incorrect app.flags.DEFINE_boolean behavior

[mborgerding]

demo python script

test.py

#!/usr/bin/python
import tensorflow as tf
tf.app.flags.DEFINE_boolean('tea',False,'Soggy Leaves')
cfg = tf.app.flags.FLAGS

if cfg.tea:
    print( "have some tea" )
else:
    print( "no tea" )

Expected behavior

When the above test.py is run with any of {(no arguments), --notea, or --tea=True}, the script behaves as expected...

[markb@schur tmp]$ ./test.py
no tea
[markb@schur tmp]$ ./test.py --notea
no tea
[markb@schur tmp]$ ./test.py --tea=True
have some tea

Weird stuff

Strangely --tea=False is incorrectly interpreted as True. Having both tea and no tea is only possible in certain Infocom games.

[markb@schur tmp]$ ./test.py --tea=False
have some tea

Apparently, only those who truly want tea will get it.

[markb@schur tmp]$ ./test.py --tea
usage: test.py [-h] [--tea TEA] [--notea]
test.py: error: argument --tea: expected one argument

C'mon. It's a boolean flag !
"Say something once. Why say it again?" -- Talking Heads

[vrv]

Looks like we're getting hit by: http://stackoverflow.com/questions/15008758/parsing-boolean-values-with-argparse

It's doing bool("False") --> True.

I'll try to get it to support:

--tea --> tea is True
--tea=True -> tea is True
--tea=False -> tea is False
--notea --> tea is False

[gokul-uf]

@vrv
this issue still persists as of TF 1.12.0
All the examples above work as expected
but ./test.py --tea False fails as it gives output have some tea . Is this still expected behavior?

[vrv]

TensorFlow's implementation of flags was replaced by absl: https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/platform/flags.py

So any flag related issues should probably be filed at https://github.com/abseil/abseil-py/issues now, unless there's evidence that the wrapper in TensorFlow is the issue.

[gokul-uf]

Thanks, I will raise it there

[MartinPeris]

I know this has been closed for a long time and tf.app.flags are not really supported. But defining tea like this will give you the expected behavior:

tf.app.flags.DEFINE_boolean('tea',None,'Soggy Leaves')