2019-02-21 21:32:59 +00:00
|
|
|
{- git-annex import from remotes
|
|
|
|
-
|
2021-10-06 21:05:32 +00:00
|
|
|
- Copyright 2019-2021 Joey Hess <id@joeyh.name>
|
2019-02-21 21:32:59 +00:00
|
|
|
-
|
|
|
|
- Licensed under the GNU AGPL version 3 or higher.
|
|
|
|
-}
|
|
|
|
|
2019-02-27 17:15:02 +00:00
|
|
|
{-# LANGUAGE OverloadedStrings #-}
|
2019-02-26 19:25:28 +00:00
|
|
|
|
2019-02-23 19:47:55 +00:00
|
|
|
module Annex.Import (
|
|
|
|
ImportTreeConfig(..),
|
|
|
|
ImportCommitConfig(..),
|
|
|
|
buildImportCommit,
|
2019-02-26 19:25:28 +00:00
|
|
|
buildImportTrees,
|
add thirdPartyPopulated interface
This is to support, eg a borg repo as a special remote, which is
populated not by running git-annex commands, but by using borg. Then
git-annex sync lists the content of the remote, learns which files are
annex objects, and treats those as present in the remote.
So, most of the import machinery is reused, to a new purpose. While
normally importtree maintains a remote tracking branch, this does not,
because the files stored in the remote are annex object files, not
user-visible filenames. But, internally, a git tree is still generated,
of the files on the remote that are annex objects. This tree is used
by retrieveExportWithContentIdentifier, etc. As with other import/export
remotes, that the tree is recorded in the export log, and gets grafted
into the git-annex branch.
importKey changed to be able to return Nothing, to indicate when an
ImportLocation is not an annex object and so should be skipped from
being included in the tree.
It did not seem to make sense to have git-annex import do this, since
from the user's perspective, it's not like other imports. So only
git-annex sync does it.
Note that, git-annex sync does not yet download objects from such
remotes that are preferred content. importKeys is run with
content downloading disabled, to avoid getting the content of all
objects. Perhaps what's needed is for seekSyncContent to be run with these
remotes, but I don't know if it will just work (in particular, it needs
to avoid trying to transfer objects to them), so I skipped that for now.
(Untested and unused as of yet.)
This commit was sponsored by Jochen Bartl on Patreon.
2020-12-18 18:52:57 +00:00
|
|
|
recordImportTree,
|
2020-09-28 19:29:08 +00:00
|
|
|
canImportKeys,
|
2020-07-03 17:41:57 +00:00
|
|
|
importKeys,
|
2019-05-21 18:38:00 +00:00
|
|
|
makeImportMatcher,
|
2020-09-30 14:10:03 +00:00
|
|
|
getImportableContents,
|
2019-02-23 19:47:55 +00:00
|
|
|
) where
|
2019-02-21 21:32:59 +00:00
|
|
|
|
|
|
|
import Annex.Common
|
|
|
|
import Types.Import
|
2019-02-26 17:11:25 +00:00
|
|
|
import qualified Types.Remote as Remote
|
2019-02-21 21:32:59 +00:00
|
|
|
import Git.Types
|
|
|
|
import Git.Tree
|
2019-02-22 16:41:17 +00:00
|
|
|
import Git.Sha
|
2019-02-21 21:32:59 +00:00
|
|
|
import Git.FilePath
|
2019-04-24 19:13:07 +00:00
|
|
|
import Git.History
|
2019-02-22 16:41:17 +00:00
|
|
|
import qualified Git.Ref
|
|
|
|
import qualified Git.Branch
|
|
|
|
import qualified Annex
|
2019-02-21 21:32:59 +00:00
|
|
|
import Annex.Link
|
2019-02-22 16:41:17 +00:00
|
|
|
import Annex.LockFile
|
2019-02-27 17:15:02 +00:00
|
|
|
import Annex.Content
|
2019-03-01 17:26:15 +00:00
|
|
|
import Annex.Export
|
2019-05-01 17:13:00 +00:00
|
|
|
import Annex.RemoteTrackingBranch
|
2020-06-23 20:07:18 +00:00
|
|
|
import Annex.HashObject
|
2020-09-04 17:49:57 +00:00
|
|
|
import Annex.Transfer
|
2020-09-30 14:41:59 +00:00
|
|
|
import Annex.CheckIgnore
|
2020-12-23 19:21:33 +00:00
|
|
|
import Annex.VectorClock
|
2019-03-08 16:33:44 +00:00
|
|
|
import Command
|
2019-02-27 17:15:02 +00:00
|
|
|
import Backend
|
|
|
|
import Types.Key
|
|
|
|
import Types.KeySource
|
2019-03-08 16:43:03 +00:00
|
|
|
import Messages.Progress
|
2019-02-27 17:15:02 +00:00
|
|
|
import Utility.DataUnits
|
2019-06-25 15:37:52 +00:00
|
|
|
import Utility.Metered
|
2019-02-22 16:41:17 +00:00
|
|
|
import Logs.Export
|
2019-02-27 17:58:03 +00:00
|
|
|
import Logs.Location
|
2019-05-21 18:38:00 +00:00
|
|
|
import Logs.PreferredContent
|
|
|
|
import Types.FileMatcher
|
|
|
|
import Annex.FileMatcher
|
2020-09-28 17:22:16 +00:00
|
|
|
import qualified Utility.Matcher
|
2019-02-26 19:25:28 +00:00
|
|
|
import qualified Database.Export as Export
|
2019-03-06 22:04:30 +00:00
|
|
|
import qualified Database.ContentIdentifier as CIDDb
|
|
|
|
import qualified Logs.ContentIdentifier as CIDLog
|
2020-06-11 20:07:36 +00:00
|
|
|
import Backend.Utilities
|
2019-02-26 19:25:28 +00:00
|
|
|
|
|
|
|
import Control.Concurrent.STM
|
|
|
|
import qualified Data.Map.Strict as M
|
2019-03-08 16:33:44 +00:00
|
|
|
import qualified Data.Set as S
|
2020-09-30 14:10:03 +00:00
|
|
|
import qualified System.FilePath.Posix.ByteString as Posix
|
2019-12-09 17:49:05 +00:00
|
|
|
import qualified System.FilePath.ByteString as P
|
2019-03-04 20:02:56 +00:00
|
|
|
|
2019-02-23 19:47:55 +00:00
|
|
|
{- Configures how to build an import tree. -}
|
|
|
|
data ImportTreeConfig
|
|
|
|
= ImportTree
|
|
|
|
-- ^ Import the tree as-is from the remote.
|
|
|
|
| ImportSubTree TopFilePath Sha
|
|
|
|
-- ^ Import a tree from the remote and graft it into a subdirectory
|
|
|
|
-- of the existing tree whose Sha is provided, replacing anything
|
|
|
|
-- that was there before.
|
|
|
|
deriving (Show)
|
|
|
|
|
|
|
|
{- Configures how to build an import commit. -}
|
|
|
|
data ImportCommitConfig = ImportCommitConfig
|
make import tree from remote generate a merge commit
This way no history is lost, neither what was exported to the remote,
or the history of changes that is imported from it. No complicated
correlation of two possibly very different histories is needed, just
record what we know and then git merge will do a good job.
Also, it notices when the remote tracking branch doesn't need to be updated,
and avoids doing anything, so noop remotes are super cheap.
The only catch here is that, since the commits generated for imports
from the remote don't have a stable date or author/committer, each
(non-noop) import generates different commits for the same imported
trees. So, when the imported remote tracking branch is merged into master
and then a change is imported again, there will be an extra series of
commits, which will get more and more expensive each time.
This seems to call for making stable commits for imports. Also that
seems a good idea to make importing in several repositories have the
same result.
2019-04-30 20:13:21 +00:00
|
|
|
{ importCommitTracking :: Maybe Sha
|
|
|
|
-- ^ Current commit on the remote tracking branch.
|
2019-02-23 19:47:55 +00:00
|
|
|
, importCommitMode :: Git.Branch.CommitMode
|
|
|
|
, importCommitMessage :: String
|
|
|
|
}
|
|
|
|
|
make import tree from remote generate a merge commit
This way no history is lost, neither what was exported to the remote,
or the history of changes that is imported from it. No complicated
correlation of two possibly very different histories is needed, just
record what we know and then git merge will do a good job.
Also, it notices when the remote tracking branch doesn't need to be updated,
and avoids doing anything, so noop remotes are super cheap.
The only catch here is that, since the commits generated for imports
from the remote don't have a stable date or author/committer, each
(non-noop) import generates different commits for the same imported
trees. So, when the imported remote tracking branch is merged into master
and then a change is imported again, there will be an extra series of
commits, which will get more and more expensive each time.
This seems to call for making stable commits for imports. Also that
seems a good idea to make importing in several repositories have the
same result.
2019-04-30 20:13:21 +00:00
|
|
|
{- Buils a commit for an import from a special remote.
|
2019-02-21 21:32:59 +00:00
|
|
|
-
|
make import tree from remote generate a merge commit
This way no history is lost, neither what was exported to the remote,
or the history of changes that is imported from it. No complicated
correlation of two possibly very different histories is needed, just
record what we know and then git merge will do a good job.
Also, it notices when the remote tracking branch doesn't need to be updated,
and avoids doing anything, so noop remotes are super cheap.
The only catch here is that, since the commits generated for imports
from the remote don't have a stable date or author/committer, each
(non-noop) import generates different commits for the same imported
trees. So, when the imported remote tracking branch is merged into master
and then a change is imported again, there will be an extra series of
commits, which will get more and more expensive each time.
This seems to call for making stable commits for imports. Also that
seems a good idea to make importing in several repositories have the
same result.
2019-04-30 20:13:21 +00:00
|
|
|
- When there are no changes to make (importCommitTracking
|
|
|
|
- already matches what was imported), returns Nothing.
|
2019-02-23 19:47:55 +00:00
|
|
|
-
|
2019-02-21 21:32:59 +00:00
|
|
|
- After importing from a remote, exporting the same thing back to the
|
2019-02-22 16:41:17 +00:00
|
|
|
- remote should be a no-op. So, the export log and database are
|
|
|
|
- updated to reflect the imported tree.
|
2019-02-21 21:32:59 +00:00
|
|
|
-
|
2019-02-23 19:47:55 +00:00
|
|
|
- This does not download any content from a remote. But since it needs the
|
2019-02-22 16:41:17 +00:00
|
|
|
- Key of imported files to be known, its caller will have to first download
|
2019-02-21 21:32:59 +00:00
|
|
|
- new files in order to generate keys for them.
|
|
|
|
-}
|
|
|
|
buildImportCommit
|
2019-02-22 16:41:17 +00:00
|
|
|
:: Remote
|
2019-02-23 19:47:55 +00:00
|
|
|
-> ImportTreeConfig
|
|
|
|
-> ImportCommitConfig
|
2021-10-06 21:05:32 +00:00
|
|
|
-> ImportableContentsChunkable Annex (Either Sha Key)
|
2019-02-26 17:11:25 +00:00
|
|
|
-> Annex (Maybe Ref)
|
2019-02-23 19:47:55 +00:00
|
|
|
buildImportCommit remote importtreeconfig importcommitconfig importable =
|
make import tree from remote generate a merge commit
This way no history is lost, neither what was exported to the remote,
or the history of changes that is imported from it. No complicated
correlation of two possibly very different histories is needed, just
record what we know and then git merge will do a good job.
Also, it notices when the remote tracking branch doesn't need to be updated,
and avoids doing anything, so noop remotes are super cheap.
The only catch here is that, since the commits generated for imports
from the remote don't have a stable date or author/committer, each
(non-noop) import generates different commits for the same imported
trees. So, when the imported remote tracking branch is merged into master
and then a change is imported again, there will be an extra series of
commits, which will get more and more expensive each time.
