At 06/13/2012 12:10 PM, Eric Blake Wrote:
On 06/12/2012 09:07 PM, Wen Congyang wrote:
We really shouldn't need to compile a file for a virgin 'make dist' to work, but now I have enough information to repeat it. It may be a while before I have a clean solution, though, since every thing I try implies nuking my tree and starting from a fresh clone module my patch attempt.
I now have 'make dist' works; if it also passes the longer 'make distcheck' on a virgin tree, then I will push my patch under the build-breaker rule (patch in separate mail). The bug was that we had a file in the tarball that depended on a generated file, which is a no-no. We really need to ship remote_protocol-structs, and we also want 'make check' to ensure that file is up-to-date (which includes a dependency on a generated file), but a little bit of refactoring makes it so that 'make check' need not interfere with 'make dist', by having the two targets depend on different names.
I apply your patch, and the problem does not exist. Thanks Wen Congyang
I revert some commits and test the building. I find that this problem is introduced by the commit 7bff56a0d1514cb955eb14adc14281626e80e96c.
That was fixing real bugs, but I'm not surprised that other latent bugs were exposed in the process.
Yes, that commit fixes a bug, but introduce a new bug.
Not a new bug, but a latent one. The bug has been present since July 2011 (commit 62dee6f), but it was the refactoring of commit 7bff56a that exposed it better.