Failing test with whitespace mismatches for hls-splice-plugin

[maralorn]

Your environment

Trying to build hls-splice-plugin-1.0.0.1.

Which OS do you use: Nixos

Steps to reproduce

I ran the tests.

Expected behaviour

Tests pass.

Actual behaviour

  TSimpleDecl.hs (golden):      FAIL (111.47s)
    test/Main.hs:87:
    expected: "{-# LANGUAGE TemplateHaskell #-}\r\n{-# LANGUAGE QuasiQuotes #-}\r\nmodule TSimpleDecl where\r\nimport Language.Haskell.TH ( mkName, clause, normalB, funD, sigD )\r\n\r\n-- Foo\r\n--  Bar\r\nfoo :: Int\r\nfoo = 42\r\n-- Bar\r\n-- ee\r\n-- dddd\r\n"
     but got: "{-# LANGUAGE TemplateHaskell #-}\r\n{-# LANGUAGE QuasiQuotes #-}\r\nmodule TSimpleDecl where\r\nimport Language.Haskell.TH ( mkName, clause, normalB, funD, sigD )\r\n\r\n-- Foo\r\n--  Bar\r\nfoo :: Int\nfoo = 42\r\n-- Bar\r\n-- ee\r\n-- dddd\r\n"
  TQQDecl.hs (golden):          FAIL (133.83s)
    test/Main.hs:87:
    expected: "{-# LANGUAGE QuasiQuotes #-}\r\nmodule TQQDecl where\r\nimport QQ (str)\r\n\r\nfoo :: String\r\nfoo = \"foo\"\r\n"
     but got: "{-# LANGUAGE QuasiQuotes #-}\r\nmodule TQQDecl where\r\nimport QQ (str)\r\n\r\nfoo :: String\nfoo = \"foo\"\r\n"
  TTypeKindError.hs (golden):   OK (125.31s)
  TDeclKindError.hs (golden):   FAIL (118.40s)
    test/Main.hs:120:
    expected: "{-# LANGUAGE TemplateHaskell #-}\r\n{-# LANGUAGE QuasiQuotes #-}\r\nmodule TSimpleDecl where\r\nimport Language.Haskell.TH ( mkName, clause, normalB, funD, sigD )\r\n\r\n-- Foo\r\n--  Bar\r\nfoo :: Int\r\nfoo = 42\r\nbar :: Int\r\n-- Bar\r\n-- ee\r\n-- dddd\r\n"
     but got: "{-# LANGUAGE TemplateHaskell #-}\r\n{-# LANGUAGE QuasiQuotes #-}\r\nmodule TSimpleDecl where\r\nimport Language.Haskell.TH ( mkName, clause, normalB, funD, sigD )\r\n\r\n-- Foo\r\n--  Bar\r\nfoo :: Int\nfoo = 42\nbar :: Int\r\n-- Bar\r\n-- ee\r\n-- dddd\r\n"

3 out of 31 tests failed (138.46s)

This looks like some issue with unix vs. windows line endings?

[felixonmars]

I am getting the same errors when building for Arch:

  TSimpleDecl.hs (golden):      FAIL (88.82s)
    test/Main.hs:87:
    expected: "{-# LANGUAGE TemplateHaskell #-}\r\n{-# LANGUAGE QuasiQuotes #-}\r\nmodule TSimpleDecl where\r\nimport Language.Haskell.TH ( mkName, clause, normalB, funD, sigD )\r\n\r\n-- Foo\r\n--  Bar\r\nfoo :: Int\r\nfoo = 42\r\n-- Bar\r\n-- ee\r\n-- dddd\r\n"
     but got: "{-# LANGUAGE TemplateHaskell #-}\r\n{-# LANGUAGE QuasiQuotes #-}\r\nmodule TSimpleDecl where\r\nimport Language.Haskell.TH ( mkName, clause, normalB, funD, sigD )\r\n\r\n-- Foo\r\n--  Bar\r\nfoo :: Int\nfoo = 42\r\n-- Bar\r\n-- ee\r\n-- dddd\r\n"
  TQQDecl.hs (golden):          FAIL (102.13s)
    test/Main.hs:87:
    expected: "{-# LANGUAGE QuasiQuotes #-}\r\nmodule TQQDecl where\r\nimport QQ (str)\r\n\r\nfoo :: String\r\nfoo = \"foo\"\r\n"
     but got: "{-# LANGUAGE QuasiQuotes #-}\r\nmodule TQQDecl where\r\nimport QQ (str)\r\n\r\nfoo :: String\nfoo = \"foo\"\r\n"
  TTypeKindError.hs (golden):   OK (99.71s)
  TDeclKindError.hs (golden):   FAIL (93.87s)
    test/Main.hs:120:
    expected: "{-# LANGUAGE TemplateHaskell #-}\r\n{-# LANGUAGE QuasiQuotes #-}\r\nmodule TSimpleDecl where\r\nimport Language.Haskell.TH ( mkName, clause, normalB, funD, sigD )\r\n\r\n-- Foo\r\n--  Bar\r\nfoo :: Int\r\nfoo = 42\r\nbar :: Int\r\n-- Bar\r\n-- ee\r\n-- dddd\r\n"
     but got: "{-# LANGUAGE TemplateHaskell #-}\r\n{-# LANGUAGE QuasiQuotes #-}\r\nmodule TSimpleDecl where\r\nimport Language.Haskell.TH ( mkName, clause, normalB, funD, sigD )\r\n\r\n-- Foo\r\n--  Bar\r\nfoo :: Int\nfoo = 42\nbar :: Int\r\n-- Bar\r\n-- ee\r\n-- dddd\r\n"

