Closed kiwi0fruit closed 6 years ago
This looks interesting -- it looks like a nice tool! The reproducibility problem is a big and important one. It's good to see this being addressed.
Thanks for response! Today I've learned about your Panzer. And it seems that you tried to solve alike problem that I did (but differently). So I was interested in your opinion.
I didn't do much on the reproducibility front -- it was more to add an extra layer of abstraction ('styles') so an individual user doesn't need to remember a million settings for document compilation via pandoc. They can just set one metadata value (style: Article
) and forget about it. Or they can combine 2 styles (style: [Article, BoldHeadings]
)
I didn't do much work though on how to reproduce the workflow produced by styles in a reliable way between users. Doing this properly means tracking all the dependencies of all programs invoked in document compilation, which (as you know!) is really hard.
I see. I didn't mind to dig into command line arguments at all so I went another way. I was irritated by copy-pasting terminal commands, adjusting names and like. So it's not a reproducibility that inspired me. I also needed extra level of abstraction. But I've chosen the one that is the easiest to understand it's workings and to tune in a moment.
Reproducibility came next. But I don't really track versions - I simply hope that they won't break anything and simply add used Pandoc and text filers to dependencies of python module.
You even have preflight
and postflight
options: this functionaliy together with filters themselves is what I initially wanted to put into some kind of profiles. So this ended with bash scripts that define everything.
But in your case it's metadata section that controls panzer app.
@msprev See the dream description.
What do you think of Pandoctools: profile manager of text processing pipelines (uses Pandoc filters and any text CLI filters)?