josh-filter --reverse on shallow clone creates unconnected branch

[thkoch2001]

When I wrote down my workflow for #1425 I first tried it on a monorepo clones with --depth 1. However the result of josh-filter [...] --reverse was a branch unconnected to the main branch of the monorepo.

I kind of understand that josh does not have enough information on a shallow clone to create the expected result. However I have a few questions:

• I'd like to understand how josh actually works. Splitting out a workspace from a monorepo does not seem too hard, but connecting a workspace back into the main branch seems like black magic. Could you help me understand how this works?
• Is there a way to detect that the result is not what the user wants and to produce an error instead of a faulty(?) branch?
• Is there a way to fetch just the necessary history to make this work instead of cloning the full repo history?

[christian-schilling]

When I wrote down my workflow for #1425 I first tried it on a monorepo clones with --depth 1. However the result of josh-filter [...] --reverse was a branch unconnected to the main branch of the monorepo.

I kind of understand that josh does not have enough information on a shallow clone to create the expected result. However I have a few questions:

* I'd like to understand how josh actually works. Splitting out a workspace from a monorepo does not seem too hard, but connecting a workspace back into the main branch seems like black magic. Could you help me understand how this works?

It is black magic ;)
Basically there are a few parts to this:

• Being able to invert filters so that all files are in the right place. For example the invert of :/a is :prefix=a, some filters invert to themselves (like ::file). This is why we have our own language for filters instead of arbitary scripts, to be able to define inversions.
• Finding the right point in the target branch to apply the changes to.
• Putting the resulting inverse mapped partial trees back into the full tree. This is essentally a merge but with some special cases.

* Is there a way to detect that the result is not what the user wants and to produce an error instead of a faulty(?) branch?

Definetly something to look into 🤔
Apart from that it always makes sense to inspect the results (like via PR) and not push them directly somewhere that matters.

* Is there a way to fetch just the necessary history to make this work instead of cloning the full repo history?

Not yet, but we have plans for this.