Support user-defined verbs

[balki]

miller verbs are great. Easy to use, saves key strokes and easy to understand. It would be nice to extend those benefits to user-defined verbs.

Proposal

Users can create new verbs by registering them using below function. This function will be called in a miller file loaded using --[m]load.

register_verb(func, name, description, flags)

• name is string
• description is string for use in mlr help
• flags is an array of flag which is a map with keys short, long, description, type, default. At least one of short or long is required. Rest are of them are optional.
• type can be "bool", "single", or "multi". default is "single". "multi" means it can occur multiple times. e.g. verb -f arg1-f arg2 -f arg3. Accessed in the function as a list arg["f"] will be ["arg1", "arg2", "arg3"].
• when type is bool, it can be turned off with --no- prefix. i.e. --no-longname.
• bool flags can be coalesced. i.e. -xyz foo is same as -x -y -z foo.
• one of the flag can have long set to [POS] which refers to positional argument. type is always single
• func is a miller function that will be called with one argument args. It is a map with passed flags.

Example

register_verb(bconv, "bconv", "Convert bytes to KB, MB, GB or TB", [
{
"short": "s",
"long":"source",
"type":"single",
"default":"B",
"description":"Source unit. One of B, KB, MB, GB"
},
{
"short": "t",
"long":"target",
"type":"single",
"default": "auto",
"description":"Target unit. One of B, KB, MB, GB, auto. If auto, appropriate unit is chosen"
},
{
"short": "x",
"long":"suffix",
"type":"bool",
"default":true,
"description":"Add suffix after conversion. E.g. 2.4 MB"
},
{
"long":"[POS]",
"description":"comma separated list of fields to convert"
}
]
)

func bconf(args) {
if (args["suffix"]) { ... }
}

Namespace

1. Keep user defined verbs same namespace as standard ones. (similar to user defined functions).
2. Have user-defined verbs start with capital letter. Similar to go functions.
3. [my choice] Prefix user-defined verbs with u: or a different prefix. E.g. mlr --l2p u:bconv Memory,Swap then head.

[johnkerl]

@balki so the verb-implementation language would be the Miller DSL?

[aborruso]

It seems to me a great idea

…

______ Scusami per la brevità, ti sto scrivendo dal cellulare. website: https://medium.com/tantotanto 38° 7' 48" N, 13° 21' 9" E EPSG:4326 Il gio 17 lug 2025, 17:24 Balki ***@***.***> ha scritto:

*balki* created an issue (johnkerl/miller#1836) <#1836> miller verbs are great. Easy to use, saves key strokes and easy to understand. It would be nice to extend those benefits to user-defined verbs. Proposal Users can create new verbs by registering them using below function. This function will be called in a miller file loaded using --[m]load. register_verb(func, name, description, flags) - name is string - description is string for use in mlr help - flags is an array of flag which is a map with keys short, long, description, type, default. Atleast one of short or long is required. Rest are of them are optional. - type can be "bool", "single", or "multi". default is "single". "multi" means it can occur multiple times. e.g. verb -f arg1-f arg2 -f arg3. Accessed in the function as a list arg["f"] will be ["arg1", "arg2", "arg3"]. - when type is bool, it can be turned off with --no- prefix. i.e. --no-longname. - bool flags can be coalesced. i.e. -xyz foo is same as -x -y -z foo. - one of the flag can have long set to [POS] which refers to positional argument. type is always single - func is a miller function that will be called with one argument args. It is a map with passed flags. Example register_verb(bconv, "bconv", "Convert bytes to KB, MB, GB or TB", [ { "short": "s", "long":"source", "type":"single", "default":"B", "description":"Source unit. One of B, KB, MB, GB" }, { "short": "t", "long":"target", "type":"single", "default": "auto", "description":"Target unit. One of B, KB, MB, GB, auto. If auto, appropriate unit is chosen" }, { "short": "x", "long":"suffix", "type":"flag", "default":true, "description":"Add suffix after conversion. E.g. 2.4 MB" }, { "long":"[POS]", "description":"comma seperated list of fields to convert" } ] ) func bconf(args) { if (args["suffix"]) { ... } } Namespace 1. Keep user defined verbs same namespace as standard ones. (similar to user defined functions). 2. Have user-defined verbs start with capiltal letter. Similar to go functions. 3. *[my choice]* Prefix user-defined verbs with u: or a different prefix. E.g. mlr --l2p u:bconv Memory,Swap then head. — Reply to this email directly, view it on GitHub <#1836>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAHPD6QAVWB2EPGEQ5LR533I65ZTAVCNFSM6AAAAACBYFN5K6VHI2DSMVQWIX3LMV43ASLTON2WKOZTGIZTSOJZGE4TQMI> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[balki]

@balki so the verb-implementation language would be the Miller DSL?

Yes. I think miller DSL can handle most use-cases and would be fast enough. Using golang plugin system will require to recompile and not easy for many users.

This proposal does not cover verbs that need to look across multiple records. May be we can have another register function like
register_verb3(func_begin, func_process, func_end, name, desc, flags). func_end, can emit all records after aggregating.

[johnkerl]

Using golang plugin system will require to recompile and not easy for many users.

100% agree 😎