GLUT should not have been dropped from nightly

[andreasabel]

Build flags seem to be ignored when judging whether a package can stay in nightly.

GLUT was dropped from nightly, because of a bound random < 1.3, but random is only used in some executables (examples) rather than the GLUT library. These examples are only built with flag BuildExamples, which is off by default.

Thus, I think GLUT (and its dependants like gloss) should not have been disabled.

stackage/build-constraints.yaml

Line 6126 in c671e09

- GLUT < 0 # tried GLUT-2.7.0.16, but its *executable* requires random >=1.0 && < 1.3 and the snapshot contains random-1.3.1

stackage/build-constraints.yaml

Line 6170 in c671e09

- Vis < 0 # tried Vis-1.0.0, but its *library* requires the disabled package: GLUT

stackage/build-constraints.yaml

Line 6957 in c671e09

- gloss < 0 # tried gloss-1.13.2.2, but its *library* requires the disabled package: GLUT

stackage/build-constraints.yaml

Line 6964 in c671e09

- gloss-rendering < 0 # tried gloss-rendering-1.13.2.1, but its *library* requires the disabled package: GLUT

With resolver: nightly-2025-07-23, stack build succeeds for GLUT.

It seems that the dependency checker of Stackage does not take build flags into account, does it?

References:

• random-1.3.1 #7648
• v 2.7.0.16 revision 4: allow random-1.3, bump CI to GHC 9.12 haskell-opengl/GLUT#36

[juhp]

Thanks for reporting this - wonder if https://github.com/bergmark/commenter could be improved to handle this better

[juhp]

I generally try to look over the output of commenter before committing, but the random-1.3 lift was really large so I guess I missed detail like this: anyway thanks for catching and reporting.

I opened a commenter ticket to track this.