* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * THIS RELEASE STREAM IS OPEN TO BUG FIXES. * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * This file tracks the status of releases in the 1.8.x line. See http://subversion.apache.org/docs/community-guide/releasing.html#release-stabilization for details on how release lines and voting work, what kinds of bugs can delay a release, etc. Status of 1.8.1: Candidate changes: ================== * r1469519, r1469520, r1481848, r1481870 Silence compiler warnings, tighten scopes and improve code. Justification: Clean builds make it more likely to notice real compiler warnings. Branch: ^/subversion/branches/1.8.x-r1469519-and-friends Votes: +1: blair, brane -0: philip (see dev thread: false alarm, trades one warning for another) * r1498000 Improve error handling for inherited property requests by using the transition based xml parser in libsvn_ra_serf. Justification: This maintenance work which usually doesn't belong in a patch release makes it possible to properly detect if this request should be restarted on non-chunk-enabled proxies. Votes: +1: rhuijben, ivan * r1496127 Fix occasional failure of checkout_tests.py 12. Justification: Spurious FAILs confuse people testing the release. Votes: +1: philip, rhuijben -0: danielsh (loses functionality; patch sent with alternative fix) stsp: where exactly is the alternative patch? danielsh: http://mid.gmane.org/20130705040533.GL3244@lp-shahaf.local * r1501199, r1501207, r1502777 Verify the result of the xml parser in ra serf, instead of assuming the parsing always succeeds. Justification: While in general we can just rely on the parsed result being incomplete, this makes many network (and server implementation) problems easier to diagnose. Votes: +1: rhuijben * r1501293, r1501338, r1501360, r1501533 Fix issue #4390, "parallel nested checkout not possible with wc-ng" Justification: Regression from 1.6. User complained. See issue for details. Branch: ^/subversion/branches/1.8.x-issue4390 Votes: +1: stsp +1: rhuijben (for 1.8.2+) -0: ivan (not regression from 1.7.x, destructive change in libsvn_wc core) * r1501656, r1501702 Fix the ability to build Debug Build with BDB FS on Windows. Justification: Our build system should work properly. Votes: +1: breser, stefan2 -0: rhuijben (Not a regression and I don't think we should encourage building with just a debug BDB (even for release mode) by backporting this change. That would require using the non-distributable debug CRT. Trunk has a cleaner detection of BDB (via db.h) and even while debugging a maintainer build of Subversion a release BDB works fine) * r1502267 Enforce strict version equality between cmdline tools and libraries. Justification: Without this, a 1.8.1 client will use a 1.8.2 libsvn_* library if that's what the runtime linker finds. This may be a (hard-to-diagnose) problem due to possible incompatibilities in the subversion/include/private/ APIs used. Branch: 1.8.x-r1502267 Notes: Seen in the wild under 1.7.x. Votes: +1: danielsh * r1502901 Make ra_serf sessions reusable after a log callback uses the SVN_ERR_CEASE_INVOCATION error. Justification: Without this patch a new session would be required for each of these usages. Required by the r1468980 group. Notes: Resets the connection used by ra_serf to avoid having to read all data, which is mostly the same behavior as not re-using the ra session. Votes: +1: rhuijben, stefan2 * r1468980, r1478987, r1502811 Reduce the number of ra sessions for automatic merge calculations to two. Justification: 10 concurrent ra sessions open to a single server at the same time breaks in certain setups (e.g. xinetd svnserve), while we really don't need this many. Opening ra session is still a performance killer in all other scenarios, so reducing the number of sessions is a good thing. Notes: Depends on r1502901 and r1502909, to avoid ra-session reuse errors. r1468980 is a small performance tweak needed for a clean merge. r1478987 removes a common ra session. r1502811 avoids all ra session creations via svn_client__mergeinfo_log. Votes: +1: rhuijben (for 1.8.2) +1: stefan2 * r1503010, r1503058, r1503061, r1503211 svn_utf_*: Describe APR errors as coming from APR. Justification: It is not obvious what needs to be recompiled to resolve "This function has not been implemented on this platform". We do the same for SQLite. . The current code will not enter the error path in the ENOTIMPL case; that is being discussed on dev@ still. The "Justification" field uses that error code is used as an example only. Votes: +1: danielsh, rhuijben +1 (without r1503211): stefan2 * r1503318 ra_serf: Do not retry HTTP requests if we started to parse them. Discussion: http://svn.haxx.se/dev/archive-2013-07/0290.shtml Justification: There is absolutely no guarantee that REPORT or PROPFIND responses will be equal for every request. Before r1503318 after request we skipped recieved N bytes of original request response and continue with remaining bytes from retried request. Votes: +1: ivan +0: danielsh Veto-blocked changes: ===================== * r1501049, r1501371 ra_serf: Return Subversion error codes in svn_error_t. Justification: Other error codes are useless to API users. Votes: +1: danielsh -1: rhuijben (We also report apr and OS statuses, which are certainly not useless, as GUI clients depend on this extensive error codes to provide proper workarounds. This use-case should be fixed by directly setting the right message in the error chain when leaving ra_serf. This adds up to a -0 for me. The reason for the -1 is the re-use of a specific error code that is automatically unwrapped in some ra_serf code, to avoid handling codes like APR_EOF as non fatal [ without r1501371 ] : Responded in detail in the r1501049 thread. tldr: outermost svn_error_t->apr_err should be defined by Subversion or by APR, not by serf; can't find evidence of the "automatic unwrapping", and in any case happy to change the error code to something else if that's the problem.) -1: gstein (apr_status_t was intended/designed to hold error codes from multiple apps/libs; check for known values, and propagate all unkown values) Approved changes: =================