Clean up context ID generation. #30
Conversation
Using a sha256 hash is major overkill for the purposes of finding a short hash of a string. This change switches _short_hash to use Python's built-in hash() function (used for hashing in dicts and sets). which works just fine for this, and is much faster. In addition, since context IDs are length-limited, use base-85 encoding to pack as many bits of the hash and index into the context as possible.
|
Hey @basnijholt, could you please take a look at this? |
|
Hi @blueshiftlabs, thanks a lot for taking the time! I am still thinking about whether I want to change the numbering to base85. The numbering has helped me a lot when debugging, in base85 it's not really readable anymore. |
|
I can limit the base85 to just the entity ID hash without affecting anything. My real concern there had to do with repeating context IDs, but since the counter gets reset to 0 at every HA restart, 10 million should be enough - even at one call per second, that's still 115 days of HA uptime before we run out of IDs. One note - since context IDs are persisted to the database, it might be worth saving the counter value across Home Assistant restarts to avoid collisions on every restart. Not sure if it matters in this case, but if it does, base85 might be a lot more useful. I could also do base36 or something similar that might be a bit more human-readable, in exchange for giving up some keyspace. |
basnijholt
force-pushed
the
master
branch
7 times, most recently
from
191055d
to
8a97480
Compare
|
Thanks a lot @blueshiftlabs! This has taken me 2 years to merge, but now that we have tests, it makes me more confident that it works. |
Using a SHA-256 cryptographic hash is major overkill for the purposes of finding a short representation of a string. This change switches
_short_hashto use Python's built-inhash()function (used for hashing in dicts and sets), which works just fine for this, and is much faster.In addition, since context IDs are length-limited, use base-85 encoding to pack as many bits of the hash and index into the context as possible. This slightly changes the context format, but the new format should be a bit easier to read.