Fixing the ci/build.ps1 to work with Visual Studio 2017 without running it. I assume that the user will build neovim on cmd.exe or powershell without running the startup batchfiles from Visual Studio.

Notes:

• default to MSVC if %CONFIGURATION% env var is undefined
  • detect MSVC and MINGW folders to know if building is possible
• host/providers are optional because hardcoded paths may not exist in user environment.
• modifications to $env forces the user to run powershell -NoProfile -File ci\build.ps1 to build Neovim to avoid environment corruption
• path validation (ie. Test-Path) to ensure assumptions are met before proceeding
  • ie. mingw paths for 32/64 bit builds and mingw32-make
  • python/ruby/node directories
• Set-PSDebug -Off in case the user runs .\ci\build.ps1 in a powershell session.

I don't follow. Did the change to GetBinaryDep(...) somehow cause MSVC build to fail? Will build.ps1 be required for developers to build now?

It didn't fail but I don't know how to make cmake copy the nvim-qt runtime directory such that, in the build/share/nvim-qt is not created. I tried to add back the same command prior to adding gui_shim.vim but changing share/nvim/runtime to share/nvim-qt/runtime so that nvim-qt can set its env var for its runtime directory and append a valid path to rtp.

Since I don't want to second-guess which directory nvim-qt runtime files are, I'm running ci/build.ps1 on my machine without running Visual Studio 2017 GUI. I'm not even using its command prompt since cmake can find the Visual Studio folders anyway.

ci/build.ps1 isn't required but I want to support running it on a regular cmd/powershell session without GUIs, same as in Appveyor.

cmake warnings are "transferred" to the next cmake invocation so the functional tests always fails on my machine. Not sure if this is cmake or msbuild. I'm getting the same warnings as the passing MSVC builds on Appveyor and the warnings are from building Neovim.

Are there cmake cache files/directories that I have to delete?

Anyway, skipping the tests, only cpack is failing because of CPACK_CMAKE_GENERATOR. Where is it defined?

this?

we do not explicitly set CPACK_CMAKE_GENERATOR as far as I know.

Yes. I'll deal with that later.

Debugging the following failures on my machine

 [  FAILED  ] 5 tests, listed below:
  [  FAILED  ] ...d/repo/git/neovim/test/functional\ex_cmds\write_spec.lua @ 34: :write &backupcopy=auto preserves symlinks
  ...d/repo/git/neovim/test/functional\ex_cmds\write_spec.lua:47: Expected objects to be the same.
  Passed in:
  (table) {
   *[1] = 'content0' }
  Expected:
  (table) {
   *[1] = 'content1' }

  stack traceback:
        ...d/repo/git/neovim/test/functional\ex_cmds\write_spec.lua:47: in function <...d/repo/git/neovim/test/functional\ex_cmds\write_spec.lua:34>

  [  FAILED  ] ...d/repo/git/neovim/test/functional\ex_cmds\write_spec.lua @ 68: :write appends FIFO file
  ...d/repo/git/neovim/test/functional\ex_cmds\write_spec.lua:82: test_fifo: No such file or directory

  stack traceback:
        ...d/repo/git/neovim/test/functional\ex_cmds\write_spec.lua:82: in function <...d/repo/git/neovim/test/functional\ex_cmds\write_spec.lua:68>

  [  FAILED  ] .../neovim/test/functional\legacy\011_autocommands_spec.lua @ 77: file reading, writing and bufnew and filter autocommands BufReadPre, BufReadPost (using gzip)
  .../neovim/test/functional\legacy\011_autocommands_spec.lua:91: Expected objects to be the same.
  Passed in:
  (string) '▼┬ì<├┐Z ♥Xtestfile +.I,*Q├êOS(I-.I├ï├îI├Ñ├è├ë├îKU0├ótLJNIMK├Å├ê├î├è├Ä├ë├ì├ï/(,*.)-+┬»┬¿┬¼┬é┬
¿0├ª┬¼├Ç♂ ┬¬┬ÜcJ┬ö9f♦├ì1'├è∟♂┬é├ªX↕e┬Ä┬í┬ü☻▲┬âR├│RP┬é¶ !qs&e☺  '
  Expected:
  (string) 'start of testfile
  line 2        Abcdefghijklmnopqrstuvwxyz
  line 3        xxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
  line 4        Abcdefghijklmnopqrstuvwxyz
  line 5        xxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
  line 6        Abcdefghijklmnopqrstuvwxyz
  line 7        xxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
  line 8        Abcdefghijklmnopqrstuvwxyz
  line 9        xxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
  line 10 Abcdefghijklmnopqrstuvwxyz
  end of testfile'

  stack traceback:
        .../neovim/test/functional\legacy\011_autocommands_spec.lua:91: in function <.../neovim/test/functional\legacy\011_autocommands_spec.lua:77>

  [  FAILED  ] .../neovim/test/functional\legacy\011_autocommands_spec.lua @ 98: file reading, writing and bufnew and filter autocommands FileReadPre, FileReadPost
  .../neovim/test/functional\legacy\011_autocommands_spec.lua:32: Expected objects to be the same.
  Passed in:
  (table) {
    [access] = 1526676621
    [change] = 1526676621
    [dev] = 2
    [gid] = 0
    [ino] = 0
    [mode] = 'file'
    [modification] = 1526676621
    [nlink] = 1
    [permissions] = 'rw-rw-rw-'
    [rdev] = 2
    [size] = 357
    [uid] = 0 }
  Expected:
  (nil)

  stack traceback:
        .../neovim/test/functional\legacy\011_autocommands_spec.lua:32: in function 'prepare_gz_file'
        .../neovim/test/functional\legacy\011_autocommands_spec.lua:99: in function <.../neovim/test/functional\legacy\011_autocommands_spec.lua:98>

  [  FAILED  ] ...d/repo/git/neovim/test/functional\legacy\delete_spec.lua @ 44: Test for delete() symlink delete
  ...d/repo/git/neovim/test/functional\legacy\delete_spec.lua:56: Expected objects to be the same.
  Passed in:
  (number) -1
  Expected:
  (number) 0

  stack traceback:
        ...d/repo/git/neovim/test/functional\legacy\delete_spec.lua:56: in function <...d/repo/git/neovim/test/functional\legacy\delete_spec.lua:44>

Is there a convenient way to query for admin privs other than os.execute('mklink tmplink tmpfile')?
I can do it via powershell and C# for more accurate detection.

You mean from vim? Not that I know of. Try-and-fail is a fine approach, as you mentioned.

@justinmk Does cmake still have an issue with sh/bash in PATH? What was the reason why it's removed again? I disabled some tests (ie. gzip, fifo) because they come with msysgit and I don't want to deal with them now.

I copied only the shim because I don't know how to copy the entire runtime/ for nvim-qt. Feedback is appreciated.

In the meantime, I'll work on more path checks and adding a mingw fallback if Visual Studio is not installed.

Does cmake still have an issue with sh/bash in PATH ? What was the reason why it's removed again?

It was only a problem when building with mingw, it comes from f53c825. Only @equalsraf knows why. If it doesn't cause a problem, maybe it's not a problem anymore.

I disabled some tests (ie. gzip, fifo) because they come with msysgit and I don't want to deal with them now.

Ok. But I'm confused, how is that related to the gui shim? Aren't those tests already passing on master?

Aren't those tests already passing on master?

They were skipped because Windows doesnt have gzip and mkfifo by default. Appveyor has them because of Git. I have them in my PATH because I install Git in a different folder so the build script didnt remove sh from PATH on my machine. sh is not a problem in MSVC build.

https://github.com/Kitware/CMake/blob/master/Modules/CMakeMinGWFindMake.cmake unsets CMAKE_MAKE_PROGRAM if sh.exe is in PATH.

Having issues compiling LuaJIT because of sjlj exceptions. Not sure if it's caused by -O2

cmake requires -DCMAKE_SH=CMAKE_SH-NOTFOUND -DCMAKE_MAKE_PROGRAM=C:/msys64/mingw$bits/bin/mingw32-make to make cmake -G work with Mingw Makefiles. mingw32-make requires passing CMAKE_SH and CMAKE_MAKE_PROGRAM to the generated Makefile. Maybe it's faster to patch https://github.com/Kitware/CMake/blob/master/Modules/CMakeMinGWFindMake.cmake.

Building LuaJIT on Mingw must omit -O2 to disable sjlj exceptions. I'm using the Debug build for now to force it.

@janlazo How about installing mingw-w64-$arch-ninja and using cmake - G Ninja? It does not complain even if there is sh in the PATH.

@erw7 I'll test Ninja after going through all the affected dependencies.

Is there a mingw compiler that uses seh exception for LuaJIT? I don't know how to force it for gcc so I'm leaving this PR as RFC for now.

I don't know how to answer that, maybe @jszakmeister @jamessan or some other build expert can help.

makedeps.bat was included to conveniently build .deps/ so it will be deprecated since I'm getting tired of running everything after updating the build script. I'll rewrite ci/build.ps1, similar to scripts/vim-patch.sh.

build.ps1 is getting complicated, maybe it isn't worth trying to make it work outside of AppVeyor. Instead use a separate script, then we can potentially use that from ci/build.ps1 if it has a lot of duplicated ideas. But I would guess it might end up not needing much code at all, e.g. makedeps.bat is trivial.

What exactly do you use build.ps1 for, other than building .deps ?

@justinmk I want a convenient script to run everything and ci/build.ps1 has that use-case. I wasn't compiling nvim at all prior to this PR. Problem is the port from batchfile to powershell script is incomplete (for me) because it didn't take full advantage of powershell features.

exitIfFailed might be redundant now. There is makedeps.bat but I don't want to write a batchfile when a powershell function suffices.

I can't reproduce the error for xgettext on MSVC_64. It could be a path escaping issue with one of the arguments if it uses cmd.exe.

MINGW_32 is completely broken because of msvcrt.dll. gdb detects stack buffer overrun on ntdll.dll. How did mingw (32-bit) build before? Could be a segfault (overflow) because of the negative exit code. Strange that all of the dependencies are fine such that nvim can be compiled.