Looks super interesting! Why not bundle it as a separate app rather than trying to get it merged into core. That, I think, will be a tough sell.
2 Likes
Practically impossible (nonwithstanding a theoretical possibility), it requires to become a default-ish column, like owner, idx, docstatus, etc. and hook into insert and map_doc, at the very least. Some things need to grow from the trunk 
ā This is such one thing.
Hmmā¦nothing I saw in the proposal canāt be done (and done better) encapsulated as an app, but itās certainly possible I missed something.
In any case, if this really can only be done in core, my opinion on this PR shifts from keen interest to hard opposition. No orchestration philosophy this opinionated should be implemented in the framework. If there are hooks you need that donāt currently exist, Iād support seeing those merged, but this is too much to force on every user.
Thatās just my opinion, of course, but based on the history of other similar proposals in the past, Iād be very surprised if the maintainers didnāt take the same stance. If I were a betting man, Iād wager the chances of something like this ever being merged are extremely low.
1 Like
Youāre absolutely right! Thereās a middle ground, that will emerge.
But please if you have concrete ideas how to implement it as an app (at reasonable cost or beyond), share your ideas in detail with me. I do appreciate them very much!
I probably just canāt see it atm, so Iāll need an eye-opener 
In any case, if this really can only be done in core, my opinion on this PR shifts from keen interest to hard opposition.
Hm? ā How would you expect me to handle such a position, gracefully?
No orchestration philosophy this opinionated should be implemented in the framework.
I canāt see much of an opinion on the PR thus far, though? Maybe the description lays out stuff that is in the realm of possibilities rather than the realm of necessities?
Iām not really sure what barriers you see. You might see barriers I donāt, to be sure.
You mention adding fields to the DocType definition schema itself, for example, but why would that be preferable to a simple linked table? What do you gain by polling every table definition rather than just the ones involved in a āFlowā?
It is a simple linked table. 
ā orthogonal to the core schema addition.
Thereās a simple guard (cached, and all) to ensure that only the ones who have flow tracing enabled are eligible for the implementing code path. We are talking about the runtime equivalent cost of the most basic equality check still without pythonās type casting.
You might see barriers I donāt, to be sure.
Show me the code!