Why doesn't log <path> respect revsets.log?

[BatmanAoD]

log <path> restricts the list of commits to ones that modified the specified path(s). If -r is also given, the effect seems to be equivalent to appending & file(<path>) to the explicit revset argument.

But if -r is not provided, it appears that the revset used is just file(<path>). Why isn't it <revsets.log> & file(<path>), for consistency (and convenience)? (And of course the user could still use log <path> -r :: to get the current behavior.)

Relatedly, is there a way to explicitly use my configured revsets.log inside of a revset expression?

[martinvonz]

That's actually how it used to behave. It changed in #3410 and there's some discussion there. I was actually surprised myself when jj log README.md showed up empty because I had not edited in any commits in my default revset.

[BatmanAoD]

Huh, I don't actually see any rationale in that PR...and I see that the ticket was actually something I opened.

The new default actually seems much more surprising to me. Now, in a repo that's quite old, I get pages and pages of results when I expect just the changes from my branch. (My configured default revset is trunk()..@ | trunk().)

[martinvonz]

It's possible that just printing a warning as you suggested somewhere would be a better solution. I'm not sure. I'd like to hear some more opinions before we change it again.

[amiryal]

I'd like to hear some more opinions before we change it again.

Here 👋 thinking this new behaviour is confusingly inconsistent, although I haven’t experienced it first hand yet. My answer to #2475 would be: don’t change the output, but rather just vary the warning messages depending on whether X seems like a filename, a revset or both.