[FLINK-6987] Fix erroneous when path containing spaces

[rice668]

We will get an error Reason: "Test erroneous" when use a folder was called "flink-1.3.1 2" to build flink and in which contains spaces then that error was triggered. The reason is that the spaces will encode into flink-1.3.1%202 and then the system can not find the path.

[rice668]

Hi @twalthr , you can test it again. I think it works fine now.

[zentol]

I doubt this fixes the underlying issue, it just masks it by storing the input files in paths containing spaces.

[rice668]

I dont think it has an underlying issue as below line of code throw an error, and it is not relavant to the TextInputFormat class . tmpDir has spaces and then trigger that problem. If my understanding is wrong, please help me out.

File tempFile = File.createTempFile("TextInputFormatTest", ".tmp", tmpDir);

[rice668]

Oops. I might be wrong. I will check again.

[rice668]

Thanks @twalthr for reporting this VERY good issue. And thanks @zentol for looking over it. PR has been updated. I misread the error from JIRA before , so literally used trim() to solve and it was proved to be wrong.

[zentol]

Pretty sure this fails on windows.

[rice668]

What do you mean ? My computer is windows and operation system belongs to windows 7. I didnt get any error with this line of code and with a folder called flink-1.3.1 2 to build it with test. Test passed with flink-1.3.1 2 when build flink.

[zentol]

Typically, when going from flink paths to Strings to jave file API's on Windows you get exceptions since Flink paths on Windows always start with a "/". But this call never goes through the file API, hence it does succeed.

Anyway, an easier solution is to simply not call URI#toString when calling the Path constructor in the tests. Then the spaces aren't escaped and the tests succeed.

[rice668]

An awesome solution always welcomed. I can literally use parentDir.toString() to avoid that issue. Not to call parentDir.toURI().toString().

[rice668]

We can also use parentDir.getAbsoluteFile().toString() to hack that. Not very sure about which way is better. I would think parentDir.getAbsoluteFile().toString() is better as it can obtain the whole path. What do you think ? @zentol

[zentol]

I would keep the conversion to an URI for the sake of making less changes to the test.

[rice668]

Oh. Yeah. You are very correct, we need make the less changes to test. PR will update soon today.

[rice668]

I have updated the code. Please check again. One thing we should concern though.

FYI, though, it is not very relevant to this issue here.

We can see, for 1 and 2 output results, they are both the same. That is to say, call toString() from toURI() does not change the internal of the result. But, if we call them in context of flink.core.fs.Path. what I got is 3 and 4 outputs. It seems new Path change someting stuff. I guess that we should expect that not happen under this context. Instead, I expect new Path(parentDirTest.toURI().toString()) return the same value as new Path(parentDirTest.toURI()). Now, they are not equal.

File parentDirTest = new File("D:\\projects\\hello world");
System.out.println(parentDirTest.toURI());

1. Output: file:/D:/projects/hello%20world/

System.out.println(parentDirTest.toURI().toString());

2. Output: file:/D:/projects/hello%20world/

System.out.println(new Path(parentDirTest.toURI()));

3. Output: file:/D:/projects/hello world/

System.out.println(new Path(parentDirTest.toURI().toString()));

4. Output: file:/D:/projects/hello%20world

[zentol]

That#s a problem of URI, not of the path class. URI#toString escapes certain characters (like spaces) but they aren't escaped internally.

[rice668]

Oh. I see, thanks.

[zentol]

merging.