issue_comments: 643685669

I tried this experiment: python import sqlite3, psutil def show_things(): conn = sqlite3.connect("fixtures.db") tables = [r[0] for r in conn.execute("select * from sqlite_master").fetchall()] return tables print(psutil.Process().open_files()) print(show_things()) print(psutil.Process().open_files()) To see if the connection would be automatically released when the conn variable was garbage collected at the end of the function... and it was correctly released - the two calls to open_files() showed that the file did not remain open.

Likewise: ``` In [11]: conn = sqlite3.connect("fixtures.db")

In [12]: psutil.Process().open_files()
Out[12]: [popenfile(path='/Users/simon/.ipython/profile_default/history.sqlite', fd=4), popenfile(path='/Users/simon/.ipython/profile_default/history.sqlite', fd=5), popenfile(path='/Users/simon/Dropbox/Development/datasette/fixtures.db', fd=12)]

In [13]: del conn

In [14]: psutil.Process().open_files()
Out[14]: [popenfile(path='/Users/simon/.ipython/profile_default/history.sqlite', fd=4), popenfile(path='/Users/simon/.ipython/profile_default/history.sqlite', fd=5)] `` So presumably there's something about the way my pytest fixtures work that's causing the many differentDatasette()` instances and their underlying SQLite connections that I create not to be cleaned up later.