On Sat, Dec 27, 2008 at 9:00 AM, Stefan Sperling <stsp_at_elego.de> wrote:
> Hi folks,
>
> I got a report of a test failure in 1.5.5, trans_test.py, test 1:
...
> File
> "/b/ports/obj/subversion-1.5.5/subversion-1.5.5/subversion/tests/cmdline/
> svntest/actions.py", line 112, in guarantee_greek_repository
> if main.copy_repos(main.pristine_dir, path, 1):
> File
> "/b/ports/obj/subversion-1.5.5/subversion-1.5.5/subversion/tests/cmdline/
> svntest/main.py", line 655, in copy_repos
> load_in.close() # Tell load we are done
> IOError: [Errno 32] Broken pipe
> FAIL: trans_tests.py 1: commit new files with keywords active from birth
>
> Apparently, the test fails this way on OpenBSD/amd64.
> It works for me on OpenBSD/i386.
>
> There's an interesting comment above that test, which was
> added in r18242:
...
> +### This test is know to fail when Subversion is built in very deep
> +### directory structures, caused by SVN_KEYWORD_MAX_LEN being defined
> +### as 255.
...
> Unfortunately, the comment does not explain what kind of
> failure is to be expected. Is the reported error related
> to this comment or not?
Stefan, that comment was added based on a report from Blair, based on
a root cause analysis from Peter Lundblad:
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=675612
http://subversion.tigris.org/ds/viewMessage.do?dsMessageId=675616&dsForumId=462
The original report doesn't include the level of detail you want.
Have you checked what SVN_KEYWORD_MAX_LEN is #define'd to across your
builds on those platforms? If nothing looks suspect, try strace'ing
the load operation highlighted above.
- Dan
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=994079
Received on 2008-12-28 02:06:03 CET