pipeline file: characters to allow in stage name

[skshetry]

Should we restrict stage name to certain characters? Maybe [a-z][A-Z][0-9]{hyphen, underscore}.

YAML keys can contain any thing, even spaces.

Allowing every kind of characters might create strange errors (which we might be unaware of).
If we disallow, what characters should we disallow?

[efiop]

The stricter - the better. We can always ease up on it.

[michael-ford]

Hey DVC team, not sure if this is the best place to do this, but as a DVC user I'd like to cast a vote for including periods as allowed characters. They are allowed in filenames in all modern OS that would run DVC, and while aren't the most common punctuation in file systems, are widely used.

I bring this up because I'm converting from 0.94 to 1.0+ and I commonly use periods in the names of .dvc files, which are now becoming stage names. Manually adding them to a dvc.yaml works, as does dvc repro, but dvc run complains.

The only compromise I can see (without being familiar with DVC implementation) is that periods are used in REGEX, but I generally never run into issues. Like I said dvc repro works fine with periods. Not sure if there are any other downsides though.

If you agree, I could potentially get my hands dirty and send a pull request. If one user's vote isn't enough, maybe in time others will chime in here as well.

[efiop]

@skshetry what do you think? ^

[skshetry]

Manually adding them to a dvc.yaml works, as does dvc repro, but dvc run complains

@michael-ford, yes, you are right that we only complain on dvc run. It's unlikely that we are going to start enforcing them.

You just have to add "." (period) if you want to in the following line, but I'll suggest you slugify them instead if you are migrating.

dvc/dvc/stage/__init__.py

Line 33 in bc56119

INVALID_STAGENAME_CHARS = set(string.punctuation) - {"_", "-"}

We are working on new features that will affect the dvc.yaml and we'll be relaxing some of the restrictions and/or enforcing them if we use these characters for some features. However, the period on dvc run won't be affected, and should be safe to use as-is, irrespective of the change on INVALID_STAGENAME_CHARS.

[efiop]

@skshetry Should we reopen/create new issue?

[michael-ford]

@skshetry thanks for the reply. From what I understand:

• obviously easy to add, but does your message mean you're not interested in changing it / PR? Meaning I could do it on my own fork if I want
• Allowing periods (with the exception of dvc run) will continue to work in the near-future updates