Dvc: progress, prompt: isatty broken on Windows, default git bash and conda python

[shcheklein]

Effectively it means that we don't see any progress bars. Just empty screen for commands.

It's a well known issue on a default setup - conda + git bash (haven't tested other configurations). More on this here: https://stackoverflow.com/a/32599341/298182

To reproduce:

import sys

file = sys.stderr
print(file.isatty())

returns False if you run like this python test.py
returns True if winpty python test.py

[efiop]

DVC_IGNORE_ISATTY=true dvc add data

works, but has broken formatting :slightly_frowning_face:

[efiop]

Can reproduce the same with regular official python (not from conda).

[casperdcl]

Based on the SO link should we just document winpty on git-bash?

[casperdcl]

if DVC_IGNORE_ISATTY still had broken formatting this means that git-bash simply doesn't support \r and progress bars

[casperdcl]

@efiop does winpty fix the formatting?

[efiop]

@casperdcl It actually looks pretty strange:

$ DVC_IGNORE_ISATTY=true dvc add data
  0%|          |Add                                   0/1 [00:00<?,     ?file/s]
!
Computing hashes (only done once)                   0 [00:00<??:??,      ?md5/s]
100% Add                                          1/1 [00:16<00:00,  16.2s/file]

To track the changes with git, run:

        git add data.dvc

This is literally what I'm seeing, no line was updated :thinking:

Didn't try winpty yet :slightly_frowning_face: @shcheklein Does it work with dvc add correctly?

[shcheklein]

@efiop @casperdcl winpty works fine for me. I wonder how other tools deal with this. For example, I've seen conda printing progress bars without any winpty? What is that specific that winpty is doing after all?

[casperdcl]

Ah interesting... conda uses tqdm so there must be a way to patch it.

[shcheklein]

@casperdcl may be it does not check isatty at all?

[casperdcl]

@shcheklein @efiop DVC_IGNORE_ISATTY=true also skips checking isatty so there must be something else going on

[casperdcl]

Actually this may just be a colorama issue on windows, which may coincidentally have been fixed by my last devel-branch commit: https://github.com/tqdm/tqdm/commit/ae6de71c3460123944aa7bb80d7cb0b5bdcafb7b

@efiop / @shcheklein could you try

pip install -U git+https://github.com/tqdm/tqdm.git@ae6de71c#egg=tqdm

and check if it fixes this issue?

[efiop]

@casperdcl Installing that still shows

$ DVC_IGNORE_ISATTY=true dvc add data
  0%|          |Add                                   0/1 [00:00<?,     ?file/s]
!
Computing hashes (only done once)                   0 [00:00<??:??,      ?md5/s]
100% Add                                          1/1 [00:16<00:00,  16.2s/file]

To track the changes with git, run:

        git add data.dvc

for me :slightly_frowning_face:

[casperdcl]

hmm. @efiop was colorama also installed on your machine?

[efiop]

@casperdcl Ah, wait. I do see a change if i drop #egg part in that command(not sure what is up with that). Now it works, but it prints on a new line each time:

