the electricity supplier) outages, using data collected in simonw/pge-outages converted using pge-outages.sh
- ca-fires shows a history of fires in California reported on fire.ca.gov/incidents, from data in simonw/ca-fires-history converted using ca-fires.sh
- sf-bay-511 has records of San Francisco Bay Area traffic and transit incident data from 511.org, collected in dbreunig/511-events-history converted using sf-bay-511.sh
The demos are deployed using Datasette on Google Cloud Run by this GitHub Actions workflow.
This tool can be run against a Git repository that holds a file that contains JSON, CSV/TSV or some other format and which has multiple versions tracked in the Git history. Read Git scraping: track changes over time by scraping to a Git repository to understand how you might create such a repository.
The file
command analyzes the history of an individual file within the repository, and generates a SQLite database table that represents the different versions of that file over time.
The file is assumed to contain multiple objects - for example, the results of scraping an electricity outage map or a CSV file full of records.
Assuming you have a file called incidents.json
that is a JSON array of objects, with multiple versions of that file recorded in a repository. Each version of that file might look something like this:
[
{
"IncidentID": "abc123",
"Location": "Corner of 4th and Vermont",
"Type": "fire"
},
{
"IncidentID": "cde448",
"Location": "555 West Example Drive",
"Type": "medical"
}
]
Change directory into the GitHub repository in question and run the following:
git-history file incidents.db incidents.json
This will create a new SQLite database in the incidents.db
file with three tables:
commits
containing a row for every commit, with ahash
column, thecommit_at
date and a foreign key to anamespace
.item
containing a row for every item in every version of thefilename.json
file - with an extra_commit
column that is a foreign key back to thecommit
table.namespaces
containing a single row. This allows you to build multiple tables for different files, using the--namespace
option described below.
The database schema for this example will look like this:
CREATE TABLE [namespaces] (
[id] INTEGER PRIMARY KEY,
[name] TEXT
);
CREATE UNIQUE INDEX [idx_namespaces_name]
ON [namespaces] ([name]);
CREATE TABLE [commits] (
[id] INTEGER PRIMARY KEY,
[namespace] INTEGER REFERENCES [namespaces]([id]),
[hash] TEXT,
[commit_at] TEXT
);
CREATE UNIQUE INDEX [idx_commits_namespace_hash]
ON [commits] ([namespace], [hash]);
CREATE TABLE [item] (
[IncidentID] TEXT,
[Location] TEXT,
[Type] TEXT,
[_commit] INTEGER REFERENCES [commits]([id])
);
If you have 10 historic versions of the incidents.json
file and each one contains 30 incidents, you will end up with 10 * 30 = 300 rows in your item
table.
If your objects have a unique identifier - or multiple columns that together form a unique identifier - you can use the --id
option to de-duplicate and track changes to each of those items over time.
This provides a much more interesting way to apply this tool.
If there is a unique identifier column called IncidentID
you could run the following:
git-history file incidents.db incidents.json --id IncidentID
The database schema used here is very different from the one used without the --id
option.
If you have already imported history, the command will skip any commits that it has seen already and just process new ones. This means that even though an initial import could be slow subsequent imports should run a lot faster.
This command will create six tables - commits
, item
, item_version
, columns
, item_changed
and namespaces
.
