summaryrefslogtreecommitdiff
path: root/bugs/restore_from_gzip__39__d_repo_gives_confusing_error_message.mdwn
diff options
context:
space:
mode:
Diffstat (limited to 'bugs/restore_from_gzip__39__d_repo_gives_confusing_error_message.mdwn')
-rw-r--r--bugs/restore_from_gzip__39__d_repo_gives_confusing_error_message.mdwn44
1 files changed, 0 insertions, 44 deletions
diff --git a/bugs/restore_from_gzip__39__d_repo_gives_confusing_error_message.mdwn b/bugs/restore_from_gzip__39__d_repo_gives_confusing_error_message.mdwn
deleted file mode 100644
index 50c5016..0000000
--- a/bugs/restore_from_gzip__39__d_repo_gives_confusing_error_message.mdwn
+++ /dev/null
@@ -1,44 +0,0 @@
-Create repository using:
-obnam --compress-with=gzip --repository --repository /tmp/backup/ backup "$HOME/"
-
-Try to restore:
-obnam restore --repository /tmp/backup/ --to=/tmp/restore "$HOME/test"
-
-Error message:
-ERROR: Invalid repository format version ('x\x9c3\xe5\x02\x00\x00v\x00@') -- forgot encryption?
-
-Either the compression used should be autodetected (maybe saved with metadata?) or the error message should be adjusted (using obnam 0.25)
-
---xeen
-
----
-
-I confirm that I see the bug. And I agree that Obnam should automatically detect
-when compression has been used and that the error message is awful.
-
----
-
-
-Proposed solution (involving re-formatting the repository, for fun and profit)
-
-1. Alter the metadata store to not be filtered
- * Alter Repository to not use encryption filter in the metadata files
- * Alter HookedFS to allow you to ask that data filters be bypassed
- * Alter Repository to ask for bypass on metadata files (format)
-2. Alter the data filtering hooks to work as follows:
- * When filtering for write, the output data format should always be filtername\0data. This chains for each write filter.
- * When filtering for read, we read the filter name, apply its inverse on the data and then repeat, chaining until we read \0data (an empty filter name)
-
-Side effects:
-
-* Ordering of filters can change at will, because reading data is always deterministic on the stored data.
-* Data is leaked about (at best) the top level filter (which might be encryption).
- * Note: Encrypted repositories are obvious anyway so this should not be considered a blocker
-* All files increase in size by some number of (at minimum 1) bytes.
-* Repository metadata is now not affected by filters, so we can always guarantee to be able to read it and produce useful warnings/errors in case of format mismatches.
-
----
-
-Should be fixed in repository format 6
-
-[[done]]