bug#40482: Test failure in tests/packages.scm after merging master into core-updates

bug#40482: Test failure in tests/packages.scm after merging master into core-updates

[Marius Bakke]

[-- Attachment #1: Type: text/plain, Size: 3055 bytes --]

On the current "core-updates" branch (commit
a533c5a1835cbeafaf595c4474e2ce6adde7de8d), one test starts to fail if
commits 190ddfe21e3d87719733d12fb9b5eb176125a49f and
a187cc562890895ad41dfad00eb1d5c4a4b00936 are applied:

--8<---------------cut here---------------start------------->8---
test-name: package-transitive-supported-systems, implicit inputs
location: tests/packages.scm:362
source:
+ (test-equal
+   "package-transitive-supported-systems, implicit inputs"
+   %supported-systems
+   (let ((p (dummy-package
+              "foo"
+              (build-system gnu-build-system)
+              (supported-systems
+                `("does-not-exist" "foobar" ,@%supported-systems)))))
+     (parameterize
+       ((%current-system "armhf-linux"))
+       (package-transitive-supported-systems p))))
expected-value: ("x86_64-linux" "i686-linux" "armhf-linux" "aarch64-linux" "mips64el-linux" "i586-gnu")
actual-value: ("x86_64-linux" "i686-linux")
result: FAIL
--8<---------------cut here---------------end--------------->8---

After some trial and error I found that simply removing the test added
by commit a187cc562890895ad41dfad00eb1d5c4a4b00936 will make the above
test pass!

Moving the test further down the file (after the one above) will _also_
succeed.  Apparently there is some side effect caused by running the new
test that leaks into the test mentioned above.

For completeness, here is the new test that makes the above fail.

--8<---------------cut here---------------start------------->8---
test-name: transaction-upgrade-entry, grafts
location: tests/packages.scm:151
source:
+ (test-assert
+   "transaction-upgrade-entry, grafts"
+   (with-build-handler
+     (const 'failed!)
+     (parameterize
+       ((%graft? #t))
+       (let* ((old (dummy-package "foo" (version "1")))
+              (bar (dummy-package
+                     "bar"
+                     (version "0")
+                     (replacement old)))
+              (new (dummy-package
+                     "foo"
+                     (version "1")
+                     (inputs `(("bar" ,bar)))))
+              (tx (mock ((gnu packages)
+                         find-best-packages-by-name
+                         (const (list new)))
+                        (transaction-upgrade-entry
+                          %store
+                          (manifest-entry
+                            (inherit (package->manifest-entry old))
+                            (item (string-append
+                                    (%store-prefix)
+                                    "/"
+                                    (make-string 32 #\e)
+                                    "-foo-1")))
+                          (manifest-transaction)))))
+         (and (match (manifest-transaction-install tx)
+                     ((($ <manifest-entry> "foo" "1" "out" item))
+                      (eq? item new)))
+              (null? (manifest-transaction-remove tx)))))))
actual-value: #t
result: PASS
--8<---------------cut here---------------end--------------->8---

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 487 bytes --]

bug#40482: Test failure in tests/packages.scm after merging master into core-updates

[Ludovic Courtès]

Hello,

Marius Bakke <mbakke@fastmail.com> skribis:

> On the current "core-updates" branch (commit
> a533c5a1835cbeafaf595c4474e2ce6adde7de8d), one test starts to fail if
> commits 190ddfe21e3d87719733d12fb9b5eb176125a49f and
> a187cc562890895ad41dfad00eb1d5c4a4b00936 are applied:
>
> test-name: package-transitive-supported-systems, implicit inputs
> location: tests/packages.scm:362
> source:
> + (test-equal
> +   "package-transitive-supported-systems, implicit inputs"
> +   %supported-systems
> +   (let ((p (dummy-package
> +              "foo"
> +              (build-system gnu-build-system)
> +              (supported-systems
> +                `("does-not-exist" "foobar" ,@%supported-systems)))))
> +     (parameterize
> +       ((%current-system "armhf-linux"))
> +       (package-transitive-supported-systems p))))
> expected-value: ("x86_64-linux" "i686-linux" "armhf-linux" "aarch64-linux" "mips64el-linux" "i586-gnu")
> actual-value: ("x86_64-linux" "i686-linux")
> result: FAIL
>
>
> After some trial and error I found that simply removing the test added
> by commit a187cc562890895ad41dfad00eb1d5c4a4b00936 will make the above
> test pass!

This is fixed by e85d4cecbe253e59a8a2a42b6ce427d96ff10534.

This has to do with memoization of things that are actually
system-dependent.  This is why you’d see this side effect: in one case,
the first thing memoized would be the armhf graph, in the other case it
would be the x86_64 one.

Concretely,

  guix build coreutils -s x86_64-linux -s armhf-linux

would return the right derivation for x86_64 and a bogus one for armhf
(one that depends on Mes, Gash, etc.).

Thanks,
Ludo’.