Record special field abstraction leakage

  • Open
  • quality assurance status badge
Details
2 participants
  • Leo Prikler
  • Ludovic Courtès
Owner
unassigned
Submitted by
Ludovic Courtès
Severity
important
L
L
Ludovic Courtès wrote on 26 Mar 2019 10:38
(address . bug-Guix@gnu.org)
87zhpiht6k.fsf@gnu.org
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’.
L
L
Ludovic Courtès wrote on 4 Apr 2019 13:26
control message for bug #34999
(address . control@debbugs.gnu.org)
87v9zurozb.fsf@gnu.org
severity 34999 important
L
L
Leo Prikler wrote on 24 Aug 2021 17:47
Re: Record special field abstraction leakage
bd16e5d6ff2c424945f883ac6892c5e7b0385c28.camel@student.tugraz.at
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:
Toggle quote (11 lines)
> 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?
From 1f38ff4c8b93cde533cf3d3f67358aafe9cf3dfa Mon Sep 17 00:00:00 2001
From: Leo Prikler <leo.prikler@student.tugraz.at>
Date: Tue, 24 Aug 2021 17:32:33 +0200
Subject: [PATCH] guix: records: let*-bind raw values, wrap them in
constructor.

This fixes the abstraction leakage mentioned in https://bugs.gnu.org/34999.

* guix/records.scm (make-syntactic-constructor)[field-bindings]: Bind to raw
value.
[field-value]: Always wrap the value.
[record-inheritance]: Wrap "inherited" values.
---
guix/records.scm | 17 ++++++++++-------
1 file changed, 10 insertions(+), 7 deletions(-)

Toggle diff (44 lines)
diff --git a/guix/records.scm b/guix/records.scm
index ed94c83dac..074f1650c8 100644
--- a/guix/records.scm
+++ b/guix/records.scm
@@ -153,7 +153,10 @@ of TYPE matches the expansion-time ABI."
#`(make-struct/no-tail type
#,@(map (lambda (field index)
- (or (field-inherited-value field)
+ (or (and=>
+ (field-inherited-value field)
+ (lambda (value)
+ (wrap-field-value field value)))
(if (innate-field? field)
(wrap-field-value
field (field-default-value field))
@@ -211,8 +214,7 @@ of TYPE matches the expansion-time ABI."
(map (lambda (field+value)
(syntax-case field+value ()
((field value)
- #`(field
- #,(wrap-field-value #'field #'value)))))
+ #`(field value))))
field+value))
(syntax-case s (inherit expected ...)
@@ -224,10 +226,11 @@ of TYPE matches the expansion-time ABI."
((_ (field value) (... ...))
(let ((fields (map syntax->datum #'(field (... ...)))))
(define (field-value f)
- (or (find (lambda (x)
- (eq? f (syntax->datum x)))
- #'(field (... ...)))
- (wrap-field-value f (field-default-value f))))
+ (wrap-field-value f
+ (or (find (lambda (x)
+ (eq? f (syntax->datum x)))
+ #'(field (... ...)))
+ (field-default-value f))))
;; Pass S to make sure source location info is preserved.
(report-duplicate-field-specifier 'name s)
--
2.33.0
(use-modules (guix records)) (define-record-type* <thing> thing make-thing thing? this-thing (name thing-name (thunked)) (name2 thing-name2)) (let* ((%thing (thing (name "foo") (name2 name))) (%thing2 (thing (inherit %thing) (name "bar")))) (format #t "thing1:~% name: ~a~% name2: ~a~%~%" (thing-name %thing) (thing-name2 %thing)) (format #t "thing2:~% name: ~a~% name2: ~a~%" (thing-name %thing2) (thing-name2 %thing2)))
?