This seems to call for making stable commits for imports. Also that
seems a good idea to make importing in several repositories have the
same result.
2019-04-30 20:13:21 +00:00
|
|
|
case importCommitTracking importcommitconfig of
|
2019-04-24 19:13:07 +00:00
|
|
|
Nothing -> go Nothing
|
make import tree from remote generate a merge commit
This way no history is lost, neither what was exported to the remote,
or the history of changes that is imported from it. No complicated
correlation of two possibly very different histories is needed, just
record what we know and then git merge will do a good job.
Also, it notices when the remote tracking branch doesn't need to be updated,
and avoids doing anything, so noop remotes are super cheap.
The only catch here is that, since the commits generated for imports
from the remote don't have a stable date or author/committer, each
(non-noop) import generates different commits for the same imported
trees. So, when the imported remote tracking branch is merged into master
and then a change is imported again, there will be an extra series of
commits, which will get more and more expensive each time.
This seems to call for making stable commits for imports. Also that
seems a good idea to make importing in several repositories have the
same result.
2019-04-30 20:13:21 +00:00
|
|
|
Just trackingcommit -> inRepo (Git.Ref.tree trackingcommit) >>= \case
|
2019-04-24 19:13:07 +00:00
|
|
|
Nothing -> go Nothing
|
make import tree from remote generate a merge commit
This way no history is lost, neither what was exported to the remote,
or the history of changes that is imported from it. No complicated
correlation of two possibly very different histories is needed, just
record what we know and then git merge will do a good job.
Also, it notices when the remote tracking branch doesn't need to be updated,
and avoids doing anything, so noop remotes are super cheap.
The only catch here is that, since the commits generated for imports
from the remote don't have a stable date or author/committer, each
(non-noop) import generates different commits for the same imported
trees. So, when the imported remote tracking branch is merged into master
and then a change is imported again, there will be an extra series of
commits, which will get more and more expensive each time.
This seems to call for making stable commits for imports. Also that
seems a good idea to make importing in several repositories have the
same result.
2019-04-30 20:13:21 +00:00
|
|
|
Just _ -> go (Just trackingcommit)
|
add thirdPartyPopulated interface
This is to support, eg a borg repo as a special remote, which is
populated not by running git-annex commands, but by using borg. Then
git-annex sync lists the content of the remote, learns which files are
annex objects, and treats those as present in the remote.
So, most of the import machinery is reused, to a new purpose. While
normally importtree maintains a remote tracking branch, this does not,
because the files stored in the remote are annex object files, not
user-visible filenames. But, internally, a git tree is still generated,
of the files on the remote that are annex objects. This tree is used
by retrieveExportWithContentIdentifier, etc. As with other import/export
remotes, that the tree is recorded in the export log, and gets grafted
into the git-annex branch.
importKey changed to be able to return Nothing, to indicate when an
ImportLocation is not an annex object and so should be skipped from
being included in the tree.
It did not seem to make sense to have git-annex import do this, since
from the user's perspective, it's not like other imports. So only
git-annex sync does it.
Note that, git-annex sync does not yet download objects from such
remotes that are preferred content. importKeys is run with
content downloading disabled, to avoid getting the content of all
objects. Perhaps what's needed is for seekSyncContent to be run with these
remotes, but I don't know if it will just work (in particular, it needs
to avoid trying to transfer objects to them), so I skipped that for now.
(Untested and unused as of yet.)
This commit was sponsored by Jochen Bartl on Patreon.
2020-12-18 18:52:57 +00:00
|
|
|
where
|
|
|
|
go trackingcommit = do
|
|
|
|
(imported, updatestate) <- recordImportTree remote importtreeconfig importable
|
|
|
|
buildImportCommit' remote importcommitconfig trackingcommit imported >>= \case
|
|
|
|
Just finalcommit -> do
|
|
|
|
updatestate
|
|
|
|
return (Just finalcommit)
|
|
|
|
Nothing -> return Nothing
|
|
|
|
|
|
|
|
{- Builds a tree for an import from a special remote.
|
|
|
|
-
|
|
|
|
- Also returns an action that can be used to update
|
|
|
|
- all the other state to record the import.
|
|
|
|
-}
|
|
|
|
recordImportTree
|
|
|
|
:: Remote
|
|
|
|
-> ImportTreeConfig
|
2021-10-06 21:05:32 +00:00
|
|
|
-> ImportableContentsChunkable Annex (Either Sha Key)
|
add thirdPartyPopulated interface
This is to support, eg a borg repo as a special remote, which is
populated not by running git-annex commands, but by using borg. Then
git-annex sync lists the content of the remote, learns which files are
annex objects, and treats those as present in the remote.
So, most of the import machinery is reused, to a new purpose. While
normally importtree maintains a remote tracking branch, this does not,
because the files stored in the remote are annex object files, not
user-visible filenames. But, internally, a git tree is still generated,
of the files on the remote that are annex objects. This tree is used
by retrieveExportWithContentIdentifier, etc. As with other import/export
remotes, that the tree is recorded in the export log, and gets grafted
into the git-annex branch.
importKey changed to be able to return Nothing, to indicate when an
ImportLocation is not an annex object and so should be skipped from
being included in the tree.
It did not seem to make sense to have git-annex import do this, since
from the user's perspective, it's not like other imports. So only
git-annex sync does it.
Note that, git-annex sync does not yet download objects from such
remotes that are preferred content. importKeys is run with
content downloading disabled, to avoid getting the content of all
objects. Perhaps what's needed is for seekSyncContent to be run with these
remotes, but I don't know if it will just work (in particular, it needs
to avoid trying to transfer objects to them), so I skipped that for now.
(Untested and unused as of yet.)
This commit was sponsored by Jochen Bartl on Patreon.
2020-12-18 18:52:57 +00:00
|
|
|
-> Annex (History Sha, Annex ())
|
|
|
|
recordImportTree remote importtreeconfig importable = do
|
|
|
|
imported@(History finaltree _) <- buildImportTrees basetree subdir importable
|
|
|
|
return (imported, updatestate finaltree)
|
2019-02-21 21:32:59 +00:00
|
|
|
where
|
2019-02-23 19:47:55 +00:00
|
|
|
basetree = case importtreeconfig of
|
|
|
|
ImportTree -> emptyTree
|
|
|
|
ImportSubTree _ sha -> sha
|
|
|
|
subdir = case importtreeconfig of
|
|
|
|
ImportTree -> Nothing
|
|
|
|
ImportSubTree dir _ -> Just dir
|
|
|
|
|
add thirdPartyPopulated interface
This is to support, eg a borg repo as a special remote, which is
populated not by running git-annex commands, but by using borg. Then
git-annex sync lists the content of the remote, learns which files are
annex objects, and treats those as present in the remote.
So, most of the import machinery is reused, to a new purpose. While
normally importtree maintains a remote tracking branch, this does not,
because the files stored in the remote are annex object files, not
user-visible filenames. But, internally, a git tree is still generated,
of the files on the remote that are annex objects. This tree is used
by retrieveExportWithContentIdentifier, etc. As with other import/export
remotes, that the tree is recorded in the export log, and gets grafted
into the git-annex branch.
importKey changed to be able to return Nothing, to indicate when an
ImportLocation is not an annex object and so should be skipped from
being included in the tree.
It did not seem to make sense to have git-annex import do this, since
from the user's perspective, it's not like other imports. So only
git-annex sync does it.
Note that, git-annex sync does not yet download objects from such
remotes that are preferred content. importKeys is run with
content downloading disabled, to avoid getting the content of all
objects. Perhaps what's needed is for seekSyncContent to be run with these
remotes, but I don't know if it will just work (in particular, it needs
to avoid trying to transfer objects to them), so I skipped that for now.
(Untested and unused as of yet.)
This commit was sponsored by Jochen Bartl on Patreon.
2020-12-18 18:52:57 +00:00
|
|
|
updatestate finaltree = do
|
2019-03-11 17:44:23 +00:00
|
|
|
importedtree <- case subdir of
|
add thirdPartyPopulated interface
This is to support, eg a borg repo as a special remote, which is
populated not by running git-annex commands, but by using borg. Then
git-annex sync lists the content of the remote, learns which files are
annex objects, and treats those as present in the remote.
So, most of the import machinery is reused, to a new purpose. While
normally importtree maintains a remote tracking branch, this does not,
because the files stored in the remote are annex object files, not
user-visible filenames. But, internally, a git tree is still generated,
of the files on the remote that are annex objects. This tree is used
by retrieveExportWithContentIdentifier, etc. As with other import/export
remotes, that the tree is recorded in the export log, and gets grafted
into the git-annex branch.
importKey changed to be able to return Nothing, to indicate when an
ImportLocation is not an annex object and so should be skipped from
being included in the tree.
It did not seem to make sense to have git-annex import do this, since
from the user's perspective, it's not like other imports. So only
git-annex sync does it.
Note that, git-annex sync does not yet download objects from such
remotes that are preferred content. importKeys is run with
content downloading disabled, to avoid getting the content of all
objects. Perhaps what's needed is for seekSyncContent to be run with these
remotes, but I don't know if it will just work (in particular, it needs
to avoid trying to transfer objects to them), so I skipped that for now.
(Untested and unused as of yet.)
This commit was sponsored by Jochen Bartl on Patreon.
2020-12-18 18:52:57 +00:00
|
|
|
Nothing -> pure finaltree
|
2019-03-11 17:44:23 +00:00
|
|
|
Just dir ->
|
|
|
|
let subtreeref = Ref $
|
add thirdPartyPopulated interface
This is to support, eg a borg repo as a special remote, which is
populated not by running git-annex commands, but by using borg. Then
git-annex sync lists the content of the remote, learns which files are
annex objects, and treats those as present in the remote.
So, most of the import machinery is reused, to a new purpose. While
normally importtree maintains a remote tracking branch, this does not,
because the files stored in the remote are annex object files, not
user-visible filenames. But, internally, a git tree is still generated,
of the files on the remote that are annex objects. This tree is used
by retrieveExportWithContentIdentifier, etc. As with other import/export
remotes, that the tree is recorded in the export log, and gets grafted
into the git-annex branch.
importKey changed to be able to return Nothing, to indicate when an
ImportLocation is not an annex object and so should be skipped from
being included in the tree.
It did not seem to make sense to have git-annex import do this, since
from the user's perspective, it's not like other imports. So only
git-annex sync does it.
Note that, git-annex sync does not yet download objects from such
remotes that are preferred content. importKeys is run with
content downloading disabled, to avoid getting the content of all
objects. Perhaps what's needed is for seekSyncContent to be run with these
remotes, but I don't know if it will just work (in particular, it needs
to avoid trying to transfer objects to them), so I skipped that for now.
(Untested and unused as of yet.)
This commit was sponsored by Jochen Bartl on Patreon.
