Slice environment support

[rokf]

Hello,

I was wondering if you've ever considered adding slices as a supported environment type - currently only structs and map[string]any are supported - https://expr-lang.org/docs/environment.

I've checked previous issues and discussions but I haven't managed to find the reasoning behind this limitation anywhere. I apologise in advance if I've missed something and opened a duplicate.

Is there a technical reason behind the fact that slices aren't supported? A note in the documentation would be super helpful if this is the case.

My use case: let's say that I'd like to filter and transform arrays (slices) of data with Expr where the data comes in from arbitrary NDJSON, JSON array, and CSV files, where I would like to skip the creation of an intermediary variable into which I'd pass the slice of data. Having slices supported as an environment type would also mean that a single expression execution could transform the whole payload instead of re-running the expression on each row (I know that they can be compiled for efficiency and all that, but still) 🙂 Essentially I'm looking at Expr from the perspective of something like https://github.com/itchyny/gojq - as a replacement (alternative) for it.

Thank you!

[sknoyon324234234234]

I'm not a maintainer, but I think the reason is conceptual rather than technical: in Expr the environment is the namespace that identifiers are resolved against. foo in an expression means env["foo"] (map) or env.Foo (struct), and the language docs even describe $env as "a map of all variables passed to the expression". A slice has no names, so there'd be nothing for identifiers to resolve to, and no equivalent of jq's implicit . input at the top level.

The good news is that a one-key wrapper gives you exactly the jq-style "transform the whole payload in one run" you're after:

program, _ := expr.Compile(`map(filter(rows, .age > 30), .name)`,
    expr.Env(map[string]any{"rows": []any{}}))

out, _ := expr.Run(program, map[string]any{"rows": data}) // data = your decoded NDJSON/JSON/CSV slice

filter, map, reduce, groupBy, sortBy etc. all operate on rows in a single evaluation, and the program can be compiled once and reused for every file. If you want a jq-like feel, pick a short name like _ or data as the key.

A note in the environment docs saying "to evaluate over a list, put it under a key" would probably help others coming from jq, so worth suggesting as a docs PR too.