make test fails early on (Win64)

[joa-quim]

Actually, I never managed to run the the make test till the end on Windows, but since a week or two it's failing quite earlier than it used to.

/c/programs/julia64> make test
    JULIA test/all
        From worker 5:       * linalg4
        From worker 3:       * linalg2
        From worker 4:       * linalg3
        From worker 2:       * linalg1
        From worker 4:       * core
        From worker 4:       * keywordargs
        From worker 5:       * numbers
        From worker 4:       * strings
exception on 4: ERROR: test failed: @mstr "\r\n      foo\r\n      \tbar" == "foo\n\tbar"
 in error at error.jl:21
 in default_handler at test.jl:19
 in do_test at test.jl:39
while loading strings.jl, in expression starting on line 812
ERROR: test failed: @mstr "\r\n      foo\r\n      \tbar" == "foo\n\tbar"
 in anonymous at task.jl:1350
 in anonymous at task.jl:335
while loading strings.jl, in expression starting on line 812
while loading C:\programs\julia64\test\runtests.jl, in expression starting on line 46

[ivarne]

Might be related to #7165

[joa-quim]

Nope, unfortunately no. Still errors for the same reason.

[tkelman]

I think what @ivarne meant is the test was recently added to resolve that issue, but leads to trouble on Windows depending on how your git is configured. If your git is configured to download the source with Windows line endings (or if you open and modify the file with some brain-dead editor that always replaces line endings with \r\n), looks like you'll run into this problem. PITA carriage returns.

[joa-quim]

Indeed I went check and the files have the Windows EOLs.
I didn't (at least voluntarily) configured git to save files with the one or the other EOL type. I have both Turtoise and a Win git version installed and both do the same thing.
But, isn't this something that is set at the repertory level? I mean, on SVN we set the EOL type that each system will use (at the repertory level). Does Git do it differently?

[tkelman]

There are multiple ways of doing it in Git, at the user level or at the repository level. See https://help.github.com/articles/dealing-with-line-endings#platform-all

Since there's no .gitattributes file in the Julia repository (or its various submodules), git falls back to the user's configuration. The defaults there depend on where you installed git from (I think Cygwin and MSYS2 gits default to \n, other git-for-Windows installations default to \r\n).

Can you check if the simple modification to the test in #7220 fixes things on your end?

[joa-quim]

Yes, it does solve the issue thanks.

And now I'm back to my other long standing failure.
Shall I open another issue? I hesitate to do this for every test failure because I assume the developers are seing them as well (all it take is to run the test in a Win machine)

Joaquim

        From worker 4:       * mpfr
exception on 2: ERROR: test error during !(ispath(file))
stat: operation not permitted (EPERM)
 in ispath at stat.jl:105
while loading file.jl, in expression starting on line 244
exception on 4: ERROR: test failed: f > i1
 in error at error.jl:21
 in default_handler at test.jl:19
 in do_test at test.jl:39
while loading mpfr.jl, in expression starting on line 778
ERROR: test error during !(ispath(file))
stat: operation not permitted (EPERM)
while loading file.jl, in expression starting on line 244
while loading C:\programs\julia64\test\runtests.jl, in expression starting on line 46

[tkelman]

Usually best to do a search through the issues to see whether or not the error messages look new. You're seeing two failures there. The failure in test/file.jl is #6144, the failute in test/mpfr.jl is being discussed in #3399

[joa-quim]

Right, but it's not such easy thing to do. For example the repetition of

ERROR: test error during !(ispath(file))
stat: operation not permitted (EPERM)

made me think that there was only one error.