couchdb-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Adam Kocoloski (JIRA)" <j...@apache.org>
Subject [jira] Commented: (COUCHDB-196) Bug in _all_docs: extra incorrect row returned when ?key= is specified
Date Fri, 09 Jan 2009 03:36:59 GMT

    [ https://issues.apache.org/jira/browse/COUCHDB-196?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12662235#action_12662235
] 

Adam Kocoloski commented on COUCHDB-196:
----------------------------------------

Hi Bob, thanks for looking into this.  I'd argue for replacing that entire function with couch_view:less_json/2

diff --git a/src/couchdb/couch_db_updater.erl b/src/couchdb/couch_db_updater.erl
index 049a354..38885f8 100644
--- a/src/couchdb/couch_db_updater.erl
+++ b/src/couchdb/couch_db_updater.erl
@@ -253,13 +253,8 @@ init_db(DbName, Filepath, Fd, Header0) ->
     Header = simple_upgrade_record(Header0, #db_header{}),
     {ok, SummaryStream} = couch_stream:open(Header#db_header.summary_stream_state, Fd),
     ok = couch_stream:set_min_buffer(SummaryStream, 10000),
-    Less =
-        fun(A,B) when A==B -> false;
-        (nil, _) -> true; % nil - special key sorts before all
-        ({}, _) -> false; % {} -> special key sorts after all
-        (A, B) -> A < B
-        end,
-            
+    Less = fun couch_view:less_json/2,
+    
     {ok, IdBtree} = couch_btree:open(Header#db_header.fulldocinfo_by_id_btree_state, Fd,
         [{split, fun(X) -> btree_by_id_split(X) end},
         {join, fun(X,Y) -> btree_by_id_join(X,Y) end},


The DB behavior seems to be the same either way, and the fewer places we have this collaction
algorithm coded up the better.  I looked into the 2 test suite failures that show up as a
result of applying either of these patches.  The failure in design docs occurs because the
test suite expects to find the document "_design/test" in the _all_docs view between "_design%2F"
and "_design%2FZZZ".  That expectation is erroneous.  I'm not sure why it worked before.

The second failure occurs in view_include_docs where an _all_docs view is queried with limit=2&skip=1.
 In the trunk code _design docs sort after docs with stringified integers for ids.  Your patch
causes the _design docs to sort first, which I believe is the correct behavior.  Again, the
test suite's expectation is probably wrong.


> Bug in _all_docs: extra incorrect row returned when ?key= is specified
> ----------------------------------------------------------------------
>
>                 Key: COUCHDB-196
>                 URL: https://issues.apache.org/jira/browse/COUCHDB-196
>             Project: CouchDB
>          Issue Type: Bug
>          Components: HTTP Interface
>            Reporter: Jason Davies
>         Attachments: collation_ids.diff, collation_test.2.diff, collation_test.diff,
couch_db_updater.diff, view_coll.js
>
>
> I first noticed this when attempting to reverse the order of docs displayed in Futon,
when listing the design docs.  When the order is reversed, non-design docs are displayed.
> After digging deeper I discovered that querying _all_docs with key="Z" can potentially
return two rows, with keys "Z" and "a".
> I will attach a test case to reproduce this.  My locale is set to en_GB.UTF-8, which
could be related to this problem.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message