Here's the full schema:
CREATE TABLE [namespaces] (
[id] INTEGER PRIMARY KEY,
[name] TEXT
);
CREATE UNIQUE INDEX [idx_namespaces_name]
ON [namespaces] ([name]);
CREATE TABLE [commits] (
[id] INTEGER PRIMARY KEY,
[namespace] INTEGER REFERENCES [namespaces]([id]),
[hash] TEXT,
[commit_at] TEXT
);
CREATE UNIQUE INDEX [idx_commits_namespace_hash]
ON [commits] ([namespace], [hash]);
CREATE TABLE [item] (
[_id] INTEGER PRIMARY KEY,
[_item_id] TEXT
, [IncidentID] TEXT, [Location] TEXT, [Type] TEXT, [_commit] INTEGER);
CREATE UNIQUE INDEX [idx_item__item_id]
ON [item] ([_item_id]);
CREATE TABLE [item_version] (
[_id] INTEGER PRIMARY KEY,
[_item] INTEGER REFERENCES [item]([_id]),
[_version] INTEGER,
[_commit] INTEGER REFERENCES [commits]([id]),
[IncidentID] TEXT,
[Location] TEXT,
[Type] TEXT,
[_item_full_hash] TEXT
);
CREATE TABLE [columns] (
[id] INTEGER PRIMARY KEY,
[namespace] INTEGER REFERENCES [namespaces]([id]),
[name] TEXT
);
CREATE UNIQUE INDEX [idx_columns_namespace_name]
ON [columns] ([namespace], [name]);
CREATE TABLE [item_changed] (
[item_version] INTEGER REFERENCES [item_version]([_id]),
[column] INTEGER REFERENCES [columns]([id]),
PRIMARY KEY ([item_version], [column])
);
CREATE VIEW item_version_detail AS select
commits.commit_at as _commit_at,
commits.hash as _commit_hash,
item_version.*,
(
select json_group_array(name) from columns
where id in (
select column from item_changed
where item_version = item_version._id
)
) as _changed_columns
from item_version
join commits on commits.id = item_version._commit;
CREATE INDEX [idx_item_version__item]
ON [item_version] ([_item]);
The item
table will contain the most recent version of each row, de-duplicated by ID, plus the following additional columns:
_id
- a numeric integer primary key, used as a foreign key from theitem_version
table._item_id
- a hash of the values of the columns specified using the--id
option to the command. This is used for de-duplication when processing new versions._commit
- a foreign key to thecommit
table, representing the most recent commit to modify this item.
The item_version
table will contain a row for each captured differing version of that item, plus the following columns:
_id
- a numeric ID for the item version record._item
- a foreign key to theitem
table._version
- the numeric version number, starting at 1 and incrementing for each captured version._commit
- a foreign key to thecommit
table._item_full_hash
- a hash of this version of the item. This is used internally by the tool to identify items that have changed between commits.
The other columns in this table represent columns in the original data that have changed since the previous version. If the value has not changed, it will be represented by a null
.
If a value was previously set but has been changed back to null
it will still be represented as null
in the item_version
row. You can identify these using the item_changed
many-to-many table described below.
You can use the --full-versions
option to store full copies of the item at each version, rather than just storing the columns that have changed.
This SQL view joins item_version
against commits
to add three further columns: _commit_at
with the date of the commit, and _commit_hash
with the Git commit hash.
This many-to-many table indicates exactly which columns were changed in an item_version
.
item_version
is a foreign key to a row in theitem_version
table.column
is a foreign key to a row in thecolumns
table.
This table with have the largest number of rows, which is why it stores just two integers in order to save space.
The columns
table stores column names. It is referenced by item_changed
.
id
- an integer ID.name
- the name of the column.namespace
- a foreign key tonamespaces
, for if multiple file histories are sharing the same database.
Note that _id
, _item_full_hash
, _item
, _item_id
, _version
, _commit
, _item_id
, _commit_at
, _commit_hash
, _changed_columns
, rowid
are considered reserved column names for the purposes of this tool.
If your data contains any of these they will be renamed to add a trailing underscore, for example _id_
, _item_
, _version_
, to avoid clashing with the reserved columns.
If you have a column with a name such as _commit_
it will be renamed too, adding an additional trailing underscore, so _commit_
becomes _commit__
and _commit__
becomes _commit___
.
