InstallScript GetLine() can not read text file contains result from command prompt

Question

My Installation needs to check the result of a command from cmd.exe. Thus, I redirect the result of the command to a text file and then try to read the file to get the result as follows:

// send command to cmd to execute and redirect the result to a text file

// try to read the file
szDir = "D:\\";
szFileName = "MyFile.txt";

if Is(FILEEXISTS, szDir ^ szFileName) then
    listID = ListCreate(STRINGLIST);
    if listID != LIST_NULL then
        if OpenFIleMode(FILE_MODE_NORMAL) = 0 then
            if OpenFile(nFileHandle, szDir, szFileName) = 0 then

                // I run into problem here
                while (GetLine(nFileHandle, szCurLine) = 0 )
                    ListAddString(listID, szCurLine, AFTER);
                endwhile;
                CloseFile(nFileHandle);

            endif;
        endif;
    endif;
endif;

The problem is that right after the command prompt is executed and the result is redirected to MyFile.txt, I can set open file mode, open the file but I can not read any text into my list. ListReadFromFile() does not helps. If I open the file, edit and save it manually, my script works.

After debugging, I figured that GetLine() returns an error code (-1) which means the file pointer must be at the end of file or other errors. However, FILE_MODE_NORMAL sets the file as read only and SET THE FILE POINTER AT THE BEGINNING OF THE FILE.

What did I possibly do wrong? Is this something to do with read/write access of the file? I tried this command without result:

icacls D:\MyFile.txt /grant Administrator:(R,W)

I am using IstallShield 2018 and Windows 10 64-bit btw. Your help is much appreciated.

EDIT 1: I suspected the encoding and tried a few things:

1. After running "wslconfig /l", the content of MyFile.txt opened in Notepad++ is without an encoding, but still appeared normal and readable. I tried to converted the content to UTF-8 but it did not work.

2. If I add something to the file (echo This line is appended >> MyFile.txt), the encoding changed to UTF-8, but the content in step 1 is changeed also. NULL (\0) is added to between every character and even repelace new line character. Maybe this is why GetLine() failed to read the file.

3. Work around: after step 1, I run "find "my_desired_content" MyFile.txt" > TempFile.txt and read TempFile.txt (which is encoded in UTF-8).

My ultimate goal is to check if "my_desired_content" apeears in the result of "wslconfig /l" so this is fine. However, what I don't understand is that both MyFile.txt and TempFile.txt are created from cmd command but they are encoded differently?

Answer 1

The problem is due to the contents of the file. Assuming this is the file generated by your linked question, you can examine its contents in a hex editor to find out the following facts:

• Its contents are encoded in UTF-16 (LE) without a BOM
• Its newlines are encoded as CR or CR CR instead of CR LF

I thought the newlines would be more important than the text encoding, but it turns out I had it backwards. If I change each of these things independently, GetLine seems to function correctly for either CR, CR CR, or CR LF, but only handles UTF-16 when the BOM is present. (That is, in a hex editor, the file starts with FF FE 57 00 instead of 57 00 for a file starting with the character W.)

I'm at a bit of a loss for the best way to address this. If you're up for a challenge, you could read the file with FILE_MODE_BINARYREADONLY, and can use your extra knowledge about what should be in the file to ensure you interpret its encoding correctly. Note that for most of UTF-16, you can create a single code unit by combining two bytes in the following manner:

szResult[i] = (nHigh << 8) + nLow;

where nHigh and nLow are probably values like szBuffer[2*i + 1] and szBuffer[2*i], assuming you filled a STRING szBuffer by calling ReadBytes.

Other unproven ideas include editing it in binary to ensure the BOM (FF FE) is present, figuring out ways to ensure the file is originally created with the BOM, figuring out ways to create it in an alternate encoding, finding another command you can invoke to "fix" the file, or lodging a request with the vendor (my employer) and hoping the development team changes something to better handle this case.

---

Here's an easier workaround. If you can safely assume that the command will append UTF-16 characters without a signature, you can append this output to a file that has just a signature. How do you get such a file?

• You could create a file with just the BOM in your development environment, and add it to your Support Files. If you need to use it multiple times, copy it around first.

• You could create it with code. Just call the following (error checking omitted for clarity)

  OpenFileMode(FILE_MODE_APPEND_UNICODE);
  CreateFile(nFileHandle, szDir, szFileName);
  CloseFile(nFileHandle);
  

  and if szDir ^ szFileName didn't exist, it will now be a file with just the UTF-16 signature.

Assuming this file is called sig.txt, you can then invoke the command wslconfig /l >> sig.txt to write to that file. Note the doubled >> for append. The resulting file will include the Unicode signature you created ahead of time, plus the Unicode data output from wslconfig, and GetLine should interpret things correctly.

The biggest problem here is that this hardcodes around the behavior of wslconfig, and that behavior may change at any point. This is why Christopher alludes to recommending an API, and I agree completely. In the mean time, You could try to make this more robust by invoking it in a cmd /U (but my understanding of what that does or guarantees is fuzzy at best), or by trying the original way and then with the BOM.

Answer 2

This whole WSL thing is pretty new. I don't see any APIs it but rather then screen scrapping command outputs you might want to look at this registry key:

HKEY_CURRENT_USER\SOFTWARE\Microsoft\Windows\CurrentVersion\Lxss

It seems to have the list of installed distros that come from the store. Coming from the store probably explains why this is HKCU and not HKLM.

A brave new world.... sigh.