2020-12-18 18:52:57 +00:00
|
|
|
fromRef' finaltree
|
2020-04-07 21:41:09 +00:00
|
|
|
<> ":"
|
|
|
|
<> getTopFilePath dir
|
2019-03-11 17:44:23 +00:00
|
|
|
in fromMaybe emptyTree
|
|
|
|
<$> inRepo (Git.Ref.tree subtreeref)
|
|
|
|
updateexportdb importedtree
|
|
|
|
oldexport <- updateexportlog importedtree
|
|
|
|
updatelocationlog oldexport importedtree
|
2019-04-24 19:13:07 +00:00
|
|
|
|
2019-03-07 19:59:44 +00:00
|
|
|
updateexportdb importedtree = do
|
|
|
|
db <- Export.openDb (Remote.uuid remote)
|
|
|
|
Export.writeLockDbWhile db $ do
|
2019-02-22 16:41:17 +00:00
|
|
|
prevtree <- liftIO $ fromMaybe emptyTree
|
2019-02-26 19:25:28 +00:00
|
|
|
<$> Export.getExportTreeCurrent db
|
2019-02-22 16:41:17 +00:00
|
|
|
when (importedtree /= prevtree) $ do
|
2019-02-27 19:29:41 +00:00
|
|
|
Export.updateExportDb db prevtree importedtree
|
2019-02-26 19:25:28 +00:00
|
|
|
liftIO $ Export.recordExportTreeCurrent db importedtree
|
2019-03-07 19:59:44 +00:00
|
|
|
Export.closeDb db
|
2019-02-23 19:47:55 +00:00
|
|
|
|
2019-02-22 16:41:17 +00:00
|
|
|
updateexportlog importedtree = do
|
2019-03-01 17:26:15 +00:00
|
|
|
oldexport <- getExport (Remote.uuid remote)
|
2020-12-23 18:27:38 +00:00
|
|
|
recordExport (Remote.uuid remote) importedtree $ ExportChange
|
2019-03-01 17:26:15 +00:00
|
|
|
{ oldTreeish = exportedTreeishes oldexport
|
2019-02-22 16:41:17 +00:00
|
|
|
, newTreeish = importedtree
|
|
|
|
}
|
2019-03-01 17:26:15 +00:00
|
|
|
return oldexport
|
|
|
|
|
|
|
|
-- downloadImport takes care of updating the location log
|
|
|
|
-- for the local repo when keys are downloaded, and also updates
|
|
|
|
-- the location log for the remote for keys that are present in it.
|
|
|
|
-- That leaves updating the location log for the remote for keys
|
|
|
|
-- that have had the last copy of their content removed from it.
|
|
|
|
--
|
|
|
|
-- This must run after the export database has been updated
|
|
|
|
-- and flushed to disk, so it can query it.
|
|
|
|
updatelocationlog oldexport finaltree = do
|
|
|
|
let stillpresent db k = liftIO $ not . null
|
|
|
|
<$> Export.getExportedLocation db k
|
2021-03-05 18:03:51 +00:00
|
|
|
let updater db moldkey _newkey _ = case moldkey of
|
2021-03-05 18:17:48 +00:00
|
|
|
Just oldkey | not (isGitShaKey oldkey) ->
|
|
|
|
unlessM (stillpresent db oldkey) $
|
2021-03-05 18:03:51 +00:00
|
|
|
logChange oldkey (Remote.uuid remote) InfoMissing
|
2021-03-05 18:17:48 +00:00
|
|
|
_ -> noop
|
2022-10-11 17:04:33 +00:00
|
|
|
-- When the remote is versioned, it still contains keys
|
|
|
|
-- that are not present in the new tree.
|
|
|
|
unless (Remote.versionedExport (Remote.exportActions remote)) $ do
|
|
|
|
db <- Export.openDb (Remote.uuid remote)
|
|
|
|
forM_ (exportedTreeishes oldexport) $ \oldtree ->
|
|
|
|
Export.runExportDiffUpdater updater db oldtree finaltree
|
|
|
|
Export.closeDb db
|
2019-02-21 21:32:59 +00:00
|
|
|
|
2019-05-20 20:37:04 +00:00
|
|
|
buildImportCommit' :: Remote -> ImportCommitConfig -> Maybe Sha -> History Sha -> Annex (Maybe Sha)
|
|
|
|
buildImportCommit' remote importcommitconfig mtrackingcommit imported@(History ti _) =
|
make import tree from remote generate a merge commit
This way no history is lost, neither what was exported to the remote,
or the history of changes that is imported from it. No complicated
correlation of two possibly very different histories is needed, just
record what we know and then git merge will do a good job.
Also, it notices when the remote tracking branch doesn't need to be updated,
and avoids doing anything, so noop remotes are super cheap.
The only catch here is that, since the commits generated for imports
from the remote don't have a stable date or author/committer, each
(non-noop) import generates different commits for the same imported
trees. So, when the imported remote tracking branch is merged into master
and then a change is imported again, there will be an extra series of
commits, which will get more and more expensive each time.
This seems to call for making stable commits for imports. Also that
seems a good idea to make importing in several repositories have the
same result.
2019-04-30 20:13:21 +00:00
|
|
|
case mtrackingcommit of
|
2019-05-21 15:32:54 +00:00
|
|
|
Nothing -> Just <$> mkcommitsunconnected imported
|
make import tree from remote generate a merge commit
This way no history is lost, neither what was exported to the remote,
or the history of changes that is imported from it. No complicated
correlation of two possibly very different histories is needed, just
record what we know and then git merge will do a good job.
Also, it notices when the remote tracking branch doesn't need to be updated,
and avoids doing anything, so noop remotes are super cheap.
The only catch here is that, since the commits generated for imports
from the remote don't have a stable date or author/committer, each
(non-noop) import generates different commits for the same imported
trees. So, when the imported remote tracking branch is merged into master
and then a change is imported again, there will be an extra series of
commits, which will get more and more expensive each time.
This seems to call for making stable commits for imports. Also that
seems a good idea to make importing in several repositories have the
same result.
2019-04-30 20:13:21 +00:00
|
|
|
Just trackingcommit -> do
|
|
|
|
-- Get history of tracking branch to at most
|
2019-05-01 18:20:26 +00:00
|
|
|
-- one more level deep than what was imported,
|
|
|
|
-- so we'll have enough history to compare,
|
|
|
|
-- but not spend too much time getting it.
|
2019-05-21 15:32:54 +00:00
|
|
|
let maxdepth = succ importeddepth
|
make import tree from remote generate a merge commit
This way no history is lost, neither what was exported to the remote,
or the history of changes that is imported from it. No complicated
correlation of two possibly very different histories is needed, just
record what we know and then git merge will do a good job.
Also, it notices when the remote tracking branch doesn't need to be updated,
and avoids doing anything, so noop remotes are super cheap.
The only catch here is that, since the commits generated for imports
from the remote don't have a stable date or author/committer, each
(non-noop) import generates different commits for the same imported
trees. So, when the imported remote tracking branch is merged into master
and then a change is imported again, there will be an extra series of
commits, which will get more and more expensive each time.
This seems to call for making stable commits for imports. Also that
seems a good idea to make importing in several repositories have the
same result.
2019-04-30 20:13:21 +00:00
|
|
|
inRepo (getHistoryToDepth maxdepth trackingcommit)
|
|
|
|
>>= go trackingcommit
|
2019-04-26 14:17:02 +00:00
|
|
|
where
|
2019-05-21 15:32:54 +00:00
|
|
|
go _ Nothing = Just <$> mkcommitsunconnected imported
|
make import tree from remote generate a merge commit
This way no history is lost, neither what was exported to the remote,
or the history of changes that is imported from it. No complicated
correlation of two possibly very different histories is needed, just
record what we know and then git merge will do a good job.
Also, it notices when the remote tracking branch doesn't need to be updated,
and avoids doing anything, so noop remotes are super cheap.
The only catch here is that, since the commits generated for imports
from the remote don't have a stable date or author/committer, each
(non-noop) import generates different commits for the same imported
trees. So, when the imported remote tracking branch is merged into master
and then a change is imported again, there will be an extra series of
commits, which will get more and more expensive each time.
This seems to call for making stable commits for imports. Also that
seems a good idea to make importing in several repositories have the
same result.
2019-04-30 20:13:21 +00:00
|
|
|
go trackingcommit (Just h)
|
2019-05-01 16:37:54 +00:00
|
|
|
-- If the tracking branch head is a merge commit
|
|
|
|
-- and one side of the merge matches the history,
|
|
|
|
-- nothing new needs to be committed.
|
2019-05-21 15:32:54 +00:00
|
|
|
| t == ti && any sametodepth (S.toList s) = return Nothing
|
2019-05-01 18:20:26 +00:00
|
|
|
-- If the tracking branch matches the history,
|
|
|
|
-- nothing new needs to be committed.
|
|
|
|
-- (This is unlikely to happen.)
|
2019-05-21 15:32:54 +00:00
|
|
|
| sametodepth h' = return Nothing
|
make import tree from remote generate a merge commit
This way no history is lost, neither what was exported to the remote,
or the history of changes that is imported from it. No complicated
correlation of two possibly very different histories is needed, just
record what we know and then git merge will do a good job.
Also, it notices when the remote tracking branch doesn't need to be updated,
and avoids doing anything, so noop remotes are super cheap.
The only catch here is that, since the commits generated for imports
from the remote don't have a stable date or author/committer, each
(non-noop) import generates different commits for the same imported
trees. So, when the imported remote tracking branch is merged into master
and then a change is imported again, there will be an extra series of
commits, which will get more and more expensive each time.
This seems to call for making stable commits for imports. Also that
seems a good idea to make importing in several repositories have the
same result.
2019-04-30 20:13:21 +00:00
|
|
|
| otherwise = do
|
2019-05-01 18:20:26 +00:00
|
|
|
importedcommit <- case getRemoteTrackingBranchImportHistory h of
|
2019-05-21 15:32:54 +00:00
|
|
|
Nothing -> mkcommitsunconnected imported
|
|
|
|
Just oldimported@(History oldhc _)
|
|
|
|
| importeddepth == 1 ->
|
|
|
|
mkcommitconnected imported oldimported
|
|
|
|
| otherwise -> do
|
|
|
|
let oldimportedtrees = mapHistory historyCommitTree oldimported
|
|
|
|
mknewcommits oldhc oldimportedtrees imported
|
2019-05-21 18:38:00 +00:00
|
|
|
ti' <- addBackExportExcluded remote ti
|
2019-05-01 17:13:00 +00:00
|
|
|
Just <$> makeRemoteTrackingBranchMergeCommit'
|
2019-05-20 20:37:04 +00:00
|
|
|
trackingcommit importedcommit ti'
|
make import tree from remote generate a merge commit
This way no history is lost, neither what was exported to the remote,
or the history of changes that is imported from it. No complicated
correlation of two possibly very different histories is needed, just
record what we know and then git merge will do a good job.
Also, it notices when the remote tracking branch doesn't need to be updated,
and avoids doing anything, so noop remotes are super cheap.
The only catch here is that, since the commits generated for imports
from the remote don't have a stable date or author/committer, each
(non-noop) import generates different commits for the same imported
trees. So, when the imported remote tracking branch is merged into master
and then a change is imported again, there will be an extra series of
commits, which will get more and more expensive each time.
This seems to call for making stable commits for imports. Also that
seems a good idea to make importing in several repositories have the
same result.
2019-04-30 20:13:21 +00:00
|
|
|
where
|
|
|
|
h'@(History t s) = mapHistory historyCommitTree h
|
2019-04-23 20:34:19 +00:00
|
|
|
|
2019-05-21 15:32:54 +00:00
|
|
|
importeddepth = historyDepth imported
|
|
|
|
|
|
|
|
sametodepth b = imported == truncateHistoryToDepth importeddepth b
|
2019-04-23 20:34:19 +00:00
|
|
|
|
make import tree from remote generate a merge commit
This way no history is lost, neither what was exported to the remote,
or the history of changes that is imported from it. No complicated
correlation of two possibly very different histories is needed, just
record what we know and then git merge will do a good job.
Also, it notices when the remote tracking branch doesn't need to be updated,
and avoids doing anything, so noop remotes are super cheap.
The only catch here is that, since the commits generated for imports
from the remote don't have a stable date or author/committer, each
(non-noop) import generates different commits for the same imported
trees. So, when the imported remote tracking branch is merged into master
and then a change is imported again, there will be an extra series of
commits, which will get more and more expensive each time.
