A python list implementation that uses the disk to handle very large collections of pickle-able objects.

Now as a PyPi package: pip install disklist!

DiskList will create a unamed temporary file on disk and store your objects in it. The most commonly used list methods and operators are implemented so you can use it as an almost drop-in replacement.

Here's an example!

from disklist import DiskList

dl = DiskList()

for i in range(0, 1000):
    dl.append(i)

dl[0] # Boring indexing works!
dl[0:2] # So does slicing!

# What if you want to insert something?

dl.insert(0, 2)

# ...or setting an element

dl[0] = 2

# Concatenation is functional too

dl2 = DiskList()
dl2.append(4)

dl = dl + dl2

# Why not iterate through it?

for item in dl:
    print(dl) # 1, 2, 3, 4, 5, 6... You get the point

Anything that you can do with a list that is impossible with a DiskList deserves an issue!

Mostly for speed concerns, setting an index to something doesn't clean the old object in the TemporaryFile so while the "list" will dereference it, the space on your disk will still be used. While this will not have any real impacts in most usecases, avoid doing stuff like this:

for i in range(1000000):
    dl[0] = i

Because while dl[0] will be equal to 999999, you effectively created 7.868 MB (yes I did the math) of useless data on your disk.

As with anything that uses the disk, expect every action to 2+ order of magnitude slower than a regular list. Here are some benchmarks:

|---------- Instanciation ----------|
List: 0.0000000444 sec
DiskList: 0.0000147268 sec
|---------- Instanciation with big object ----------|
List: 0.0000000341 sec
DiskList: 0.0000169193 sec
|---------- Appending ----------|
List: 0.0000000967 sec
DiskList: 0.0000034166 sec
|---------- Appending big object ----------|
List: 0.0000000715 sec
DiskList: 0.0000214165 sec
|---------- Inserting ----------|
List: 0.0000005487 sec
DiskList: 0.0000044418 sec
|---------- Inserting big object ----------|
List: 0.0000004602 sec
DiskList: 0.0000218233 sec
|------ Getting with index ------|
List: 0.0000000430 sec
DiskList: 0.0000030653 sec
|------ Setting with index ------|
List: 0.0000000898 sec
DiskList: 0.0000034653 sec
|---------- Iterating ----------|
List: 0.0000154686 sec
DiskList without cache: 0.0028857501 sec
DiskList with cache: 0.0013473100 sec
|---------- Iterating with big object ----------|
List: 0.0000151939 sec
DiskList without cache: 0.0312750774 sec
DiskList with cache: 0.0160398417 sec

Full disclosure, these tests were done using an SSD

Excellent question! My honest answer would be "as rarely as possible". It is slow, not scalable (like a database would be for example). My only legitimate use right now is for machine learning to store my training batches until I need them. If you feel like your usecase is legitimate send me an email at [email protected]. I am very curious!