#34999 Record special field abstraction leakage

Message #5 received at submit@debbugs.gnu.org (full text, mbox, reply):

From: Ludovic Courtès <ludo@gnu.org>

To: bug-Guix@gnu.org

Subject: Record special field abstraction leakage

Date: Tue, 26 Mar 2019 10:38:11 +0100

The changes I made in version-control.scm and gnucash.scm in commit
e6301fb76d0a8d931ece2e18d197e3c2cc53fc6c revealed an abstraction leakage
I wasn’t aware of: there’s a pattern where users “see” that thunked
fields are thunked:

  (package
    ;; …
    (inputs …)
    (arguments `(foo bar ,(inputs) …)))  ;<- here ‘inputs’ is seen as a thunk

Fortunately I could only find two occurrences of this and this use case
is more elegantly replaced by:

  (package-inputs this-record)

… which also has better semantics.  It’s remains a bug, though.

Ludo’.

Message #10 received at 34999@debbugs.gnu.org (full text, mbox, reply):

From: Leo Prikler <leo.prikler@student.tugraz.at>

To: Ludovic Courtès <ludo@gnu.org>, 34999@debbugs.gnu.org

Subject: Re: Record special field abstraction leakage

Date: Tue, 24 Aug 2021 17:47:30 +0200

[Message part 1 (text/plain, inline)]

Hi Ludo,

I think I have found out why users see the thunked fields as below.
Am Dienstag, den 26.03.2019, 10:38 +0100 schrieb Ludovic Courtès:
> The changes I made in version-control.scm and gnucash.scm in commit
> e6301fb76d0a8d931ece2e18d197e3c2cc53fc6c revealed an abstraction
> leakage
> I wasn’t aware of: there’s a pattern where users “see” that thunked
> fields are thunked:
> 
>   (package
>     ;; …
>     (inputs …)
>     (arguments `(foo bar ,(inputs) …)))  ;<- here ‘inputs’ is seen as
> a thunk
The issue is that for constructing the records, we let*-bind the field
names to their values before calling the constructor.  In these let*-
bindings the fields are already wrapped, e.g. inputs will be bound to
the value that the record field inputs will have, not to the raw value.

I've attached a patch to fix this issue as well as a MWE to try it out.
I'm not sure about the broader semantics of this patch, though.  I fear
that exposing raw values through let-binding probably eliminates the
delayed/thunked nature of said fields in some ways.  WDYT?

[0001-guix-records-let-bind-raw-values-wrap-them-in-constr.patch (text/x-patch, attachment)]

[34999-mwe.scm (text/x-scheme, attachment)]

Send a report that this bug log contains spam.