This seems to call for making stable commits for imports. Also that
seems a good idea to make importing in several repositories have the
same result.
2019-04-30 20:13:21 +00:00
|
|
|
mkcommit parents tree = inRepo $ Git.Branch.commitTree
|
|
|
|
(importCommitMode importcommitconfig)
|
|
|
|
(importCommitMessage importcommitconfig)
|
|
|
|
parents
|
|
|
|
tree
|
|
|
|
|
2019-05-21 15:32:54 +00:00
|
|
|
-- Start a new history of import commits, not connected to any
|
|
|
|
-- prior import commits.
|
|
|
|
mkcommitsunconnected (History importedtree hs) = do
|
|
|
|
parents <- mapM mkcommitsunconnected (S.toList hs)
|
|
|
|
mkcommit parents importedtree
|
|
|
|
|
|
|
|
-- Commit the new history connected with the old history.
|
|
|
|
-- Used when the import is not versioned, so the history depth is 1.
|
|
|
|
mkcommitconnected (History importedtree _) (History oldhc _) = do
|
|
|
|
let parents = [historyCommit oldhc]
|
2019-05-01 18:20:26 +00:00
|
|
|
mkcommit parents importedtree
|
|
|
|
|
2019-05-01 19:34:07 +00:00
|
|
|
-- Reuse the commits from the old imported History when possible.
|
|
|
|
mknewcommits oldhc old new@(History importedtree hs)
|
|
|
|
| old == new = return $ historyCommit oldhc
|
2019-05-01 18:20:26 +00:00
|
|
|
| otherwise = do
|
2019-05-01 19:34:07 +00:00
|
|
|
parents <- mapM (mknewcommits oldhc old) (S.toList hs)
|
2019-05-01 18:20:26 +00:00
|
|
|
mkcommit parents importedtree
|
|
|
|
|
2019-02-22 16:41:17 +00:00
|
|
|
{- Builds a history of git trees reflecting the ImportableContents.
|
|
|
|
-
|
|
|
|
- When a subdir is provided, imported tree is grafted into the basetree at
|
|
|
|
- that location, replacing any object that was there.
|
|
|
|
-}
|
2019-02-21 21:32:59 +00:00
|
|
|
buildImportTrees
|
2021-10-06 21:05:32 +00:00
|
|
|
:: Ref
|
|
|
|
-> Maybe TopFilePath
|
|
|
|
-> ImportableContentsChunkable Annex (Either Sha Key)
|
|
|
|
-> Annex (History Sha)
|
|
|
|
buildImportTrees basetree msubdir (ImportableContentsComplete importable) = do
|
|
|
|
repo <- Annex.gitRepo
|
|
|
|
withMkTreeHandle repo $ buildImportTrees' basetree msubdir importable
|
|
|
|
buildImportTrees basetree msubdir importable@(ImportableContentsChunked {}) = do
|
|
|
|
repo <- Annex.gitRepo
|
|
|
|
withMkTreeHandle repo $ \hdl ->
|
|
|
|
History
|
|
|
|
<$> go hdl
|
|
|
|
<*> buildImportTreesHistory basetree msubdir
|
|
|
|
(importableHistoryComplete importable) hdl
|
|
|
|
where
|
|
|
|
go hdl = do
|
|
|
|
tree <- gochunks [] (importableContentsChunk importable) hdl
|
|
|
|
importtree <- liftIO $ recordTree' hdl tree
|
|
|
|
graftImportTree basetree msubdir importtree hdl
|
|
|
|
|
|
|
|
gochunks l c hdl = do
|
|
|
|
let subdir = importChunkSubDir $ importableContentsSubDir c
|
|
|
|
-- Full directory prefix where the sub tree is located.
|
|
|
|
let fullprefix = asTopFilePath $ case msubdir of
|
|
|
|
Nothing -> subdir
|
|
|
|
Just d -> getTopFilePath d Posix.</> subdir
|
|
|
|
Tree ts <- convertImportTree (Just fullprefix) $
|
|
|
|
map (\(p, i) -> (mkImportLocation p, i))
|
|
|
|
(importableContentsSubTree c)
|
|
|
|
-- Record this subtree before getting next chunk, this
|
|
|
|
-- avoids buffering all the chunks into memory.
|
|
|
|
tc <- liftIO $ recordSubTree hdl $
|
|
|
|
NewSubTree (asTopFilePath subdir) ts
|
|
|
|
importableContentsNextChunk c >>= \case
|
|
|
|
Nothing -> return (Tree (tc:l))
|
|
|
|
Just c' -> gochunks (tc:l) c' hdl
|
|
|
|
|
|
|
|
buildImportTrees'
|
2019-02-22 16:41:17 +00:00
|
|
|
:: Ref
|
|
|
|
-> Maybe TopFilePath
|
2020-06-23 20:07:18 +00:00
|
|
|
-> ImportableContents (Either Sha Key)
|
2021-10-06 21:05:32 +00:00
|
|
|
-> MkTreeHandle
|
2019-02-21 21:32:59 +00:00
|
|
|
-> Annex (History Sha)
|
2021-10-06 21:05:32 +00:00
|
|
|
buildImportTrees' basetree msubdir importable hdl = History
|
|
|
|
<$> buildImportTree basetree msubdir (importableContents importable) hdl
|
|
|
|
<*> buildImportTreesHistory basetree msubdir (importableHistory importable) hdl
|
|
|
|
|
|
|
|
buildImportTree
|
|
|
|
:: Ref
|
|
|
|
-> Maybe TopFilePath
|
|
|
|
-> [(ImportLocation, Either Sha Key)]
|
|
|
|
-> MkTreeHandle
|
|
|
|
-> Annex Sha
|
|
|
|
buildImportTree basetree msubdir ls hdl = do
|
|
|
|
importtree <- liftIO . recordTree' hdl =<< convertImportTree msubdir ls
|
|
|
|
graftImportTree basetree msubdir importtree hdl
|
|
|
|
|
|
|
|
graftImportTree
|
|
|
|
:: Ref
|
|
|
|
-> Maybe TopFilePath
|
|
|
|
-> Sha
|
|
|
|
-> MkTreeHandle
|
|
|
|
-> Annex Sha
|
|
|
|
graftImportTree basetree msubdir tree hdl = case msubdir of
|
|
|
|
Nothing -> return tree
|
|
|
|
Just subdir -> inRepo $ \repo ->
|
|
|
|
graftTree' tree subdir basetree repo hdl
|
|
|
|
|
|
|
|
convertImportTree :: Maybe TopFilePath -> [(ImportLocation, Either Sha Key)] -> Annex Tree
|
|
|
|
convertImportTree msubdir ls = treeItemsToTree <$> mapM mktreeitem ls
|
2019-02-21 21:32:59 +00:00
|
|
|
where
|
2020-06-23 20:07:18 +00:00
|
|
|
mktreeitem (loc, v) = case v of
|
|
|
|
Right k -> do
|
|
|
|
relf <- fromRepo $ fromTopFilePath topf
|
2020-11-03 22:34:27 +00:00
|
|
|
symlink <- calcRepo $ gitAnnexLink relf k
|
2020-06-23 20:07:18 +00:00
|
|
|
linksha <- hashSymlink symlink
|
|
|
|
return $ TreeItem treepath (fromTreeItemType TreeSymlink) linksha
|
|
|
|
Left sha ->
|
|
|
|
return $ TreeItem treepath (fromTreeItemType TreeFile) sha
|
|
|
|
where
|
|
|
|
lf = fromImportLocation loc
|
|
|
|
treepath = asTopFilePath lf
|
|
|
|
topf = asTopFilePath $
|
2019-12-09 17:49:05 +00:00
|
|
|
maybe lf (\sd -> getTopFilePath sd P.</> lf) msubdir
|
2019-02-26 19:25:28 +00:00
|
|
|
|
2021-10-06 21:05:32 +00:00
|
|
|
buildImportTreesHistory
|
|
|
|
:: Ref
|
|
|
|
-> Maybe TopFilePath
|
|
|
|
-> [ImportableContents (Either Sha Key)]
|
|
|
|
-> MkTreeHandle
|
|
|
|
-> Annex (S.Set (History Sha))
|
|
|
|
buildImportTreesHistory basetree msubdir history hdl = S.fromList
|
|
|
|
<$> mapM (\ic -> buildImportTrees' basetree msubdir ic hdl) history
|
|
|
|
|
2020-09-28 19:29:08 +00:00
|
|
|
canImportKeys :: Remote -> Bool -> Bool
|
|
|
|
canImportKeys remote importcontent =
|
|
|
|
importcontent || isJust (Remote.importKey ia)
|
|
|
|
where
|
|
|
|
ia = Remote.importActions remote
|
|
|
|
|
2020-07-03 17:41:57 +00:00
|
|
|
{- Downloads all new ContentIdentifiers, or when importcontent is False,
|
|
|
|
- generates Keys without downloading.
|
2019-02-26 19:25:28 +00:00
|
|
|
-
|
2020-07-03 17:41:57 +00:00
|
|
|
- Generates either a Key or a git Sha, depending on annex.largefiles.
|
|
|
|
- But when importcontent is False, it cannot match on annex.largefiles
|
|
|
|
- (or generate a git Sha), so always generates Keys.
|
|
|
|
-
|
|
|
|
- Supports concurrency when enabled.
|
|
|
|
-
|
|
|
|
- If it fails on any file, the whole thing fails with Nothing,
|
2019-04-10 21:02:56 +00:00
|
|
|
- but it will resume where it left off.
|
2020-06-23 20:07:18 +00:00
|
|
|
-
|
2020-07-03 17:41:57 +00:00
|
|
|
- Note that, when a ContentIdentifier has been imported before,
|
|
|
|
- generates the same thing that was imported before, so annex.largefiles
|
|
|
|
- is not reapplied.
|
2019-02-26 19:25:28 +00:00
|
|
|
-}
|
2020-07-03 17:41:57 +00:00
|
|
|
importKeys
|
|
|
|
:: Remote
|
|
|
|
-> ImportTreeConfig
|
|
|
|
-> Bool
|
add thirdPartyPopulated interface
This is to support, eg a borg repo as a special remote, which is
populated not by running git-annex commands, but by using borg. Then
git-annex sync lists the content of the remote, learns which files are
annex objects, and treats those as present in the remote.
So, most of the import machinery is reused, to a new purpose. While
normally importtree maintains a remote tracking branch, this does not,
because the files stored in the remote are annex object files, not
user-visible filenames. But, internally, a git tree is still generated,
of the files on the remote that are annex objects. This tree is used
by retrieveExportWithContentIdentifier, etc. As with other import/export
remotes, that the tree is recorded in the export log, and gets grafted
into the git-annex branch.
importKey changed to be able to return Nothing, to indicate when an
ImportLocation is not an annex object and so should be skipped from
being included in the tree.
It did not seem to make sense to have git-annex import do this, since
from the user's perspective, it's not like other imports. So only
git-annex sync does it.
Note that, git-annex sync does not yet download objects from such
remotes that are preferred content. importKeys is run with
content downloading disabled, to avoid getting the content of all
objects. Perhaps what's needed is for seekSyncContent to be run with these
remotes, but I don't know if it will just work (in particular, it needs
to avoid trying to transfer objects to them), so I skipped that for now.
(Untested and unused as of yet.)
This commit was sponsored by Jochen Bartl on Patreon.
