[ https://issues.apache.org/jira/browse/COUCHDB-1607?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13499782#comment-13499782 ] Robert Newson commented on COUCHDB-1607: ---------------------------------------- Nathan, nice sleuthing. If the HashPart (aka _C) begins with a ":" we'll get a verification failure, the same is true for a HashPart with any two or more consecutive ":"'s, for the same reason. All because string:tokens silently omits empty tokens, rather than leaving a spacer. The re:split/2 method fixes that neatly. The list/binary stuff is annoying but not strictly related. And here's the difference itself; string:tokens("User:Time:Foo::Bar",":") -> ["User","Time","Foo","Bar"] re:split("User:Time:Foo::Bar",":") -> [<<"User">>,<<"Time">>,<<"Foo">>,<<>>,<<"Bar">>] That's a hash part that, by chance, has two adjacent colon characters. Since it's silently dropped in the first case, a subsequent join with ":" will not produce the original string. boom. > cookie_authentication_handler does not properly handle AuthSession signatures starting with ":" character(s) > ------------------------------------------------------------------------------------------------------------ > > Key: COUCHDB-1607 > URL: https://issues.apache.org/jira/browse/COUCHDB-1607 > Project: CouchDB > Issue Type: Bug > Affects Versions: 1.2 > Reporter: Nathan Vander Wilt > Priority: Critical > > AuthSession cookies will intermittently "break" — a user will have a perfectly valid session, but suddenly after their cookie gets refreshed they can randomly get "logged out" for practical purposes. > The cause is that Erlang's `string:tokens` behaviour does act as this code expects: > https://github.com/apache/couchdb/blob/7d4181346626c0cdb50b44f7e5e33435a8ccae0f/src/couchdb/couch_httpd_auth.erl#L163 > https://github.com/apache/couchdb/blob/7d4181346626c0cdb50b44f7e5e33435a8ccae0f/src/couchdb/couch_httpd_auth.erl#L183 > After evaluating `[A,B | C] = string:tokens("a:b::c:d").` the value of C is not `["","c","d"]` but rather `["c","d"]`. So when rejoined, the signature becomes "c:d" instead of the original ":c:d"! > It appears that using re.split/2 would avoid this problem, but yields a list/array result containing <<"">> instead of "" types, which string:join does not like. Should be pretty quick fix though for someone who knows just a bit more of the Erlang way. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira