Join GitHub today
GitHub is home to over 50 million developers working together to host and review code, manage projects, and build software together.
Sign upv2 feature: relax ordering requirements for flags/commands to improve upon cli ergonomics #1113
Comments
|
I'm tentatively in favor of this, we would just have to think about any potential unexpected impacts.
|
|
For my use case, I actually prefer the behavior that flags must be placed specifically next to their declaration site. To give an example of why this is important—I have an application called # Run tusk in verbose mode, tests in normal mode
tusk --verbose test
# Run tests in verbose mode, tusk in normal mode
tusk test --verbose
# Run both tusk and tests in verbose mode
tusk --verbose test --verboseI am not opposed to adding this as an option, but it is an option that I would personally keep toggled off in my applications. |
|
@rliebz I have thought about that problem but to even distinguish between which levels a given flag is defined in v2 you have to go thru hoops. I'm not sure whether that's even intended behaviour but I had to do something like this: app := &cli.App{
Flags: []cli.Flag{
&cli.BoolFlag{
Name: "verbose1",
Aliases:[]string{"verbose"},
},
},
Commands: []*cli.Command{
{
Name: "l1",
Flags: []cli.Flag{
&cli.BoolFlag{
Name: "verbose2",
Aliases:[]string{"verbose"},
},
},
Subcommands: []*cli.Command{
{
Name: "l2",
Flags: []cli.Flag{
&cli.BoolFlag{
Name: "verbose3",
Aliases:[]string{"verbose"},
},
},
Action: func(c *cli.Context) error {
fmt.Printf("v %+v\n",c.Bool("verbose"))
fmt.Printf("v1 %+v\n",c.Bool("verbose1"))
fmt.Printf("v2 %+v\n",c.Bool("verbose2"))
fmt.Printf("v3 %+v\n",c.Bool("verbose3"))
fmt.Println("new task template: ", c.Args().First())
return nil
},
},
},
},
},
}to even find a way to do it and considering using duplicate flag names is not even documented it looks like an implementation quirk rather than intended feature (or I am doing it wrong?). v1 had distinction between flag and global flag but honestly I made more errors in using them than actual useful functionality... |
|
I'm actually still on v1. I'm not sure what the current behavior is for this, or what the impetus for changing it was. |
|
@rliebz in v2 it would change some edge cases with using cli.GlobalType vs just cli.Type but v2 has no differentiation of that. |
|
I would love if this ordering requirement were relaxed, its my primary complaint about this library. |
|
I'd be interested in implementing it. Haven't looked at code yet but I'd probably also add a flag to opt in for "strict" ordering if it won't mess up code too much. |


Checklist
What problem does this solve?
CLI ergonomics when editing/adding flags to command are very bad.
Quick example, consider how many times you have to move cursor back in commandline to go from
to
vs
It also "reads" worse:
"output CSV on query project kittens on users with permissions to deploy:
vs
" query users that can deploy in project kittens, output csv"
Without forced ordering you can add a flag at the end at any point so ad-hoc usage of commands is much more streamlined, it also allows writing CLI scripts with arguments grouped by what makes more sense, not by how program author structured it.
Now the long winded example, consider the following:
Let's say app has a subcommand to query systems state
Which has
--filterstring option, and this command have subcommands for various parts of systemwith various sub-subcommand specific switches.
Let's say user wants to only see stuff about a certain project. No worries, we have
--filterflag for itUser now wants to see who has access to that project. But we have options just for that
So far so good. Let's see which of them can deploy the application
Okay, now let's compare it with another project...
Now we need to go back 4 words (4x
C-b) and we can start changing it. But we have few more projects to check? How about just move filter at the end ?Okay, we can't. Bit annoying but it is only few commands.
Right, it does what we wanted. Now onto writing that report. There is a
csvoutput format so it should be trivial.Oh, right, go C-b NINE times because that needs to be after a command but before subcommand.
Want to bump it to verbose or enable debug ? Go back nearly to the start of the line. Want to set environment or target server ? Same. Only real option to have convenient ability to change the parameters used in whole app is duplicating them as command-local variables.
The migration manual mentions something that "it is more POSIX-like" but POSIX conventions never had a notion of subcommands in the first place, the "stuff after options" was not a subcommand but also command's parameters (usually file to operate on), and even now most of the say
coreutilstools allow for having options in any place just for convenience (GNU'sls -la /tmpandls /tmt -lahave same effect for example)Solution description
Every
--optionin commandline should be considered for a flag, from closest "leaf" subcommand to the rootso
should be equivalent. Sometimes even having options before subcommand makes sense, like
Describe alternatives you've considered
Copying every flag to local "works" but it isn't exactly elegant, and messes up with generated help.