2020-12-18 18:52:57 +00:00
|
|
|
-> Bool
|
2021-10-06 21:05:32 +00:00
|
|
|
-> ImportableContentsChunkable Annex (ContentIdentifier, ByteSize)
|
|
|
|
-> Annex (Maybe (ImportableContentsChunkable Annex (Either Sha Key)))
|
2020-12-21 20:03:27 +00:00
|
|
|
importKeys remote importtreeconfig importcontent thirdpartypopulated importablecontents = do
|
2020-09-28 19:29:08 +00:00
|
|
|
unless (canImportKeys remote importcontent) $
|
2020-07-03 17:41:57 +00:00
|
|
|
giveup "This remote does not support importing without downloading content."
|
2019-02-26 19:25:28 +00:00
|
|
|
-- This map is used to remember content identifiers that
|
2020-07-03 17:41:57 +00:00
|
|
|
-- were just imported, before they have necessarily been
|
2019-02-26 19:25:28 +00:00
|
|
|
-- stored in the database. This way, if the same content
|
|
|
|
-- identifier appears multiple times in the
|
|
|
|
-- importablecontents (eg when it has a history),
|
2020-07-03 17:41:57 +00:00
|
|
|
-- they will only be imported once.
|
2019-02-26 19:25:28 +00:00
|
|
|
cidmap <- liftIO $ newTVarIO M.empty
|
2019-03-08 16:33:44 +00:00
|
|
|
-- When concurrency is enabled, this set is needed to
|
2020-07-03 17:41:57 +00:00
|
|
|
-- avoid two threads both importing the same content identifier.
|
|
|
|
importing <- liftIO $ newTVarIO S.empty
|
2021-10-06 21:05:32 +00:00
|
|
|
withciddb $ \db -> do
|
|
|
|
CIDDb.needsUpdateFromLog db
|
|
|
|
>>= maybe noop (CIDDb.updateFromLog db)
|
|
|
|
(prepclock (run cidmap importing db))
|
2019-02-26 19:25:28 +00:00
|
|
|
where
|
2020-12-23 19:21:33 +00:00
|
|
|
-- When not importing content, reuse the same vector
|
|
|
|
-- clock for all state that's recorded. This can save
|
|
|
|
-- a little bit of disk space. Individual file downloads
|
|
|
|
-- while downloading take too long for this optimisation
|
|
|
|
-- to be safe to do.
|
2021-10-06 21:05:32 +00:00
|
|
|
prepclock a
|
2020-12-23 19:21:33 +00:00
|
|
|
| importcontent = a
|
|
|
|
| otherwise = reuseVectorClockWhile a
|
|
|
|
|
2022-08-11 20:57:44 +00:00
|
|
|
withciddb a = do
|
|
|
|
cidlck <- calcRepo' gitAnnexContentIdentifierLock
|
|
|
|
withExclusiveLock cidlck $
|
|
|
|
bracket CIDDb.openDb CIDDb.closeDb a
|
2021-10-06 21:05:32 +00:00
|
|
|
|
|
|
|
run cidmap importing db = do
|
2020-06-23 20:07:18 +00:00
|
|
|
largematcher <- largeFilesMatcher
|
2021-10-06 21:05:32 +00:00
|
|
|
case importablecontents of
|
|
|
|
ImportableContentsComplete ic ->
|
|
|
|
go False largematcher cidmap importing db ic >>= return . \case
|
|
|
|
Nothing -> Nothing
|
|
|
|
Just v -> Just $ ImportableContentsComplete v
|
|
|
|
ImportableContentsChunked {} -> do
|
|
|
|
c <- gochunked db (importableContentsChunk importablecontents)
|
|
|
|
gohistory largematcher cidmap importing db (importableHistoryComplete importablecontents) >>= return . \case
|
|
|
|
Nothing -> Nothing
|
|
|
|
Just h -> Just $ ImportableContentsChunked
|
|
|
|
{ importableContentsChunk = c
|
|
|
|
, importableHistoryComplete = h
|
|
|
|
}
|
|
|
|
|
|
|
|
go oldversion largematcher cidmap importing db (ImportableContents l h) = do
|
2019-03-08 16:33:44 +00:00
|
|
|
jobs <- forM l $ \i ->
|
2020-12-21 20:03:27 +00:00
|
|
|
if thirdpartypopulated
|
2021-10-06 21:05:32 +00:00
|
|
|
then Left <$> thirdpartypopulatedimport db i
|
2020-12-21 20:03:27 +00:00
|
|
|
else startimport cidmap importing db i oldversion largematcher
|
2019-03-08 16:33:44 +00:00
|
|
|
l' <- liftIO $ forM jobs $
|
|
|
|
either pure (atomically . takeTMVar)
|
2019-02-26 19:25:28 +00:00
|
|
|
if any isNothing l'
|
|
|
|
then return Nothing
|
2021-10-06 21:05:32 +00:00
|
|
|
else gohistory largematcher cidmap importing db h >>= return . \case
|
|
|
|
Nothing -> Nothing
|
|
|
|
Just h' -> Just $ ImportableContents (catMaybes l') h'
|
|
|
|
|
|
|
|
gohistory largematcher cidmap importing db h = do
|
|
|
|
h' <- mapM (go True largematcher cidmap importing db) h
|
|
|
|
if any isNothing h'
|
|
|
|
then return Nothing
|
|
|
|
else return $ Just $ catMaybes h'
|
2019-02-26 19:25:28 +00:00
|
|
|
|
2021-10-06 21:05:32 +00:00
|
|
|
gochunked db c
|
|
|
|
-- Downloading cannot be done when chunked, since only
|
|
|
|
-- the first chunk is processed before returning.
|
|
|
|
| importcontent = error "importKeys does not support downloading chunked import"
|
|
|
|
-- Chunked import is currently only used by thirdpartypopulated
|
|
|
|
-- remotes.
|
|
|
|
| not thirdpartypopulated = error "importKeys does not support chunked import when not thirdpartypopulated"
|
|
|
|
| otherwise = do
|
|
|
|
l <- forM (importableContentsSubTree c) $ \(loc, i) -> do
|
|
|
|
let loc' = importableContentsChunkFullLocation (importableContentsSubDir c) loc
|
|
|
|
thirdpartypopulatedimport db (loc', i) >>= return . \case
|
|
|
|
Just (_loc, k) -> Just (loc, k)
|
|
|
|
Nothing -> Nothing
|
|
|
|
return $ ImportableContentsChunk
|
|
|
|
{ importableContentsSubDir = importableContentsSubDir c
|
|
|
|
, importableContentsSubTree = catMaybes l
|
|
|
|
, importableContentsNextChunk =
|
|
|
|
importableContentsNextChunk c >>= \case
|
|
|
|
Nothing -> return Nothing
|
|
|
|
Just c' -> withciddb $ \db' ->
|
|
|
|
prepclock $
|
|
|
|
Just <$> gochunked db' c'
|
|
|
|
}
|
|
|
|
|
2020-07-03 17:41:57 +00:00
|
|
|
waitstart importing cid = liftIO $ atomically $ do
|
|
|
|
s <- readTVar importing
|
2019-03-08 16:33:44 +00:00
|
|
|
if S.member cid s
|
|
|
|
then retry
|
2020-07-03 17:41:57 +00:00
|
|
|
else writeTVar importing $ S.insert cid s
|
2019-03-08 16:33:44 +00:00
|
|
|
|
2020-07-03 17:41:57 +00:00
|
|
|
signaldone importing cid = liftIO $ atomically $ do
|
|
|
|
s <- readTVar importing
|
|
|
|
writeTVar importing $ S.delete cid s
|
2019-03-08 16:33:44 +00:00
|
|
|
|
2020-07-03 17:41:57 +00:00
|
|
|
startimport cidmap importing db i@(loc, (cid, _sz)) oldversion largematcher = getcidkey cidmap db cid >>= \case
|
2020-06-23 20:07:18 +00:00
|
|
|
(k:ks) ->
|
|
|
|
-- If the same content was imported before
|
2023-03-14 02:39:16 +00:00
|
|
|
-- yielding multiple different keys, it's not clear
|
2020-06-23 20:07:18 +00:00
|
|
|
-- which is best to use this time, so pick the
|
|
|
|
-- first in the list. But, if any of them is a
|
|
|
|
-- git sha, use it, because the content must
|
|
|
|
-- be included in the git repo then.
|
|
|
|
let v = case mapMaybe keyGitSha (k:ks) of
|
|
|
|
(sha:_) -> Left sha
|
|
|
|
[] -> Right k
|
|
|
|
in return $ Left $ Just (loc, v)
|
2019-03-08 16:33:44 +00:00
|
|
|
[] -> do
|
|
|
|
job <- liftIO $ newEmptyTMVarIO
|
2023-04-08 19:48:32 +00:00
|
|
|
let ai = ActionItemOther (Just (QuotedPath (fromImportLocation loc)))
|
2020-09-14 20:49:33 +00:00
|
|
|
let si = SeekInput []
|
|
|
|
let importaction = starting ("import " ++ Remote.name remote) ai si $ do
|
2019-04-19 19:05:08 +00:00
|
|
|
when oldversion $
|
|
|
|
showNote "old version"
|
2020-07-03 17:41:57 +00:00
|
|
|
tryNonAsync (importordownload cidmap db i largematcher) >>= \case
|
2019-03-08 16:33:44 +00:00
|
|
|
Left e -> next $ do
|
|
|
|
warning (show e)
|
|
|
|
liftIO $ atomically $
|
|
|
|
putTMVar job Nothing
|
|
|
|
return False
|
|
|
|
Right r -> next $ do
|
|
|
|
liftIO $ atomically $
|
|
|
|
putTMVar job r
|
|
|
|
return True
|
|
|
|
commandAction $ bracket_
|
2020-07-03 17:41:57 +00:00
|
|
|
(waitstart importing cid)
|
|
|
|
(signaldone importing cid)
|
|
|
|
importaction
|
2019-03-08 16:33:44 +00:00
|
|
|
return (Right job)
|
|
|
|
|
2021-10-06 21:05:32 +00:00
|
|
|
thirdpartypopulatedimport db (loc, (cid, sz)) =
|
2020-12-21 20:03:27 +00:00
|
|
|
case Remote.importKey ia of
|
2021-10-06 21:05:32 +00:00
|
|
|
Nothing -> return Nothing
|
2020-12-21 20:03:27 +00:00
|
|
|
Just importkey ->
|
|
|
|
tryNonAsync (importkey loc cid sz nullMeterUpdate) >>= \case
|
|
|
|
Right (Just k) -> do
|
2021-10-06 21:05:32 +00:00
|
|
|
recordcidkey' db cid k
|
|
|
|
logChange k (Remote.uuid remote) InfoPresent
|
|
|
|
return $ Just (loc, Right k)
|
|
|
|
Right Nothing -> return Nothing
|
2020-12-21 20:03:27 +00:00
|
|
|
Left e -> do
|
|
|
|
warning (show e)
|
2021-10-06 21:05:32 +00:00
|
|
|
return Nothing
|
2020-12-21 20:03:27 +00:00
|
|
|
|
2020-09-28 19:03:15 +00:00
|
|
|
importordownload cidmap db (loc, (cid, sz)) largematcher= do
|
|
|
|
f <- locworktreefile loc
|
2020-11-04 18:20:37 +00:00
|
|
|
matcher <- largematcher f
|
2020-09-28 19:03:15 +00:00
|
|
|
-- When importing a key is supported, always use it rather
|
|
|
|
-- than downloading and retrieving a key, to avoid
|
|
|
|
-- generating trees with different keys for the same content.
|
|
|
|
let act = if importcontent
|
|
|
|
then case Remote.importKey ia of
|
|
|
|
Nothing -> dodownload
|
2020-12-21 20:03:27 +00:00
|
|
|
Just _ -> if Utility.Matcher.introspect matchNeedsFileContent matcher
|
2020-09-28 19:03:15 +00:00
|
|
|
then dodownload
|
|
|
|
else doimport
|
|
|
|
else doimport
|
|
|
|
act cidmap db (loc, (cid, sz)) f matcher
|
2020-07-03 17:41:57 +00:00
|
|
|
|
2020-09-28 19:03:15 +00:00
|
|
|
doimport cidmap db (loc, (cid, sz)) f matcher =
|
2020-07-03 17:41:57 +00:00
|
|
|
case Remote.importKey ia of
|
|
|
|
Nothing -> error "internal" -- checked earlier
|
2020-09-28 17:22:16 +00:00
|
|
|
Just importkey -> do
|
2020-12-21 20:03:27 +00:00
|
|
|
when (Utility.Matcher.introspect matchNeedsFileContent matcher) $
|
2020-09-28 17:22:16 +00:00
|
|
|
giveup "annex.largefiles configuration examines file contents, so cannot import without content."
|
2020-09-28 19:03:15 +00:00
|
|
|
let mi = MatchingInfo ProvidedInfo
|
2021-03-02 16:47:23 +00:00
|
|
|
{ providedFilePath = Just f
|
2020-09-28 17:22:16 +00:00
|
|
|
, providedKey = Nothing
|
2021-03-02 16:47:23 +00:00
|
|
|
, providedFileSize = Just sz
|
2020-09-28 17:22:16 +00:00
|
|
|
, providedMimeType = Nothing
|
|
|
|
, providedMimeEncoding = Nothing
|
2021-03-02 16:47:23 +00:00
|
|
|
, providedLinkType = Nothing
|
2020-09-28 17:22:16 +00:00
|
|
|
}
|
2021-09-22 14:51:10 +00:00
|
|
|
let bwlimit = remoteAnnexBwLimit (Remote.gitconfig remote)
|
2020-12-21 20:03:27 +00:00
|
|
|
islargefile <- checkMatcher' matcher mi mempty
|
bwlimit
Added annex.bwlimit and remote.name.annex-bwlimit config that works for git
remotes and many but not all special remotes.
This nearly works, at least for a git remote on the same disk. With it set
to 100kb/1s, the meter displays an actual bandwidth of 128 kb/s, with
occasional spikes to 160 kb/s. So it needs to delay just a bit longer...
I'm unsure why.
However, at the beginning a lot of data flows before it determines the
right bandwidth limit. A granularity of less than 1s would probably improve
that.
And, I don't know yet if it makes sense to have it be 100ks/1s rather than
100kb/s. Is there a situation where the user would want a larger
granularity? Does granulatity need to be configurable at all? I only used that
format for the config really in order to reuse an existing parser.
This can't support for external special remotes, or for ones that
themselves shell out to an external command. (Well, it could, but it
would involve pausing and resuming the child process tree, which seems
very hard to implement and very strange besides.) There could also be some
built-in special remotes that it still doesn't work for, due to them not
having a progress meter whose displays blocks the bandwidth using thread.
But I don't think there are actually any that run a separate thread for
downloads than the thread that displays the progress meter.
Sponsored-by: Graham Spencer on Patreon
2021-09-21 20:58:02 +00:00
|
|
|
metered Nothing sz bwlimit $ const $ if islargefile
|
2020-09-28 19:03:15 +00:00
|
|
|
then doimportlarge importkey cidmap db loc cid sz f
|
2020-09-28 17:22:16 +00:00
|
|
|
else doimportsmall cidmap db loc cid sz
|
|
|
|
|
2020-09-28 19:03:15 +00:00
|
|
|
doimportlarge importkey cidmap db loc cid sz f p =
|
2020-09-28 17:22:16 +00:00
|
|
|
tryNonAsync importer >>= \case
|
2020-09-28 19:03:15 +00:00
|
|
|
Right (Just (k, True)) -> return $ Just (loc, Right k)
|
|
|
|
Right _ -> return Nothing
|
2020-09-28 17:22:16 +00:00
|
|
|
Left e -> do
|
|
|
|
warning (show e)
|
|
|
|
return Nothing
|
|
|
|
where
|
|
|
|
importer = do
|
add thirdPartyPopulated interface
This is to support, eg a borg repo as a special remote, which is
populated not by running git-annex commands, but by using borg. Then
git-annex sync lists the content of the remote, learns which files are
annex objects, and treats those as present in the remote.
So, most of the import machinery is reused, to a new purpose. While
normally importtree maintains a remote tracking branch, this does not,
because the files stored in the remote are annex object files, not
user-visible filenames. But, internally, a git tree is still generated,
of the files on the remote that are annex objects. This tree is used
by retrieveExportWithContentIdentifier, etc. As with other import/export
remotes, that the tree is recorded in the export log, and gets grafted
into the git-annex branch.
importKey changed to be able to return Nothing, to indicate when an
ImportLocation is not an annex object and so should be skipped from
being included in the tree.
It did not seem to make sense to have git-annex import do this, since
from the user's perspective, it's not like other imports. So only
git-annex sync does it.
Note that, git-annex sync does not yet download objects from such
remotes that are preferred content. importKeys is run with
content downloading disabled, to avoid getting the content of all
objects. Perhaps what's needed is for seekSyncContent to be run with these
remotes, but I don't know if it will just work (in particular, it needs
to avoid trying to transfer objects to them), so I skipped that for now.
(Untested and unused as of yet.)
This commit was sponsored by Jochen Bartl on Patreon.
2020-12-18 18:52:57 +00:00
|
|
|
-- Don't display progress when generating
|
|
|
|
-- key, if the content will later be
|
|
|
|
-- downloaded, which is a more expensive
|
|
|
|
-- operation generally.
|
|
|
|
let p' = if importcontent then nullMeterUpdate else p
|
2020-12-18 20:52:49 +00:00
|
|
|
importkey loc cid sz p' >>= \case
|
add thirdPartyPopulated interface
This is to support, eg a borg repo as a special remote, which is
populated not by running git-annex commands, but by using borg. Then
git-annex sync lists the content of the remote, learns which files are
annex objects, and treats those as present in the remote.
So, most of the import machinery is reused, to a new purpose. While
normally importtree maintains a remote tracking branch, this does not,
because the files stored in the remote are annex object files, not
user-visible filenames. But, internally, a git tree is still generated,
of the files on the remote that are annex objects. This tree is used
by retrieveExportWithContentIdentifier, etc. As with other import/export
remotes, that the tree is recorded in the export log, and gets grafted
into the git-annex branch.
importKey changed to be able to return Nothing, to indicate when an
ImportLocation is not an annex object and so should be skipped from
being included in the tree.
It did not seem to make sense to have git-annex import do this, since
from the user's perspective, it's not like other imports. So only
git-annex sync does it.
Note that, git-annex sync does not yet download objects from such
remotes that are preferred content. importKeys is run with
content downloading disabled, to avoid getting the content of all
objects. Perhaps what's needed is for seekSyncContent to be run with these
remotes, but I don't know if it will just work (in particular, it needs
to avoid trying to transfer objects to them), so I skipped that for now.
(Untested and unused as of yet.)
This commit was sponsored by Jochen Bartl on Patreon.
2020-12-18 18:52:57 +00:00
|
|
|
Nothing -> return Nothing
|
2020-12-21 20:03:27 +00:00
|
|
|
Just k -> checkSecureHashes k >>= \case
|
|
|
|
Nothing -> do
|
|
|
|
recordcidkey cidmap db cid k
|
|
|
|
logChange k (Remote.uuid remote) InfoPresent
|
|
|
|
if importcontent
|
|
|
|
then getcontent k
|
|
|
|
else return (Just (k, True))
|
|
|
|
Just msg -> giveup (msg ++ " to import")
|
|
|
|
|
2020-09-28 19:03:15 +00:00
|
|
|
getcontent :: Key -> Annex (Maybe (Key, Bool))
|
|
|
|
getcontent k = do
|
|
|
|
let af = AssociatedFile (Just f)
|
|
|
|
let downloader p' tmpfile = do
|
2022-05-09 19:38:21 +00:00
|
|
|
_ <- Remote.retrieveExportWithContentIdentifier
|
change retrieveExportWithContentIdentifier to take a list of ContentIdentifier
This partly fixes an issue where there are duplicate files in the
special remote, and the first file gets swapped with another duplicate,
or deleted. The swap case is fixed by this, the deleted case will need
other changes.
This makes retrieveExportWithContentIdentifier take a list of allowed
ContentIdentifier, same as storeExportWithContentIdentifier,
removeExportWithContentIdentifier, and
checkPresentExportWithContentIdentifier.
Of the special remotes that support importtree, borg is a special case
and does not use content identifiers, S3 I assume can't get mixed up
like this, directory certainly has the problem, and adb also appears to
have had the problem.
Sponsored-by: Graham Spencer on Patreon
2022-09-20 17:15:31 +00:00
|
|
|
ia loc [cid] (fromRawFilePath tmpfile)
|
2022-05-09 19:38:21 +00:00
|
|
|
(Left k)
|
2020-09-28 19:03:15 +00:00
|
|
|
(combineMeterUpdate p' p)
|
2022-05-09 19:38:21 +00:00
|
|
|
ok <- moveAnnex k af tmpfile
|
2020-09-28 19:03:15 +00:00
|
|
|
when ok $
|
|
|
|
logStatus k InfoPresent
|
2022-05-09 19:38:21 +00:00
|
|
|
return (Just (k, ok))
|
2020-09-28 19:03:15 +00:00
|
|
|
checkDiskSpaceToGet k Nothing $
|
|
|
|
notifyTransfer Download af $
|
2021-02-03 19:35:32 +00:00
|
|
|
download' (Remote.uuid remote) k af Nothing stdRetry $ \p' ->
|
2020-09-28 19:03:15 +00:00
|
|
|
withTmp k $ downloader p'
|
2020-09-28 17:22:16 +00:00
|
|
|
|
|
|
|
-- The file is small, so is added to git, so while importing
|
|
|
|
-- without content does not retrieve annexed files, it does
|
|
|
|
-- need to retrieve this file.
|
|
|
|
doimportsmall cidmap db loc cid sz p = do
|
|
|
|
let downloader tmpfile = do
|
2022-05-09 19:38:21 +00:00
|
|
|
(k, _) <- Remote.retrieveExportWithContentIdentifier
|
change retrieveExportWithContentIdentifier to take a list of ContentIdentifier
This partly fixes an issue where there are duplicate files in the
special remote, and the first file gets swapped with another duplicate,
or deleted. The swap case is fixed by this, the deleted case will need
other changes.
This makes retrieveExportWithContentIdentifier take a list of allowed
ContentIdentifier, same as storeExportWithContentIdentifier,
removeExportWithContentIdentifier, and
checkPresentExportWithContentIdentifier.
Of the special remotes that support importtree, borg is a special case
and does not use content identifiers, S3 I assume can't get mixed up
like this, directory certainly has the problem, and adb also appears to
have had the problem.