kupru@DESKTOP-LQKMJV2 /cygdrive/c/Users/kupru/myrepo
$ DVC_IGNORE_ISATTY=true dvc add data
  0%|          |Add                                   0/1 [00:00<?,     ?file/s]
!
  0%|          |Computing hashes (only done once) 0/12057 [00:00<?,      ?md5/s]
  0%|          |Computing hashes (only done o17/12057 [00:00<01:25,    142md5/s]
  0%|          |Computing hashes (only done o31/12057 [00:00<01:26,    140md5/s]
  0%|          |Computing hashes (only done o48/12057 [00:00<01:24,    142md5/s]
  0%|          |Computing hashes (only done o60/12057 [00:00<01:31,    132md5/s]
  1%|          |Computing hashes (only done o75/12057 [00:00<01:31,    132md5/s]
  1%|          |Computing hashes (only done o91/12057 [00:00<01:26,    138md5/s]
  1%|          |Computing hashes (only done 104/12057 [00:00<01:29,    134md5/s]
  1%|          |Computing hashes (only done 118/12057 [00:00<01:29,    134md5/s]
  1%|1         |Computing hashes (only done 132/12057 [00:00<01:29,    133md5/s]
  1%|1         |Computing hashes (only done 147/12057 [00:01<01:29,    133md5/s]
  1%|1         |Computing hashes (only done 161/12057 [00:01<01:31,    131md5/s]
  1%|1         |Computing hashes (only done 175/12057 [00:01<01:31,    130md5/s]
  2%|1         |Computing hashes (only done 191/12057 [00:01<01:27,    136md5/s]
  2%|1         |Computing hashes (only done 206/12057 [00:01<01:27,    136md5/s]
  2%|1         |Computing hashes (only done 220/12057 [00:01<01:28,    134md5/s]
  2%|1         |Computing hashes (only done 235/12057 [00:01<01:26,    137md5/s]
  2%|2         |Computing hashes (only done 249/12057 [00:01<01:26,    137md5/s]
  2%|2         |Computing hashes (only done 263/12057 [00:01<01:30,    131md5/s]
  2%|2         |Computing hashes (only done 277/12057 [00:02<01:32,    127md5/s]
  2%|2         |Computing hashes (only done 290/12057 [00:02<01:32,    127md5/s]
  3%|2         |Computing hashes (only done 303/12057 [00:02<01:35,    123md5/s]
  3%|2         |Computing hashes (only done 316/12057 [00:02<01:36,    122md5/s]
  3%|2         |Computing hashes (only done 329/12057 [00:02<01:34,    124md5/s]
  3%|2         |Computing hashes (only done 344/12057 [00:02<01:30,    129md5/s]
  3%|2         |Computing hashes (only done 359/12057 [00:02<01:27,    133md5/s]
  3%|3         |Computing hashes (only done 373/12057 [00:02<01:30,    129md5/s]

So looks like it almost works :)

[efiop]

@casperdcl Yes, colorama is installed(we have it as a dependency for dvc too). 0.4.1 though

[casperdcl]

hmm - so likely some sort of variant of https://github.com/tartley/colorama/issues/79 - maybe at the top of the main file @efiop try patching:

diff --git a/dvc/__main__.py b/dvc/__main__.py
index e160b02..7d1804a 100644
--- a/dvc/__main__.py
+++ b/dvc/__main__.py
@@ -4,6 +4,10 @@ from __future__ import unicode_literals
 import sys

 from dvc.main import main
+from dvc.progress import Tqdm
+import colorama
+colorama.deinit()
+colorama.init(strip=False, convert=True)

 if __name__ == "__main__":
     main(sys.argv[1:])

?

[efiop]

@casperdcl Didn't help. Same as in https://github.com/iterative/dvc/issues/2747#issuecomment-551300835 :slightly_frowning_face:

[efiop]

Btw, running wget in gitbash, it also doesn't respect \r and prints new lines each time.

[casperdcl]

unfortunately rprichard/winpty isn't exactly a small fix which could be integrated into dvc core. May have to rely on detecting MSYS/Cygwin/git-bash from os.environ and doing something special accordingly.

Would also need to ensure the "something special" (e.g. reduce progress updates to once per 2 seconds to prevent console spamming, or maybe disable progress altogether, or add a single . each iteration, etc...) isn't done when using winpty

[casperdcl]

looks like https://github.com/msys2/MINGW-packages/pull/2675 isn't yet in the latest gitforwindows... Interestingly I found (in git-bash):

# normal (dvc CLI) use
$ python -c 'import sys; print(sys.stderr.encoding)'
<_io.TextIOWrapper name='<stderr>' mode='w' encoding='cp1252'>
# emulate file redirection
$ python -c 'import sys; print(sys.stderr.encoding)' 2>/dev/null
<_io.TextIOWrapper name='<stderr>' mode='w' encoding='cp850'>

maybe we could use isatty() or encoding == 'cp1252'? Tested on latest git-bash & conda3, though not sure if windows version affects this as well.

[casperdcl]

As a quick follow-up - it's not a good idea to check env vars for detecting if running in git-bash because we still want to disable if really redirecting to a file.