Encoding problems when include cyrillic files

[nt0xa]

Steps to reproduce:

1. create two files

test.md

# Тест

* один
* два
* три

```include
include.md
```

include.md

* четыре
* пять
* шесть

2. run

pandoc --filter pandoc-include -o result.html test.md

And result will be:

<h1 id="тест">Тест</h1>
<ul>
<li>один</li>
<li>два</li>
<li>три</li>
</ul>
<ul>
<li>╤З╨╡╤В╤Л╤А╨╡</li>
<li>╨┐╤П╤В╤М</li>
<li>╤И╨╡╤Б╤В╤М</li>
</ul>

[nt0xa]

It seems that readFile use CP866 encoding on Windows (with default russian language)

[LarsEKrueger]

Can confirm with Windows 10 and pandoc 1.16.0.2 (pandoc and pandoc-include built on the machine using stack).

As another data point for hunting down this bug:

Take a file containing the umlauts of a, o, and u (LATIN SMALL LETTER A WITH DIAERESIS etc.)

bug.markdown

ä ö ü

Run it in cmd as

chcp 65001
pandoc bug.markdown

and you get the expected:

ä ö ü

Even if you run it as

pandoc --filter pandoc-include bug.markdown

you get the expected output.

However, if you run it as

pandoc --filter pandoc-include incl.markdown

where incl.markdown only includes bug.markdown, you get messed-up characters.

[steindani]

Hi @russtone, @LarsEKrueger, thank you for your patience and the detailed issue. On a new branch I've made it possible to set the encoding to UTF-8, will this help in your case?

To use, simply add utf-8 after the include class, like so:

```include utf8
a.md
```

[LarsEKrueger]

Can't test the branch right now. Might take a few days until I find the time.

However, I don't understand why you want to make a difference in the encoding of the file that does the include and the one that is included.

Pandoc is UTF-8 on input and there's no way around it. One would expect that include files are UTF-8 too, without requesting them to be. If there's a relevant use case that I don't see right now, you definitely need to document that.

[steindani]

@LarsEKrueger This makes it even easier, thank you!

[LarsEKrueger]

Tried commit 53b0d1 on Windows 10 (with Creator's Update). Pandoc and pandoc-include compiled using stack and ghc 8.

Issue is still there.

[steindani]

Thank you for checking. Unfortunately I have no other idea what could be wrong.

[steindani]

Oh, I see. That commit is old, and does not contain the fixes. Please try the latest on the fixing branch: 913ca87 . Thanks!

[LarsEKrueger]

Still doesn't work.

Could it be that the hSetEncoding isn't evaluated due to laziness and didn't notice during testing (i.e. because your default encoding is already uft8)?

I use the following code in my filter and it does work on windows.

justReadFile :: String -> IO (Maybe [Block])
justReadFile fn = bracket (openFile fn ReadMode) hClose $ \handle -> do
  hSetEncoding handle utf8
  cont <- hGetContents handle
  case readMarkdown def cont of
      Left _ -> return Nothing
      Right (Pandoc _ blocks) -> return $ Just blocks

If I use your fmap pattern, it ceases to work correctly. The code is:

justReadFile :: String -> IO (Maybe [Block])
justReadFile fn = bracket (openFile fn ReadMode) hClose $ \handle -> do
  fmap (`hSetEncoding` utf8) $ return handle
  cont <- hGetContents handle
  case readMarkdown def cont of
      Left _ -> return Nothing
      Right (Pandoc _ blocks) -> return $ Just blocks

Your variable handle in fileContentAsString is actually of type IO Handle, not Handle. Thus the fmap typecheck correctly, but the hSetEncoding is either run never or after the hGetContents. It's the same reason I needed to add the return, because fmap wants an IO Handle.

[steindani]

Thank you, you are awesome. I've updated the branch, and really hope this will solve the bug.

[LarsEKrueger]

After removing the utf8 class from the include code block, I ran the following command:

(cd test/encoding/ ; pandoc -f markdown -t html -s -o test.html --filter ../../dist/dist-sandbox-1d3e9dda/build/pandoc-include/pandoc-include test.md)
pandoc-include: include.md: hGetContents: illegal operation (delayed read on closed handle)
Error running filter ../../dist/dist-sandbox-1d3e9dda/build/pandoc-include/pandoc-include:
Filter returned error status 1

I encountered this problem too when writing justReadFile (see previous comment). I fixed it by moving the readMarkdown inside the function.

Same error happens on Windows.