[jneira]

Usually tests fail in windows cause the same issue 🙂 , editing the source test cases should fix it, i suppose

[Ailrun]

It's partially because splice plugin does not use any golden test util, and implement its own pseudo-golden test by itself... (If we use git diff, we can ignore \r differences)

[berberman]

Yeah, eval plugin and tactics plugin have the same problem as well.

And it turns out that files in uploaded tarballs are different from those in our git source tree:

$ diff ~/Downloads/hls-splice-plugin-1.0.0.1/test/testdata/TErrorExp.hs.expected ~/haskell-language-server/plugins/hls-splice-plugin/test/testdata/TErrorExp.hs.expected   
1,6c1,6
< {-# LANGUAGE TemplateHaskell #-}
< module TErrorExp where
< import Language.Haskell.TH ( tupE, litE, integerL )
< 
< main :: IO ()
< main = return (42, ())
---
> {-# LANGUAGE TemplateHaskell #-}
> module TErrorExp where
> import Language.Haskell.TH ( tupE, litE, integerL )
> 
> main :: IO ()
> main = return (42, ())

@jneira Did you create this sdist on Windows?

[jneira]

yeah, it seems it could be my fault, maybe cause i have auto.crlf enabled (not at my laptop rigt now).
I will take a look asap

[maralorn]

It's partially because splice plugin does not use any golden test util, and implement its own pseudo-golden test by itself... (If we use git diff, we can ignore \r differences)

btw. git diff also leads to problems on nixos because it shows differences in file permissions caused by the nix-store. But it’s okay. We can of course disable them …

[Ailrun]

@maralorn That issue is now fixed on master (and probably after next version)

[jneira]

I've checked mye env and:

PS D:\hls> git config --get auto.crlf
false

and i have the .editorconfig plugin installed so it should change crlf by lf on save (although i did not open those test data files iirc)

🤔

EDIT: never mind, i've set the config option wrong (it is core.autocrlf) 🤦

[jneira]

Ok i forced a checkout after fixing the damn core.autocrlf and all files have LF now. So i will bump up the plugin to 1.0.0.2 in master, backport the change to hackage-1.1.0 and upload the tarball again
Sorry for the inconveniences caused by that stupid error 😟

[jneira]

I've uploaded https://hackage.haskell.org/package/hls-splice-plugin-1.0.0.2 after double checking files have LF.

Akthough i am gonna use unix LF cause i want to be kind with my fellow posix comrades 😝 , tests should be resilient to line endings changes, like other plugins, as suggested by @Ailrun

@maralorn tests should work now (at least they shouldn't fail due to line endings)

[felixonmars]

I've uploaded https://hackage.haskell.org/package/hls-splice-plugin-1.0.0.2 after double checking files have LF.

Akthough i am gonna use unix LF cause i want to be kind with my fellow posix comrades stuck_out_tongue_closed_eyes , tests should be resilient to line endings changes, like other plugins, as suggested by @Ailrun

@maralorn tests should work now (at least they shouldn't fail due to line endings)

Confirmed that the tests are passing now in Arch. Thanks!

archlinux/svntogit-community@ffc8dbe#diff-37538beb61ff63edebbf735dfcf39e5d732f49183d6beb097169d971875ca422L42