diff options
| -rw-r--r-- | Command/Unused.hs | 7 | ||||
| -rw-r--r-- | Utility/Bloom.hs | 4 | ||||
| -rw-r--r-- | debian/changelog | 2 | ||||
| -rw-r--r-- | doc/design/roadmap.mdwn | 4 | ||||
| -rw-r--r-- | doc/devblog/day_219__catching_up_and_looking_back.mdwn | 109 |
5 files changed, 121 insertions, 5 deletions
diff --git a/Command/Unused.hs b/Command/Unused.hs index bf98e53bc..c2179447d 100644 --- a/Command/Unused.hs +++ b/Command/Unused.hs @@ -189,7 +189,12 @@ bloomBitsHashes :: Annex (Int, Int) bloomBitsHashes = do capacity <- bloomCapacity accuracy <- bloomAccuracy - return $ suggestSizing capacity (1/ fromIntegral accuracy) + case safeSuggestSizing capacity (1 / fromIntegral accuracy) of + Left e -> do + warning $ "bloomfilter " ++ e ++ "; falling back to sane value" + -- precaulculated value for 500000 (1/1000) + return (8388608,10) + Right v -> return v {- Creates a bloom filter, and runs an action, such as withKeysReferenced, - to populate it. diff --git a/Utility/Bloom.hs b/Utility/Bloom.hs index e3000de3d..b0988a8c4 100644 --- a/Utility/Bloom.hs +++ b/Utility/Bloom.hs @@ -9,7 +9,7 @@ module Utility.Bloom ( Bloom, - suggestSizing, + safeSuggestSizing, Hashable, cheapHashes, notElemB, @@ -25,7 +25,7 @@ import qualified Data.BloomFilter as Bloom #else import qualified Data.BloomFilter as Bloom #endif -import Data.BloomFilter.Easy (suggestSizing, Bloom) +import Data.BloomFilter.Easy (safeSuggestSizing, Bloom) import Data.BloomFilter.Hash (Hashable, cheapHashes) import Control.Monad.ST.Safe (ST) diff --git a/debian/changelog b/debian/changelog index 710f72a34..c2e0ea508 100644 --- a/debian/changelog +++ b/debian/changelog @@ -20,6 +20,8 @@ git-annex (5.20140832) UNRELEASED; urgency=medium the repository, rather than just the file's base name. Note that if you're relying on such things to keep files separate with WORM, you should really be using a better backend. + * Rather than crashing when there's a problem with the requested bloomfilter + capacity/accuracy, fall back to a reasonable default bloom filter size. -- Joey Hess <joeyh@debian.org> Thu, 04 Sep 2014 16:17:22 -0400 diff --git a/doc/design/roadmap.mdwn b/doc/design/roadmap.mdwn index 7a3fa06fe..b6af4db30 100644 --- a/doc/design/roadmap.mdwn +++ b/doc/design/roadmap.mdwn @@ -1,6 +1,6 @@ ## roadmap -Now in the +Just finished the [sustaining git-annex development](https://campaign.joeyh.name/) year (starting September 2013). @@ -15,7 +15,7 @@ Now in the * Month 9 Brazil!, [[!traillink assistant/sshpassword]] * Month 10 polish [[assistant/Windows]] port * Month 11 [[!traillink assistant/chunks]] -* **Month 12** user-driven features and polishing +* Month 12 user-driven features and polishing Deferred until later: diff --git a/doc/devblog/day_219__catching_up_and_looking_back.mdwn b/doc/devblog/day_219__catching_up_and_looking_back.mdwn new file mode 100644 index 000000000..8d82d8cb9 --- /dev/null +++ b/doc/devblog/day_219__catching_up_and_looking_back.mdwn @@ -0,0 +1,109 @@ +Yesterday and today were the first good solid days working on git-annex in a +while. There's a big backlog, currently of 133 messages, so I have been +concentrating on bug reports first. Happily, not many new bugs have been +reported lately, and I've made good progress on them, fixing 5 bugs today, +including a file descriptor leak. + +## catching up + +In this end of summer rush, I've been too busy to blog for the past 20 days, +but not entirely too busy to work on git-annex. Two releases have been made +in that time, and a fair amount of improvements worked on. + +Including a new feature: When a local git repository is cloned with `git +clone --shared`, git-annex detects this and defaults to a special mode +where file contents get hard linked into the clone. It also makes the cloned +repository be untrusted, to avoid confusing numcopies counting with the +hard links. This can be useful for temporary working repositories without +the overhead of lots of copies of files. + +## looking back + +I want to look back further, over the crowdfunded year of work covered +by this devblog. There were a lot of things I wanted to accomplish this +past year, and I managed to get to most of them. As well as a few surprises. + +* Windows support improved more than I guessed in my wildest dreams. + git-annex went from working not too well on the command line to + being