--repo DIRECTORY
- the path to the Git repository, if it is not the current working directory.--branch TEXT
- the Git branch to analyze - defaults tomain
.--id TEXT
- as described above: pass one or more columns that uniquely identify a record, so that changes to that record can be calculated over time.--full-versions
- instead of recording just the columns that have changed in theitem_version
table record a full copy of each version of theh item.--ignore TEXT
- one or more columns to ignore - they will not be included in the resulting database.--csv
- treat the data is CSV or TSV rather than JSON, and attempt to guess the correct dialect--dialect
- use a spcific CSV dialect. Options areexcel
,excel-tab
andunix
- see the Python CSV documentation for details.--skip TEXT
- one or more full Git commit hashes that should be skipped. You can use this if some of the data in your revision history is corrupted in a way that prevents this tool from working.--start-at TEXT
- skip commits prior to the specified commit hash.--start-after TEXT
- skip commits up to and including the specified commit hash, then start processing from the following commit.--convert TEXT
- custom Python code for a conversion, described below.--import TEXT
- additional Python modules to import for--convert
.--ignore-duplicate-ids
- if a single version of a file has the same ID in it more than once, the tool will exit with an error. Use this option to ignore this and instead pick just the first of the two duplicates.--namespace TEXT
- use this if you wish to include the history of multiple different files in the same database. The default isitem
but you can set it to something else, which will produce tables with names likeyournamespace
andyournamespace_version
.--wal
- Enable WAL mode on the created database file. Use this if you plan to run queries against the database whilegit-history
is creating it.--silent
- don't show the progress bar.
If the data in your repository is a CSV or TSV file you can process it by adding the --csv
option. This will attempt to detect which delimiter is used by the file, so the same option works for both comma- and tab-separated values.
git-history file trees.db trees.csv --id TreeID
You can also specify the CSV dialect using the --dialect
option.
If your data is not already either CSV/TSV or a flat JSON array, you can reshape it using the --convert
option.
The format needed by this tool is an array of dictionaries, as demonstrated by the incidents.json
example above.
If your data does not fit this shape, you can provide a snippet of Python code to converts the on-disk content of each stored file into a Python list of dictionaries.
For example, if your stored files each look like this:
{
"incidents": [
{
"id": "552",
"name": "Hawthorne Fire",
"engines": 3
},
{
"id": "556",
"name": "Merlin Fire",
"engines": 1
}
]
}
You could use the following Python snippet to convert them to the required format:
json.loads(content)["incidents"]
(The json
module is exposed to your custom function by default.)
You would then run the tool like this:
git-history file database.db incidents.json \
--id id \
--convert 'json.loads(content)["incidents"]'
The content
variable is always a bytes
object representing the content of the file at a specific moment in the repository's history.
You can import additional modules using --import
. This example shows how you could read a CSV file that uses ;
as the delimiter:
git-history file trees.db ../sf-tree-history/Street_Tree_List.csv \
--repo ../sf-tree-history \
--import csv \
--import io \
--convert '
fp = io.StringIO(content.decode("utf-8"))
return list(csv.DictReader(fp, delimiter=";"))
' \
--id TreeID
You can import nested modules such as ElementTree using --import xml.etree.ElementTree
, then refer to them in your function body as xml.etree.ElementTree
. For example, if your tracked data was in an items.xml
file that looked like this:
<items>
<item id="1" name="One" />
<item id="2" name="Two" />
<item id="3" name="Three" />
</item>
You could load it using the following --convert
script:
git-history file items.xml --convert '
tree = xml.etree.ElementTree.fromstring(content)
return [el.attrib for el in tree.iter("item")]
' --import xml.etree.ElementTree --id id
If your Python code spans more than one line it needs to include a return
statement.
You can also use Python generators in your --convert
code, for example:
git-history file stats.db package-stats/stats.json \
--repo package-stats \
--convert '
data = json.loads(content)
for key, counts in data.items():
for date, count in counts.items():
yield {
"package": key,
"date": date,
"count": count
}
' --id package --id date
This conversion function expects data that looks like this:
{
"airtable-export": {
"2021-05-18": 66,
"2021-05-19": 60,
"2021-05-20": 87
}
}
To contribute to this tool, first checkout the code. Then create a new virtual environment:
cd git-history
python -m venv venv
source venv/bin/activate
Or if you are using pipenv
:
pipenv shell
Now install the dependencies and test dependencies:
pip install -e '.[test]'
To run the tests:
pytest
To update the schema examples in this README file:
cog -r README.md