When should we write separate scripts and when should we use configfiles

[mvdbeek]

That discussion came up in galaxyproject/tools-iuc#1535.

Sometimes we need a helper script.
This can either be a script next to the tool xml, or we can template out a script
using the <configfile> mechanism.
We do have examples for both, but no clear consensus on when to use which mechanism.

The arguments for a script as separate file on disk are:

• you need to mentally "parse" the galaxy variables and cheetah stuff from the script
• you can't test your script independently from the galaxy tool
• you can't step in with a debugger
• you can't have code-highlighting / linting etc.
• Putting a script in a makes it impossible to run static checks/linting on it or to run something like 2to3
• may require extra escape mechanisms as code could be directly injected var <- "{$var}" then $var may not include quotes etc.

The arguments against (for including it as <configfile> in the tool wrapper) it are:

• we need to write a separate argument parser for the script that duplicates information
• can be used as workaround if command length exceeds the shells maximum length

(Feel free to edit/add arguments)

[mvdbeek]

I guess I'd like to see a rule of thumb like "if dealing with arguments takes up less than ~ 20 % of the code, use a separate script".

[yhoogstrate]

I don't see an added value of using config files for the purpose of generating <command>-replacing code (unless there is technically no other possibility).