minor doc change proposal?

[John Clements]

This stack overflow post

https://stackoverflow.com/questions/53543191/what-is-the-different-between-filter-and-filter-map/53545115#53545115

… is written by someone confused by the documentation for `filter-map`. I went and read the documentation, and *I* was confused for about 30 seconds. I eventually proposed rewriting the existing

Returns (filter (lambda (x) x) (map proc lst ...)), but without building the intermediate list.

to

Returns (filter not-false? (map proc lst ...)), but without building the intermediate list, where not-false? can be defined as (lambda (x) x).

This text is kludgier, but I think that the use of (lambda (x) x) as “not-false?” is idiomatic and confusing. And yes, I realize that this suggestion probably applies to many places in the docs. Maybe I should just propose adding `not-false?` as a library function, defined as (lambda (x) x)….


John

[Gustavo Massaccesi]

This function is a already defined in a few libraries and it is called `true?` for example in https://docs.racket-lang.org/predicates/index.html?q=true#%28def._%28%28lib._predicates%2Fmain..rkt%29._true~3f%29%29

I think that `not-false?` is easier to understand, but `true?` is more idiomatic.

Gustavo

--
You received this message because you are subscribed to the Google Groups "Racket Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to racket-users...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Philip McGrath]

I would find `true?` confusing, since it really means "truthy." For example, in Rackunit, `check-not-false` has this behavior, whereas `check-true` checks that the result is really `eq?` to `#t`.

Personally, I think it might be better to clarify the documentation with more prose, rather than adding a new binding to the standard library. Maybe something like this?

Like (map proc lst ...), except that, when `proc` returns `#f`, that element is omitted from the resulting list. In other words, filter-map is equivalent to (filter (lambda (x) x) (map proc lst ...)), but more efficient, because filter-map avoids building the intermediate list.

-Philip

[John Clements]

I like your idea better than mine. Do you have time to make it a pull request? If not, I’ll do it.

John

[Philip McGrath]

I can do it.

-Philip

[Robby Findler]

What about using the function

(lambda (x) (and x #true))

And putting a short discussion of truthy and a link to elsewhere in the docs?

Robby

[Greg Hendershott]

On Fri, Nov 30, 2018 at 5:58 AM Robby Findler
<ro...@eecs.northwestern.edu> wrote:
> What about using the function
>
> (lambda (x) (and x #true))
>
> And putting a short discussion of truthy and a link to elsewhere in the docs?

Yeah. I feel like the `filter-map` doc would be fine with just that change.

Already it tells you `filter-map` is made of `filter` and `map` --
which are already clickable links right there!

I think really it's the `filter` doc that needs improving. It talks
about the pred function returning "a true value". Should it say "any
value except #f"?

(In the User's Guide, the only discussion of "truthy" I can find is
for things like `if`. The discussion is pretty much just "any value
except #f". Truthy is simpler in Racket than some other languages.
That's probably the only discussion needed -- to help people coming
from other langs? The only false value is #f. Everything else --
including 0 '() null -- is true. Right?)

[John Clements]

LGTM, but I should note that I already accepted Philip’s earlier suggestion. If you want to make further changes, go for it!

John