Testing equality between svnrdump and svnadmin dump
From: Julian Foad <julianfoad_at_btopenworld.com>
Date: Fri, 9 Jan 2015 18:18:58 +0000
For Your Interest,
I am having some success with testing for equality between the outputs of svnrdump and svnadmin dump.
The attached patch makes the test suite run both 'svnadmin dump' and 'svnrdump dump' on the sandbox repository (if created) after each test, and compare the results. This is just one way of getting access to a moderately wide range of different repository contents. It's not the only way.
The few differences I have found seem to be already known (but not necessarily adequately addressed). These include: svnrdump always sets 'Prop-delta: true' even when unnecessary; svnrdump doesn't output some of the checksum headers that svnadmin does; they put different numbers of blank lines between sections in some cases.
In its current state, about 4 or 5 tests fail. A few of these are svnadmin tests that deliberately create an invalid repo, that then fails to dump. At least one other fail appears to be a bug in the test suite's dumpfile parser. That parser needs some more work. For now, the patch changes the test suite's parsing of headers in ways that make it more generic (happier to read anything) and does less checking of the expected sequence (so in this respect it reduces an aspect of the test coverage that is used for some existing tests).
I'm not planning to commit this in its current state.
Mostly I'm planning to use it to help me check for mistakes while I deduplicate the dumpfile handling code in svnadmin, svnrdump and svndumpfilter.
This is an archived mail posted to the Subversion Dev mailing list.