A test `windows_newlines` seems wrong
From #188. I found prefix(::BufferedInput, ::Integer) has a bug and if I fixed it, the test windows_newlines fails. Also, better implementation in #188's TODO fails at the same test. I think the test windows_newlines is something wrong.
The content:
julia> str = read("test/windows_newlines.data", String)
"hello:\r\0\r"
Loading result of windows_newlines from http://ben-kiki.org/ypaste/cgi-bin/ypaste.pl
So the test should be fixed.
I don't know why only 2 tests are failed.
It's strange to use hello:\r\0\r to test for Windows. It should be hello:\r\n\r.
I asked this in YAML community, and the file hello:\r\0\r is malformed. We should fail this even for load_all_file and load_file and load but only load_all fails. Hmm, why?
This test file was added when #17 was fixed. I think it was just supposed to be hello:\r\n but ended up differently.
#211 fixes this.
We should fail this even for
load_all_fileandload_fileandloadbut onlyload_allfails. Hmm, why?
It's probably incorrect, but \0 is basically parsed like \n..., so this document is like hello:\r\n...\r. load_all fails because the second document has no content, which is known to be handled badly. load_file and load work because they stop after the first document. load_all_file works thanks to the bug pointed out in #120. YAMLDocIterator loads the first document on construction, and the second document when the first is retrieved. Due to the input file being prematurely closed, it never tries to load the second document as the closed input signals eof.
is this closed by #218 ?