Way to avoid repeating revset subexpressions

[pokey]

In this gist, I have to repeat a fairly verbose subexpression, and avoid it by defining the expression as a bash variable. It would be good if there were some way to avoid needing to do this. I have a couple ideas for how to avoid this problem

[pokey]

Define variables, as per #553 (reply in thread). This would give:

git query "all_ancestors = ancestors(parents(\"$branch\")) & branches() & draft(); all_ancestors - ancestors(parents(all_ancestors))"

[pokey]

Support a syntax that is more like jq. We could eg replace their | with >, and their . with input(). So for example, it would be:

git query "ancestors(parents(\"$branch\")) & branches() & draft() > input() - ancestors(parents(input()))"

[arxanas]

Personally, I'm in favor of a pipe expression. It's rather common when writing a query interactively that you want to do more stuff that you just thought of, so a postfix variable binding syntax would be useful.

Currently, both @ (from Git) and . (from Mercurial) mean HEAD. We could change . to mean "the result of the LHS of the pipe expression" with a default of HEAD when there is no pipe expression, which would work equivalently for all queries at present.

For the actual operator, I'm partial to |>, which is used in a few languages.

Proper variable binding would be useful if we get to the point of writing full scripts. The . syntax could be used for git query as well to indicate the input.

[pokey]

Support piping a list of commits into query and referring to it using eg input(). Then we'd have:

git query -r "ancestors(parents(\"$branch\")) & branches() & draft()" | git query "input() - ancestors(parents(input()))"