C port of SD, a very efficient flash cards app
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
Anthony Wang cd9b330a1f
Fix #2: Make sure 0-weight cards are never selected
2 weeks ago
LICENSE Initial commit 8 months ago
README.md Rename to SDC 1 month ago
sd.c Code cleanup 1 month ago
segmenttree.c Fix #2: Make sure 0-weight cards are never selected 2 weeks ago
segmenttree.h Finish porting SD code 7 months ago


This file contains ambiguous Unicode characters!

This file contains ambiguous Unicode characters that may be confused with others in your current locale. If your use case is intentional and legitimate, you can safely ignore this warning. Use the Escape button to highlight these characters.


C port of SD, a very efficient flash cards app


Flash cards are stored in the cards table of a SQLite database. There are four columns: idx INTEGER PRIMARY KEY, weight INTEGER, key STRING, val STRING. The idx is a unique index for each card, starting at 0. The weight is how often the card should come up. The key and value are the front and reverse sides of the card. You can use the sqlite3 CLI to create a card deck.

Now build this project with gcc sd.c segmenttree.c -o sd -lsqlite3 -O2 -march=native and run ./sd to enjoy a fast flash cards experience! The program will display the key of a randomly selected card. Press any key to show the val of the card. Now press either y or n depending on whether you got the card correct, and the program adjusts that card's weight.

If you're wondering where the name came from, this is the C port of SD, which was named after a common type of flash card. 😀


SD is designed to be extremely efficient in order to support a very large number of flash cards and should be able to handle millions of cards with ease. If N is the number of cards, initializing the program requires O(N) time and O(N) memory. Selecting a random card and adjusting its weight requires O(log N) time. Internally SD uses segment trees to achieve this time complexity.

Some benchmark results using 10 card updates:

C version: ./sd < test
  Time (mean ± σ):      57.4 ms ±   4.5 ms    [User: 6.9 ms, System: 2.4 ms]
  Range (min … max):    51.9 ms …  70.9 ms    41 runs
Go version: ./sd < test
  Time (mean ± σ):      92.7 ms ±   6.8 ms    [User: 8.4 ms, System: 4.5 ms]
  Range (min … max):    79.4 ms … 108.9 ms    33 runs

The C port is about 30% faster than the original Go code, and I'm currently working on further performance improvements using delayed database writes.