It's cleaner to have all the calls in one place, and this structure
will also make it easier to return other errors from the `dispatch()`
function.
Note that there's still a call to `process::exit()` inside `clap` when
it fails to parse arguments.
We didn't have any testing of exit codes on failure, other than
checking that they were not 0. This patch changes that so we always
check. Since we have the special exit code 2 (set by `clap`) for
incorrect command line, I've replaced some testing of error messages
by testing of just the exit code.
As part of this, I also fixed `jj branch --allow-backwards` to
actually require `-r` (it didn't before because having a default value
means the argument is considered always provided).
The function only needs a mutable reference (it doesn't store an owned
value anywhere), and this will enable the caller (i.e. `main()`) to
use the `Ui` instance after control returns from `dispatch()`.
The conditions for triggering Nix builds and other builds were
slightly different.
Nix builds triggered by PRs happened on PRs for any branch, not just
the `main` branch. That makes very little difference in practice
because PRs for other branches are very rare. Still, let's be
consistent. I decided to trigger the builds on PRs for any branch.
More importantly, Nix builds triggered by push were only done for
pushes to `master`, which is not what our main branch is called, so
those never happened.
The biggest difference in the API is that fields are now public. The
exception from that is `oneof` fields, which still require setters and
getters.
I couldn't measure any difference in performance. I didn't expect any
difference either, but it's good that it didn't seem to regress. I
timed `jj debug operation <some hash prefix>`, which will read the
whole operation log (to check that the prefix is unambiguous).
Tree merges can currently fail because of a failure to look up an
object, or because of a failure to read its contents. Both results in
`BackendError` because of a `impl From<std::io::Error> for
BackendError`. That's kind of correct in this case, but it wasn't
intentional (that impl was from `local_backend`), and we need to
making errors more specific for better error handling.
Apparently, I need to pass `--merge` option to use kdiff3 as a diff editor.
We could add `diff-editor-args` or extend `diff-editor` to a list of command
arguments, but we'll eventually add stock merge tools and the configuration
would look like:
[merge-tools.<name>]
program = ...
diff-args = [...]
edit-args = [...]
merge-args = [...]
Since 57ba9a9409, if the automatic import from git results in some
abandoned commit, that information gets recorded in the `MutableRepo`,
but I forgot to add a call to rebase the descendants. That causes a
failed assertion at
81a8cfefcb/lib/src/transaction.rs (L86). This
patch add a test showing that failure.
I think I copied the name `write_tree()` from Git, but I find it quite
confusing, since it's not clear if it write a tree to the working copy
or reads the working copy and writes a tree to the store (it's the
former).
Used to produce a message like this:
```
thread 'main' panicked at 'failed to run diff editor: Os { code: 2, kind: NotFound, message: "No such file or directory" }', src/diff_edit.rs:136:10
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
```
Now produces a message like this:
```
Error: Failed to edit diff: Error executing editor 'meld': No such file or directory (os error 2)
```
This adds `jj git push --change <revision>` which creates a branch
with a name based on the revision's change ID, and then pushes that
like with `--branch`. That can be useful so you don't have to manually
add the branch (and come up with a name for it). The created branch
behaves like any other branch, so it's possible to make it point to a
commit with a different change ID.
This is actually enough to fix#248, but I'll continue to work on
error handling for a while. I'd like to at least include the bad
object ID in this type of error messages.
Closes#248.