pretty solid there, as well as having a working + and almost polished webapp on Windows. + There are still warts -- it's Windows after all! + +* Android didn't get many improvements. Most of the time I had budgeted to + Android porting ended up being used on Windows porting instead. I did, + however, get the Android build environment cleaned up a lot from the initial + hacked together one, and generally kept it building and working on Android. + +* The [direct mode guard](http://git-annex.branchable.com/devblog/day_48__direct_mode_guard_design/) + was not planned, but the need for it became clear, and + it's dramatically reduced the amount of command-line foot-shooting + that goes on in direct mode. + +* Repository repair was planned, and I've very proud of [git-repair](http://git-repair.branchable.com/). + Also pleased with the webapp's UI for scheduling repository consistency + checks. + Always room for improvement in this kind of thing, but this brings a new + capability to both git and git-annex. + +* The [[external_special_remote_interface|special_remotes/external]] came + together beautifully. External special remotes are now just as well + supported as built-in ones, except the webapp cannot be used to configure + them. + +* Using git-remote-gcrypt for fully encrypted git repositories, including + support in the webapp for setting them (and gpg keys if necessary), + happened. Still needs testing/more use/improvements. Avoided doing + much in the area of gpg key management, which is probably good to avoid when + possible, but is probably needed to make this a really suitable option for + end users. + +* Telehash is still being built, and it's not clear if they've gotten it + to work at all yet. The v2 telehash has recently been superseded by a + a new v3. So I am not pleased that I didn't get git-annex working with + telehash, but it was outside my control. This is a problem that needs to get + solved outside git-annex first, either by telehash or something else. + The plan is to keep an eye on everything in this space, including for example, + Maidsafe. + +* In the meantime, the new notifychanges support in git-annex-shell + makes XMPP/telehash/whatever unnecessary in a lot of configurations. + git-annex's remotedaemon architecture supports that and is designed + to support other notification methods later. And the webapp has a lot of + improvements in the area of setting up ssh remotes, so fewer users will + be stuck with XMPP. + +* I didn't quite get to [[design/assistant/deltas]], but the final month + of work on chunking provides a lot of new features and hopefully a + foundation that will get to deltas eventually. There is a new haskell + library that's being developed with the goal of being used for git-annex + deltas. + +* I hadn't planned to make git-annex be able to upgrade itself, when installed + from this website. But there was a need for that, and so it happened. + Even got a gpg key trust path for the distribution of git-annex. + +* Metadata driven views was an entirely unplanned feature. The current + prototype is very exciting, it opens up entire new use cases. + I had to hold myself back to not work on it too much, + especially as it shaded into adding a caching database to git-annex. + Had too much other stuff planned to do all I wanted. + Clearly this is an area I want to spend more time on! + +Those are most of the big features and changes, but probably half +of my work on git-annex this past year was in smaller things, and general +maintenance. Lots of others have contributed, some with +code (like the large effort to switch to bootstrap3), +and others with documentation, bug reports, etc. + +Perhaps it's best to turn to `git diff --stat` to sum up the activity +and see just how much both the crowdfunding campaign and +the previous kickstarter have pushed git-annex into high gear: + + campaign: 5410 files changed, 124159 insertions(+), 79395 deletions(-) + kickstarter: 4411 files changed, 123262 insertions(+), 13935 deletions(-) + year before: 1281 files changed, 7263 insertions(+), 55831 deletions(-) + +What's next? The hope is, no more crowdfunded campaigns where I have +to promise the moon anytime soon. Instead, the goal is to move to a more +mature and sustainable funding model, and continue to grow the git-annex +community, and the spaces where it's useful. |