From: Tim Bunce <Tim.Bunce@ig.co.uk>
Date: Tue, 16 Jun 2009 13:40:11 +0000 (+0000)
Subject: Reverted doc patch r6689 for now, sadly. I'll open a ticket to explain.
X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=commitdiff_plain;h=05075aeee9eee4070919861a119a8f582d0feb48;p=dbsrgits%2FDBIx-Class-Historic.git

Reverted doc patch r6689 for now, sadly. I'll open a ticket to explain.
---

diff --git a/lib/DBIx/Class/Storage.pm b/lib/DBIx/Class/Storage.pm
index 799a6af..bf50762 100644
--- a/lib/DBIx/Class/Storage.pm
+++ b/lib/DBIx/Class/Storage.pm
@@ -168,8 +168,14 @@ In a nested transaction (calling txn_do() from within a txn_do() coderef) only
 the outermost transaction will issue a L</txn_commit>, and txn_do() can be
 called in void, scalar and list context and it will behave as expected.
 
-Unlike L</DBIx::Class::Storage/dbh_do>, the coderef will I<not> be
-automatically retried on error.
+Please note that all of the code in your coderef, including non-DBIx::Class
+code, is part of a transaction.  This transaction may fail out halfway, or
+it may get partially double-executed (in the case that our DB connection
+failed halfway through the transaction, in which case we reconnect and
+restart the txn).  Therefore it is best that any side-effects in your coderef
+are idempotent (that is, can be re-executed multiple times and get the
+same result), and that you check up on your side-effects in the case of
+transaction failure.
 
 =cut