4 patches for repository http://darcs.net/screened:
patch 2fc2dca9ee0276841de630cb81e3ca0c944d650f
Author: Ben Franksen <ben.franksen@online.de>
Date: Sat Dec 26 18:45:09 CET 2020
* add a newline to the sources file when cloning
patch 9bc126b81eee351e1f5db297735746e2be00bdbf
Author: Ben Franksen <ben.franksen@online.de>
Date: Tue Jan 5 09:39:02 CET 2021
* clean up merging of local and remote sources when cloning
This concentrates the logic that determines the new cache/sources when we
clone a repo into into the unionRemoteCaches function: if the remote repo is
not a local path, then we add its location as a non-writable repo cache
entry. Furthermore, the freshly created _toRepo has no sources file anyway,
so using appendTextFile makes no sense.
patch eb3da80ca5a6d075f01cfcee0b1c07d51705f746
Author: Ben Franksen <ben.franksen@online.de>
Date: Mon Jan 11 01:37:20 CET 2021
* unionRemoteCaches -> filterRemoteCaches
We called unionRemoteCaches with the cache values taken from the source and
the target repo. Both are the result of calling getCaches. Now, getCaches
does a lot more than just reading a single sources file. If called with Just
a remote repo location as second parameter, it merges caches from 6
different sources: the local repo (at "."), the contents of its sources
file, the remote repo, the contents of that repo's sources file, the global
sources file, and (unless --no-cache is in effect) the global cache. This
means that merging the local and the remote caches is redundant: the remote
cache already contains all the cache locations of the local cache. This
motivates removing the first argument and renaming the function.
Furthermore, some of the design decisions for unionRemoteCaches were, I
think, misguided. So I also made a semantic change here: the function no
longer removes network entries for network accessed remote repos. This means
that we can remove the third parameter, too. My rationale is this: a network
accessed repo can contain locations in its sources that are essential (that
is, if the remote repo is itself a lazy clone). Indeed the docs admitted
that this case isn't handled well. It is true that there is no guarantee
that remote source repos can be accessed from the local user/machine. But
this is what 'checkCacheReachability' is for: it blacklists inaccessible
entries for the current run and advises the user to remove these entries
from their sources file. IME this scheme works pretty well in practice.
patch e1a2cbf2ea70af254552c2d137ba8e958c0e9f10
Author: Ben Franksen <ben.franksen@online.de>
Date: Sun Jan 10 10:48:57 CET 2021
* mkCache: first sort by locatily, then nub
This is to account for the non-standard Eq instance of CacheLoc. The whole
point of making Cache abstract and providing a smart constructor is that we
want the result to be independent of the order of the CacheLoc entries in
the input list. If we first nub and then sort, then the original order
matters: it determines whether in the result two equal (==) CacheLoc will be
Writable or NotWritable.
Attachments
|
