Diffstat (limited to 'Docs')
1 files changed, 19 insertions, 9 deletions
diff --git a/Docs/source-object-backing-store b/Docs/source-object-backing-store
index e55a99db3..5d4d3049d 100644
@@ -34,8 +34,8 @@ As the backing store only holds cache data one should not expect a
great deal of effort to be expended converting formats (i.e. the cache
may simply be discarded).
-Layout version 1
+Layout version 1.1
An object has an identifier value generated from the url (NetSurf
backing stores uses the url as the unique key). The value used is
@@ -54,7 +54,7 @@ overhead of reconstructing it at initialisation and to keep the data
used to improve the eviction decisions.
Each object is stored and retrived directly into the filesystem using
-a filename generated from a base64url encoding of an address
+a filename generated from a RFC4648 base32 encoding of an address
value. The objects address is derived from the identifier by cropping
it to a shorter length.
@@ -71,17 +71,27 @@ E.g. Linux based systems can easily cope with several megabytes of
mmaped index but RISC OS might want to limit this to a few megabytes
of heap at most.
-The files are stored on disc using their base64url address value.
+The files are stored on disc using their base32 address value.
By creating a directory for each character of the encoded filename
(except the last which is of course the leafname) we create a
-directory structure where no directory has more than 64 entries.
+directory structure where no directory has more than 32 entries.
-E.g. A 19bit address of 0x1 would be base64url encoded into AAAB
+E.g. A 19bit address of 0x1 would be base32 encoded into AAAB
resulting in the data being stored in a file path of
An address of 0x00040001 encodes to BAAB and a file path of
+The version 1 layout was identical to the 1.1 except base64url
+encoding was used, this proved problematic as some systems filesystems
+were case insensitive so upper and lower case letetrs collided.
+There is no upgrade provision from the previous version simply delete
+the cache directory.
@@ -99,7 +109,7 @@ filesystem.
Each control file table entry is 28 bytes and consists of
- - signed 64 but value for last use time
+ - signed 64 bit value for last use time
- 32bit full url hash allowing for index reconstruction and
addiitonal collision detection. Also the possibility of increasing