The action is executed on every tag and creates a GitHub release. The
release contains a .tar.gz asset and a changelog consisting of all
commit messages since the previous tag. Commits with a commit message
starting with "Meta" are excluded from the changelog.
Setting the error option in bash bypasses error handling in the forgit
functions, because the script immediatley exits in case of an error.
Remove the option to make error handling possible.
Bash scripts automatically return the exit code of the last statement,
so standalone forgit will still return useful error codes.
From the standalone forgit command we call the according forgit function
by replacing a '_' with '::'. Since we have more complex function names
like 'forgit::cherry:⛏️:from::branch' we have to replace ALL
instances of '_' instead of just the first one.
fzf has the following exit codes:
0 Normal exit
1 No match
2 Error
130 Interrupted with CTRL-C or ESC
The forgit functions "log", "diff" and "stash show" are designed to
display information, not to perform any action. Hence the fzf exit code
130 should not be interpreted as an error, it is normal exit behavior.
Exclude this error value and return 0 in this case.
I added a bunch of "wraps" commands to our functions. This is basically a quick and easy way to get function completion for free. Users can now tab complete most commands that take an argument
When we want to perform a cherry pick, we had to pass the target branch
as an argument to gcp. Forgit, however, aims to make git selections
interactive where possible. Hence it seems natural to select the target
branch interactively as well.
Git stashes are named "stash@{x}", which contains the fzf placeholder
"{x}". In order to support passing stashes as arguments to forgit::diff,
we have to prevent fzf from interpreting this substring by escaping the
opening bracket.
The pager used on fzf enter key binding commands was hardcoded to
"LESS='-r' less". Transfer this to a variable in order to reduce code
duplication and make it configurable.
The git rebase command needs to be passed the parent commit of the one
you want to edit. When selecting a commit interactively via fzf, it
makes sense to automatically pass the parent commit of the selected one.
Within fzf it's otherwise quite hard to find the parent commit by hand
if you have found your commit via an fzf search string.
This reduces code duplication. We cannot implement this as a function,
because fish and zsh do not support exporting functions, but we need it
in subshells (see https://github.com/fish-shell/fish-shell/issues/6961).
Implement it as a variable in all shells instead.
The preview on cherry pick was accidentally broken since we included the
+/- information in the commit list. Fix this by using the correct string
index.
Git log displays the newest commits on top of the list, while git cherry
pick displays the newest commits at the bottom. Reverse the order of
git cherry pick so that it has the same order like git log.
The output of 'git cherry' which is used to build the fzf input list for
forgit::cherry::pick usually prefixes every line with a '-' for commits
that have an equivalent in the target branch, and a '+' for commits that
do not.
Previously forgit removed this information from the list. However, for
the actual cherry-picking process this information is relevant, so we
should keep it.
When using a fzf find string which maches multiple lines, the commits
could appear in a wrong order. Add "--tiebreak=index" to ensure that the
correct commit order is preserved.
The gcp (git cherry pick) command was the only command which did not
have a variable to set command-specific fzf options. Add the according
variable to source code and documentation.
* Adding support for git branch delete method in fish
* Add new doc to README
* Added options like gcb
* refactor: remove `--all` flag from gbd
* feat: sync forgit::branch::delete to bash and zsh
Co-authored-by: Wenxuan Zhang <wenxuangm@gmail.com>
