On February 9, 2019 18:33, Jeff King wrote: > On Sat, Feb 09, 2019 at 01:08:01PM -0500, Randall S. Becker wrote: > > > > It sounds like you might need to set FREAD_READS_DIRECTORIES in your > > > config.mak. > > > > Setting FREAD_READS_DIRECTORIES = UnfortunatelyYes > > Silly question, but you did rebuild after setting that, not just re-run the tests, > right? Yes make was run, but some doofus (me) did not run make tests first, so test-tool was out of date, post reconfigure. When just a code change is made, test-tool gets build. I was/am confused. > > still results in > > > > Value not found for "foo.bar" > > test_expect_code: command exited with 1, we wanted 2 test-tool config > > configset_get_value foo.bar a-directory not ok 23 - proper error on > directory "files" > > # > > # echo "Error (-1) reading configuration file a-directory." >expect > && > > # mkdir a-directory && > > # test_expect_code 2 test-tool config configset_get_value foo.bar > a-directory 2>output && > > # grep "^warning:" output && > > # grep "^Error" output >actual && > > # test_cmp expect actual > > # > > > > I don't think that helped. While fopen can open a directory, fread > > does not return any data in this platform. readdir or nftw/ftw are > > pretty much the only options. However, the code still goes down the > > goto exit1 path in this situation. > > Hrm. That's the exact case FREAD_READS_DIRECTORIES is supposed to help. > It does an fstat() after fopen()ing the file to see if it points to a directory, and > if so closes it immediately and returns EISDIR. > > Can you confirm via debugger (or printf statements) that we make it into > git_fopen, and what the resulting st.st_mode we see looks like? It did not get into git_fopen at all. Subtest 1 did, but that's no surprise. t1308 is passes now, finally. Patch coming. Thanks for the help, Randall