duplicity-team team mailing list archive

[Matthew Glazar <strager.nds+launchpad@xxxxxxxxx>]

> Let's please test it against lp:duplicity (0.8-series) and see if we can replicate the gains without breakage.

I rebased this change onto 0.8-series. (lp:~stragerneds/duplicity/duplicity-0.7-perf still has the 0.7-series version, in case anyone wants it.)

I'm unable to verify that this causes no regressions. For 0.8, the test suite fails for me on my macOS machine, and on my Linux machine with duplicity's Docker container. =\

> why not take the more unobstructive path and cache the result in file_naming.parse ? [...]
> of course this way more memory will be used depending on the number of cached file entries.

The optimization is already done for CollectionsStatus.get_signature_chains and SignatureChain.add_filename. My patch replicates the pattern for CollectionsStatus.get_backup_chains and BackupSet.add_filename.

(Also, I avoid memoization like you suggest. Memoization has many costs, like the one you pointed out.)
-- 
https://code.launchpad.net/~stragerneds/duplicity/duplicity/+merge/369789
Your team duplicity-team is subscribed to branch lp:duplicity.