Sponsored-by: Graham Spencer on Patreon
2022-09-20 17:15:31 +00:00
|
|
|
ia loc [cid] (fromRawFilePath tmpfile)
|
2022-05-09 19:38:21 +00:00
|
|
|
(Right (mkkey tmpfile))
|
2020-09-28 17:22:16 +00:00
|
|
|
p
|
|
|
|
case keyGitSha k of
|
|
|
|
Just sha -> do
|
|
|
|
recordcidkey cidmap db cid k
|
2020-09-28 19:03:15 +00:00
|
|
|
return sha
|
2020-09-28 17:22:16 +00:00
|
|
|
Nothing -> error "internal"
|
|
|
|
checkDiskSpaceToGet tmpkey Nothing $
|
|
|
|
withTmp tmpkey $ \tmpfile ->
|
|
|
|
tryNonAsync (downloader tmpfile) >>= \case
|
2020-09-28 19:03:15 +00:00
|
|
|
Right sha -> return $ Just (loc, Left sha)
|
2020-07-03 17:41:57 +00:00
|
|
|
Left e -> do
|
|
|
|
warning (show e)
|
|
|
|
return Nothing
|
2020-09-28 17:22:16 +00:00
|
|
|
where
|
|
|
|
tmpkey = importKey cid sz
|
|
|
|
mkkey tmpfile = gitShaKey <$> hashFile tmpfile
|
2020-07-03 17:41:57 +00:00
|
|
|
|
2020-09-28 19:03:15 +00:00
|
|
|
dodownload cidmap db (loc, (cid, sz)) f matcher = do
|
2020-09-04 17:49:57 +00:00
|
|
|
let af = AssociatedFile (Just f)
|
2020-05-15 16:51:09 +00:00
|
|
|
let downloader tmpfile p = do
|
2022-05-09 19:38:21 +00:00
|
|
|
(k, _) <- Remote.retrieveExportWithContentIdentifier
|
change retrieveExportWithContentIdentifier to take a list of ContentIdentifier
This partly fixes an issue where there are duplicate files in the
special remote, and the first file gets swapped with another duplicate,
or deleted. The swap case is fixed by this, the deleted case will need
other changes.
This makes retrieveExportWithContentIdentifier take a list of allowed
ContentIdentifier, same as storeExportWithContentIdentifier,
removeExportWithContentIdentifier, and
checkPresentExportWithContentIdentifier.
Of the special remotes that support importtree, borg is a special case
and does not use content identifiers, S3 I assume can't get mixed up
like this, directory certainly has the problem, and adb also appears to
have had the problem.
Sponsored-by: Graham Spencer on Patreon
2022-09-20 17:15:31 +00:00
|
|
|
ia loc [cid] (fromRawFilePath tmpfile)
|
2022-05-09 19:38:21 +00:00
|
|
|
(Right (mkkey tmpfile))
|
2020-09-04 17:49:57 +00:00
|
|
|
p
|
2020-06-23 20:07:18 +00:00
|
|
|
case keyGitSha k of
|
|
|
|
Nothing -> do
|
2020-11-16 18:09:55 +00:00
|
|
|
ok <- moveAnnex k af tmpfile
|
2020-06-23 20:07:18 +00:00
|
|
|
when ok $ do
|
|
|
|
recordcidkey cidmap db cid k
|
|
|
|
logStatus k InfoPresent
|
|
|
|
logChange k (Remote.uuid remote) InfoPresent
|
|
|
|
return (Right k, ok)
|
|
|
|
Just sha -> do
|
|
|
|
recordcidkey cidmap db cid k
|
|
|
|
return (Left sha, True)
|
2020-05-15 16:51:09 +00:00
|
|
|
let rundownload tmpfile p = tryNonAsync (downloader tmpfile p) >>= \case
|
2020-06-23 20:07:18 +00:00
|
|
|
Right (v, True) -> return $ Just (loc, v)
|
2020-05-15 16:51:09 +00:00
|
|
|
Right (_, False) -> return Nothing
|
|
|
|
Left e -> do
|
|
|
|
warning (show e)
|
|
|
|
return Nothing
|
2021-09-22 14:51:10 +00:00
|
|
|
let bwlimit = remoteAnnexBwLimit (Remote.gitconfig remote)
|
2019-03-08 16:33:44 +00:00
|
|
|
checkDiskSpaceToGet tmpkey Nothing $
|
2020-09-04 17:49:57 +00:00
|
|
|
notifyTransfer Download af $
|
2021-02-03 19:35:32 +00:00
|
|
|
download' (Remote.uuid remote) tmpkey af Nothing stdRetry $ \p ->
|
2020-09-04 17:49:57 +00:00
|
|
|
withTmp tmpkey $ \tmpfile ->
|
bwlimit
Added annex.bwlimit and remote.name.annex-bwlimit config that works for git
remotes and many but not all special remotes.
This nearly works, at least for a git remote on the same disk. With it set
to 100kb/1s, the meter displays an actual bandwidth of 128 kb/s, with
occasional spikes to 160 kb/s. So it needs to delay just a bit longer...
I'm unsure why.
However, at the beginning a lot of data flows before it determines the
right bandwidth limit. A granularity of less than 1s would probably improve
that.
And, I don't know yet if it makes sense to have it be 100ks/1s rather than
100kb/s. Is there a situation where the user would want a larger
granularity? Does granulatity need to be configurable at all? I only used that
format for the config really in order to reuse an existing parser.
This can't support for external special remotes, or for ones that
themselves shell out to an external command. (Well, it could, but it
would involve pausing and resuming the child process tree, which seems
very hard to implement and very strange besides.) There could also be some
built-in special remotes that it still doesn't work for, due to them not
having a progress meter whose displays blocks the bandwidth using thread.
But I don't think there are actually any that run a separate thread for
downloads than the thread that displays the progress meter.
Sponsored-by: Graham Spencer on Patreon
2021-09-21 20:58:02 +00:00
|
|
|
metered (Just p) tmpkey bwlimit $
|
2020-09-04 17:49:57 +00:00
|
|
|
const (rundownload tmpfile)
|
2019-02-27 17:15:02 +00:00
|
|
|
where
|
|
|
|
tmpkey = importKey cid sz
|
2020-09-04 17:49:57 +00:00
|
|
|
|
2020-09-28 19:03:15 +00:00
|
|
|
mkkey tmpfile = do
|
2020-09-28 17:22:16 +00:00
|
|
|
let mi = MatchingFile FileInfo
|
|
|
|
{ matchFile = f
|
2021-03-01 20:34:40 +00:00
|
|
|
, contentFile = tmpfile
|
2020-12-14 21:42:02 +00:00
|
|
|
, matchKey = Nothing
|
2020-09-28 17:22:16 +00:00
|
|
|
}
|
2020-12-21 20:03:27 +00:00
|
|
|
islargefile <- checkMatcher' matcher mi mempty
|
2020-09-28 17:22:16 +00:00
|
|
|
if islargefile
|
|
|
|
then do
|
2020-11-04 18:20:37 +00:00
|
|
|
backend <- chooseBackend f
|
2020-09-28 17:22:16 +00:00
|
|
|
let ks = KeySource
|
|
|
|
{ keyFilename = f
|
2020-11-03 22:34:27 +00:00
|
|
|
, contentLocation = tmpfile
|
2020-09-28 17:22:16 +00:00
|
|
|
, inodeCache = Nothing
|
|
|
|
}
|
|
|
|
fst <$> genKey ks nullMeterUpdate backend
|
|
|
|
else gitShaKey <$> hashFile tmpfile
|
2019-02-26 19:25:28 +00:00
|
|
|
|
2020-09-28 17:22:16 +00:00
|
|
|
ia = Remote.importActions remote
|
2019-02-27 17:15:02 +00:00
|
|
|
|
2020-09-04 17:49:57 +00:00
|
|
|
locworktreefile loc = fromRepo $ fromTopFilePath $ asTopFilePath $
|
|
|
|
case importtreeconfig of
|
|
|
|
ImportTree -> fromImportLocation loc
|
|
|
|
ImportSubTree subdir _ ->
|
|
|
|
getTopFilePath subdir P.</> fromImportLocation loc
|
2019-02-26 19:25:28 +00:00
|
|
|
|
|
|
|
getcidkey cidmap db cid = liftIO $
|
add RemoteStateHandle
This solves the problem of sameas remotes trampling over per-remote
state. Used for:
* per-remote state, of course
* per-remote metadata, also of course
* per-remote content identifiers, because two remote implementations
could in theory generate the same content identifier for two different
peices of content
While chunk logs are per-remote data, they don't use this, because the
number and size of chunks stored is a common property across sameas
remotes.
External special remote had a complication, where it was theoretically
possible for a remote to send SETSTATE or GETSTATE during INITREMOTE or
EXPORTSUPPORTED. Since the uuid of the remote is typically generate in
Remote.setup, it would only be possible to pass a Maybe
RemoteStateHandle into it, and it would otherwise have to construct its
own. Rather than go that route, I decided to send an ERROR in this case.
It seems unlikely that any existing external special remote will be
affected. They would have to make up a git-annex key, and set state for
some reason during INITREMOTE. I can imagine such a hack, but it doesn't
seem worth complicating the code in such an ugly way to support it.
Unfortunately, both TestRemote and Annex.Import needed the Remote
to have a new field added that holds its RemoteStateHandle.
2019-10-14 16:33:27 +00:00
|
|
|
CIDDb.getContentIdentifierKeys db rs cid >>= \case
|
2019-02-27 17:15:02 +00:00
|
|
|
[] -> atomically $
|
|
|
|
maybeToList . M.lookup cid <$> readTVar cidmap
|
2019-02-26 19:25:28 +00:00
|
|
|
l -> return l
|
|
|
|
|
|
|
|
recordcidkey cidmap db cid k = do
|
|
|
|
liftIO $ atomically $ modifyTVar' cidmap $
|
|
|
|
M.insert cid k
|
2021-10-06 21:05:32 +00:00
|
|
|
recordcidkey' db cid k
|
|
|
|
recordcidkey' db cid k = do
|
add RemoteStateHandle
This solves the problem of sameas remotes trampling over per-remote
state. Used for:
* per-remote state, of course
* per-remote metadata, also of course
* per-remote content identifiers, because two remote implementations
could in theory generate the same content identifier for two different
peices of content
While chunk logs are per-remote data, they don't use this, because the
number and size of chunks stored is a common property across sameas
remotes.
External special remote had a complication, where it was theoretically
possible for a remote to send SETSTATE or GETSTATE during INITREMOTE or
EXPORTSUPPORTED. Since the uuid of the remote is typically generate in
Remote.setup, it would only be possible to pass a Maybe
RemoteStateHandle into it, and it would otherwise have to construct its
own. Rather than go that route, I decided to send an ERROR in this case.
It seems unlikely that any existing external special remote will be
affected. They would have to make up a git-annex key, and set state for
some reason during INITREMOTE. I can imagine such a hack, but it doesn't
seem worth complicating the code in such an ugly way to support it.
Unfortunately, both TestRemote and Annex.Import needed the Remote
to have a new field added that holds its RemoteStateHandle.
