On 25.08.2015 23:12, Stefan Fuhrmann wrote: > On Tue, Aug 25, 2015 at 4:43 PM, Branko Čibej <br...@wandisco.com> wrote: > >> On 25.08.2015 17:31, Stefan Fuhrmann wrote: >>> On Tue, Aug 25, 2015 at 12:55 PM, Branko Čibej <br...@wandisco.com> >> wrote: >>>> On 25.08.2015 13:49, br...@apache.org wrote: >>>>> Author: brane >>>>> Date: Tue Aug 25 11:49:09 2015 >>>>> New Revision: 1697654 >>>>> >>>>> URL: http://svn.apache.org/r1697654 >>>>> Log: >>>>> * branches/1.9.x/STATUS: >>>>> - Approve r1693886. >>>>> - Temporarily veto r1694481; the change looks broken. >>>> [...] >>>> >>>>> @@ -98,5 +84,22 @@ Candidate changes: >>>>> Veto-blocked changes: >>>>> ===================== >>>>> >>>>> + * r1694481 >>>>> + Fix Unix build on systems without GPG agent. >>>>> + Justification: >>>>> + This is a user-reported issue. >>>>> + Votes: >>>>> + +1: stefan2, philip >>>>> + -1: brane (You can't just remove a public API implementation, >>>>> + even if it is deprecated. And the prototyps is still >>>>> + right there in svn_auth.h) >>>>> + >>>>> Approved changes: >>>>> ================= >>>> r1694481 (conditionally) removes the implementation of a public API, >>>> whilst leaving the prototype in svn_auth.h untouched. This is a >>>> violation of our ABI compatibility rules, and also a linking error >>>> waiting to happen. >>>> >>> Except that the very problem is that >>> svn_auth__get_gpg_agent_simple_provider >>> is not implemented either if SVN_HAVE_GPG_AGENT >>> is not defined. And that linker problem is the one being >>> already reported and fixed by the patch. >>> >>> You are still right that we introduce another linker problem >>> further down the road for some people that stumbled >>> across the first one in the past. And not implementing >>> the public API is a bad thing. >>> >>> So, I think we need to do some coding to fix this on /trunk. >>> Question is whether we want to skip r1694481 as a stop- >>> gap patch for 1.9.1 and enable people to build SVN again. >> >> Daniel suggested inserting a dummy handler if we don't have the GPG >> agent support. I think that may be the only reasonable solution for both >> trunk and 1.9.1 (or .x if we don't thing it's important enough for .1). >> >> The real effort here is double-checking that a dummy handler won't break >> credentials resolution. >> > I think just starting with a full copying the GPG agent handler and > making each call return "failed" should work. Didn't try it, though.
I'll give this a go and hopefully come up with a test case, too. -- Brane
