- User Since
- Dec 6 2017, 9:25 AM (15 w, 2 d)
Sun, Mar 11
Sat, Mar 10
Fri, Mar 9
Tue, Feb 27
Mon, Feb 26
Feb 11 2018
There was a longish discussion on IRC about this. The problem is that the flags by themselves do not document the interaction of -r and -b or multiple instances of either at all. The change was a compromise trying to address that.
Feb 7 2018
Feb 4 2018
Feb 3 2018
Feb 1 2018
Jan 28 2018
Jan 27 2018
Jan 26 2018
I haven't seen a performance difference for regular runs, but it does show up in profiles, so eliminating it avoids wasting mental power by developers down the line.
No, the change to yield a bool first is not strictly necessary for the rest of the logic. But it cheaply kills the need for the rewindhunk generator, i.e. it cuts a layer of indirection cheaply.
Jan 25 2018
I don't completely trust my numbers here, but it seems to give at least 3-4% for diffing netbsd-7 and netbsd-8.
Jan 19 2018
Jan 18 2018
Jan 17 2018
It was preparation for more tests here. Since I am moving them to a separate file now, I'm not really attached to it at all.
Jan 16 2018
Jan 15 2018
Merging into the parent review.
Jan 14 2018
I wish we could find a way to send multiple, inline, pre-generated bundles in one response. However, the existing design of compression within the bundle2 format doesn't easily facilitate this. We should think long and hard about whether to implement this feature as partial pull or extend bundle2 to allow us to do nice things and avoid the extra client-server roundtrips.
For my test case, which is a bundle of all changes in the NetBSD repo before 2014 and a yearly bundle afterwards until 2018/1/1 and normal pull for the rest, find_pullbundle needs less than 0.5s of CPU time in this iteration when it matches.
After the initial clone and with additional available changes, pull time is:
Jan 13 2018
With the last update, the obs issues are resolved.
Which status code shall we use then, just plain 400?
Jan 12 2018
This is a proof-of-concept. There is an interaction with obsoletion that I am still trying to understand (and fix).
Dec 11 2017
The bitset version has shown already that optimizing this is worthwhile and can eliminate up to 10% of the total size of the transaction object.
Dec 10 2017
This version should be committable. It introduces the necessary API for isolating further changes and gives a small improvement already. I'll be looking at provider a memory and time efficient sparce vector next, that's where the real benefits will be.
Dec 9 2017
Like I said, this is primarily a proof of concept. I'm still playing with different approaches and I don't think intbitset is a good fit for other reasons, but it worked the best from the various pre-existing implementations.
Dec 8 2017
This is a Proof-of-Concept and with a follow-up in D1622 to use a C implementation for the bitset.