2019-10-14 16:33:27 +00:00
|
|
|
liftIO $ CIDDb.recordContentIdentifier db rs cid k
|
|
|
|
CIDLog.recordContentIdentifier rs cid k
|
|
|
|
|
|
|
|
rs = Remote.remoteStateHandle remote
|
2019-02-27 17:15:02 +00:00
|
|
|
|
|
|
|
{- Temporary key used for import of a ContentIdentifier while downloading
|
|
|
|
- content, before generating its real key. -}
|
|
|
|
importKey :: ContentIdentifier -> Integer -> Key
|
2019-11-22 20:24:04 +00:00
|
|
|
importKey (ContentIdentifier cid) size = mkKey $ \k -> k
|
2020-06-11 20:07:36 +00:00
|
|
|
{ keyName = genKeyName (decodeBS cid)
|
2019-02-27 17:15:02 +00:00
|
|
|
, keyVariety = OtherKey "CID"
|
|
|
|
, keySize = Just size
|
|
|
|
}
|
2019-05-20 20:37:04 +00:00
|
|
|
|
|
|
|
{-- Export omits non-preferred content from the tree stored on the
|
|
|
|
-- remote. So the import will normally have that content
|
|
|
|
-- omitted (unless something else added files with the same names to the
|
|
|
|
-- special remote).
|
|
|
|
--
|
|
|
|
-- That presents a problem: Merging the imported tree would result
|
2019-05-21 18:38:00 +00:00
|
|
|
-- in deletion of the files that were excluded from export.
|
|
|
|
-- To avoid that happening, this adds them back to the imported tree.
|
2019-05-20 20:37:04 +00:00
|
|
|
--}
|
2019-05-21 18:38:00 +00:00
|
|
|
addBackExportExcluded :: Remote -> Sha -> Annex Sha
|
|
|
|
addBackExportExcluded remote importtree =
|
2019-05-20 20:37:04 +00:00
|
|
|
getExportExcluded (Remote.uuid remote) >>= \case
|
|
|
|
[] -> return importtree
|
|
|
|
excludedlist -> inRepo $
|
|
|
|
adjustTree
|
|
|
|
-- don't remove any
|
|
|
|
(pure . Just)
|
|
|
|
excludedlist
|
|
|
|
-- if something was imported with the same
|
|
|
|
-- name as a file that was previously
|
|
|
|
-- excluded from import, use what was imported
|
|
|
|
(\imported _excluded -> imported)
|
|
|
|
[]
|
|
|
|
importtree
|
2019-05-21 18:38:00 +00:00
|
|
|
|
|
|
|
{- Match the preferred content of the remote at import time.
|
|
|
|
-
|
|
|
|
- Only keyless tokens are supported, because the keys are not known
|
|
|
|
- until an imported file is downloaded, which is too late to bother
|
|
|
|
- excluding it from an import.
|
|
|
|
-}
|
|
|
|
makeImportMatcher :: Remote -> Annex (Either String (FileMatcher Annex))
|
|
|
|
makeImportMatcher r = load preferredContentKeylessTokens >>= \case
|
|
|
|
Nothing -> return $ Right matchAll
|
|
|
|
Just (Right v) -> return $ Right v
|
|
|
|
Just (Left err) -> load preferredContentTokens >>= \case
|
|
|
|
Just (Left err') -> return $ Left err'
|
|
|
|
_ -> return $ Left $
|
|
|
|
"The preferred content expression contains terms that cannot be checked when importing: " ++ err
|
|
|
|
where
|
|
|
|
load t = M.lookup (Remote.uuid r) . fst <$> preferredRequiredMapsLoad' t
|
|
|
|
|
2020-09-30 14:10:03 +00:00
|
|
|
{- Gets the ImportableContents from the remote.
|
|
|
|
-
|
|
|
|
- Filters out any paths that include a ".git" component, because git does
|
|
|
|
- not allow storing ".git" in a git repository. While it is possible to
|
|
|
|
- write a git tree that contains that, git will complain and refuse to
|
|
|
|
- check it out.
|
|
|
|
-
|
2020-09-30 14:41:59 +00:00
|
|
|
- Filters out new things not matching the FileMatcher or that are
|
|
|
|
- gitignored. However, files that are already in git get imported
|
|
|
|
- regardless. (Similar to how git add behaves on gitignored files.)
|
|
|
|
- This avoids creating a remote tracking branch that, when merged,
|
|
|
|
- would delete the files.
|
2020-12-22 18:20:11 +00:00
|
|
|
-
|
|
|
|
- Throws exception if unable to contact the remote.
|
2020-12-22 18:35:02 +00:00
|
|
|
- Returns Nothing when there is no change since last time.
|
2020-09-30 14:10:03 +00:00
|
|
|
-}
|
2021-10-06 21:05:32 +00:00
|
|
|
getImportableContents :: Remote -> ImportTreeConfig -> CheckGitIgnore -> FileMatcher Annex -> Annex (Maybe (ImportableContentsChunkable Annex (ContentIdentifier, ByteSize)))
|
2020-12-22 18:20:11 +00:00
|
|
|
getImportableContents r importtreeconfig ci matcher = do
|
2020-12-22 18:35:02 +00:00
|
|
|
Remote.listImportableContents (Remote.importActions r) >>= \case
|
2021-10-06 21:05:32 +00:00
|
|
|
Just (ImportableContentsComplete ic) -> do
|
|
|
|
dbhandle <- opendbhandle
|
|
|
|
Just . ImportableContentsComplete
|
|
|
|
<$> filterunwanted dbhandle ic
|
|
|
|
Just (c@(ImportableContentsChunked {})) -> do
|
|
|
|
dbhandle <- opendbhandle
|
|
|
|
Just <$> filterunwantedchunked dbhandle c
|
2020-12-22 18:35:02 +00:00
|
|
|
Nothing -> return Nothing
|
2019-05-21 18:38:00 +00:00
|
|
|
where
|
2020-09-30 14:10:03 +00:00
|
|
|
filterunwanted dbhandle ic = ImportableContents
|
|
|
|
<$> filterM (wanted dbhandle) (importableContents ic)
|
|
|
|
<*> mapM (filterunwanted dbhandle) (importableHistory ic)
|
|
|
|
|
2021-10-06 21:05:32 +00:00
|
|
|
filterunwantedchunked dbhandle c = ImportableContentsChunked
|
|
|
|
<$> filterunwantedchunk dbhandle (importableContentsChunk c)
|
|
|
|
<*> mapM (filterunwanted dbhandle) (importableHistoryComplete c)
|
|
|
|
|
|
|
|
filterunwantedchunk dbhandle c = ImportableContentsChunk
|
|
|
|
<$> pure (importableContentsSubDir c)
|
|
|
|
<*> filterM (wantedunder dbhandle (importableContentsSubDir c))
|
|
|
|
(importableContentsSubTree c)
|
|
|
|
<*> pure (
|
|
|
|
importableContentsNextChunk c >>= \case
|
|
|
|
Nothing -> return Nothing
|
|
|
|
Just c' -> Just <$> filterunwantedchunk dbhandle c'
|
|
|
|
)
|
|
|
|
|
|
|
|
opendbhandle = Export.openDb (Remote.uuid r)
|
|
|
|
|
2020-09-30 14:10:03 +00:00
|
|
|
wanted dbhandle (loc, (_cid, sz))
|
avoid git check-ignore overhead on importing known files
isKnownImportLocation does a database lookup and there's an index
to make that lookup fast, so it's probably faster than talking to git
check-ignore. Checking the matcher is faster still.
While before the gitignore check was added it did not need to always
check isknown, now it does, because it's that or the more expensive
notignored. But at least we can skip notignored when a file is known,
which will often be the common case: Importing from a remote that's been
exported to, and/or imported from before, only new files will not be
known, so only those will need to check notignored.
At first, I had this:
(matches <&&> (isknown <||> notignored)) <||> isknown
Notice that checks isknown every time, whether it matches or not.
So, it's no slower to instead do this:
isknown <||> (matches <&&> notignored)
That has the benefit that, when it's known, it doesn't need to run
matches, which while faster than isknown, is still going to use some CPU.
And it perhaps more clearly expresses the condition: Any known file is
wanted, otherwise it's down to what matches and is not ignored.
This commit was sponsored by Jack Hill on Patren.
2020-09-30 15:09:09 +00:00
|
|
|
| ingitdir = pure False
|
|
|
|
| otherwise =
|
|
|
|
isknown <||> (matches <&&> notignored)
|
|
|
|
where
|
|
|
|
-- Checks, from least to most expensive.
|
|
|
|
ingitdir = ".git" `elem` Posix.splitDirectories (fromImportLocation loc)
|
|
|
|
matches = matchesImportLocation matcher loc sz
|
|
|
|
isknown = isKnownImportLocation dbhandle loc
|
|
|
|
notignored = notIgnoredImportLocation importtreeconfig ci loc
|
2021-10-06 21:05:32 +00:00
|
|
|
|
|
|
|
wantedunder dbhandle root (loc, v) =
|
|
|
|
wanted dbhandle (importableContentsChunkFullLocation root loc, v)
|
2019-05-21 18:38:00 +00:00
|
|
|
|
2020-09-30 14:41:59 +00:00
|
|
|
isKnownImportLocation :: Export.ExportHandle -> ImportLocation -> Annex Bool
|
|
|
|
isKnownImportLocation dbhandle loc = liftIO $
|
|
|
|
not . null <$> Export.getExportTreeKey dbhandle loc
|
|
|
|
|
avoid git check-ignore overhead on importing known files
isKnownImportLocation does a database lookup and there's an index
to make that lookup fast, so it's probably faster than talking to git
check-ignore. Checking the matcher is faster still.
While before the gitignore check was added it did not need to always
check isknown, now it does, because it's that or the more expensive
notignored. But at least we can skip notignored when a file is known,
which will often be the common case: Importing from a remote that's been
exported to, and/or imported from before, only new files will not be
known, so only those will need to check notignored.
At first, I had this:
(matches <&&> (isknown <||> notignored)) <||> isknown
Notice that checks isknown every time, whether it matches or not.
So, it's no slower to instead do this:
isknown <||> (matches <&&> notignored)
That has the benefit that, when it's known, it doesn't need to run
matches, which while faster than isknown, is still going to use some CPU.
And it perhaps more clearly expresses the condition: Any known file is
wanted, otherwise it's down to what matches and is not ignored.
This commit was sponsored by Jack Hill on Patren.
2020-09-30 15:09:09 +00:00
|
|
|
matchesImportLocation :: FileMatcher Annex -> ImportLocation -> Integer -> Annex Bool
|
|
|
|
matchesImportLocation matcher loc sz = checkMatcher' matcher mi mempty
|
2019-06-04 19:14:20 +00:00
|
|
|
where
|
2020-09-30 14:10:03 +00:00
|
|
|
mi = MatchingInfo $ ProvidedInfo
|
2021-03-02 16:47:23 +00:00
|
|
|
{ providedFilePath = Just (fromImportLocation loc)
|
2020-09-30 14:10:03 +00:00
|
|
|
, providedKey = Nothing
|
2021-03-02 16:47:23 +00:00
|
|
|
, providedFileSize = Just sz
|
2020-09-30 14:10:03 +00:00
|
|
|
, providedMimeType = Nothing
|
|
|
|
, providedMimeEncoding = Nothing
|
2021-03-02 16:47:23 +00:00
|
|
|
, providedLinkType = Nothing
|
2019-06-04 19:14:20 +00:00
|
|
|
}
|
avoid git check-ignore overhead on importing known files
isKnownImportLocation does a database lookup and there's an index
to make that lookup fast, so it's probably faster than talking to git
check-ignore. Checking the matcher is faster still.
While before the gitignore check was added it did not need to always
check isknown, now it does, because it's that or the more expensive
notignored. But at least we can skip notignored when a file is known,
which will often be the common case: Importing from a remote that's been
exported to, and/or imported from before, only new files will not be
known, so only those will need to check notignored.
At first, I had this:
(matches <&&> (isknown <||> notignored)) <||> isknown
Notice that checks isknown every time, whether it matches or not.
So, it's no slower to instead do this:
isknown <||> (matches <&&> notignored)
That has the benefit that, when it's known, it doesn't need to run
matches, which while faster than isknown, is still going to use some CPU.
And it perhaps more clearly expresses the condition: Any known file is
wanted, otherwise it's down to what matches and is not ignored.
This commit was sponsored by Jack Hill on Patren.
2020-09-30 15:09:09 +00:00
|
|
|
|
|
|
|
notIgnoredImportLocation :: ImportTreeConfig -> CheckGitIgnore -> ImportLocation -> Annex Bool
|
|
|
|
notIgnoredImportLocation importtreeconfig ci loc = not <$> checkIgnored ci f
|
|
|
|
where
|
2020-11-03 22:34:27 +00:00
|
|
|
f = case importtreeconfig of
|
2020-09-30 14:41:59 +00:00
|
|
|
ImportSubTree dir _ ->
|
|
|
|
getTopFilePath dir P.</> fromImportLocation loc
|
|
|
|
ImportTree ->
|
|
|
|
fromImportLocation loc
|