Be forward compat with rootless lock files #3023
Conversation
|
(rust_highfive has picked a reviewer for you, use r? to override) |
|
I like this approach! My plan was to just not do anything, but this seems like the correct way to approach this to me. I would indeed like to one day phase them out (as they don't make sense). Looks like tests are failing though? |
matklad
force-pushed
the
rootless-lockfile
branch
from
ac3213b to
051fb17
Compare
Yep, the empty file is now a valid |
| if orig.starts_with("[[package]]") { | ||
| assert!(out.starts_with("[root]")); | ||
| out = format!("[[package]]{}", &out["[root]".len()..]); | ||
| } |
There was a problem hiding this comment.
This is interesting.
This is the case where we've read a Cargo.toml without [root] from the future. Presumably we want to preserve this rootless format.
But this approach may change the ordering of packages in the lockfile! This is not the best behavior. Hm, I think I can actually solve this...
matklad
force-pushed
the
rootless-lockfile
branch
from
051fb17 to
5e644df
Compare
|
Should be ready now |
|
Hm I think that an empty lock file should still be an error? If a package isn't listed as |
I tend to think that empty lockfile is valid, albeit makes little sense. Lockfile can specify less packages than there are members in workspace, if it is not up to date. An empty lockfile specifies zero packages. Previously at least one package was required because of the mandatory root. We can specifically bail for lockfile without packages, but I doubt that it will be useful in practice. Empty file is just a special case of a lockfile that needs updating. |
|
To confirm, though:
right? |
|
"tomorrow" being "when we actually decide to remove |
Yep, though looks like there is no tests which verifies current lockfile format specifically. |
|
Added a test for ^^^ here, if it is necessary: #3031 |
|
Ok, I'm gonna close in favor of #3031, thanks! |
Hi! So what is the plan regarding the
[root]key in the lock file?With workspaces it is meaningless, but not immediately harmful.
I think it should be possible to remove it eventually if we teach
Cargoto read rootless lockfiles now and switch to them several cycles later. Reading lockfiles with[root]should be supported forever of course.This PR adds ability to read rootless lockfiles, but they are then replaced by the current format in
write_pkg_lockfile. Should be easy to fix if we do want to evolve lockfile format.