Closed Sentewolf closed 3 years ago
What is your OS and what is your shell?
Unless you do export data
, it should not be available to the child process's shell.
My apologies for wasting your time, you are completely right. It's shell scripts all the way down and somewhere up the chain was a "set -a" which I just learned exports variables.
I guess I'll have to try and fix a few other repositories (though the added check in validate_data_dirs.sh still seems a bit unnecessary).
I think it was using that method to check for repeated command line parameters.
On Fri, Apr 30, 2021 at 11:40 PM Sentewolf @.***> wrote:
My apologies for wasting your time, you are completely right. It's shell scripts all the way down and somewhere up the chain was a "set -a" which I just learned exports variables.
I guess I'll have to try and fix a few other repositories (though the added check in validate_data_dirs.sh still seems a bit unnecessary).
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/kaldi-asr/kaldi/issues/4511#issuecomment-830182525, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAZFLO6B5EGZKCQ6PHOV32TTLLFODANCNFSM434JZCBA .
In #4094 a check is introduced in validate_data_dir.sh to see if $data is an empty string. This check always fails if validate_data_dir.sh is called by make_mfcc.sh
It seems to me we should not check if $data is an empty string in validate_data_dir.sh.
Reverting a few lines in validate_data_dir.sh fixes the issue.
I tested this on RedHat and Debian. As far as I can tell, using the parent environment is the normal way for a shell to work so I expect this to fail in most environments.
Before I make a pull request to revert the changes made in #4094, I wanted to check why this extra validation was added, if it is needed in some other workflow, and if another way can be found to implement it.