CoreGraphics configure test is incorrect

[barracuda156]

configure:17403: checking for CoreGraphics/CoreGraphics.h
configure:17403: /opt/local/bin/gcc-mp-16 -c -pipe -Os -arch ppc -I/opt/local/include conftest.c >&5
conftest.c:81:10: fatal error: CoreGraphics/CoreGraphics.h: No such file or directory
   81 | #include <CoreGraphics/CoreGraphics.h>
      |          ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~
compilation terminated.
configure:17403: $? = 1
configure: failed program was:
| /* confdefs.h */

This is on macOS 10.6, where CoreGraphics exists, of course. The problem is that it cannot be linked or included directly, since it is a part of ApplicationServices framework. So the configure test unnecessarily disables CoreGraphics-based loader.

[barracuda156]

Looks like the only issue with the code (i.e. aside of headers and frameworks) is unconditional usage of this https://developer.apple.com/documentation/imageio/kcgimagepropertygifdelaytime (10.7+).

[barracuda156]

This works for 10.6: https://github.com/macos-powerpc/powerpc-ports/blob/a1c6b29035e5a960ca5ac6350a03081d311674dc/graphics/chafa/files/patch-fix-CG-loader.diff

[hpjansson]

Hi, thanks for this. It looks like the patch is against configure, but it needs to be against configure.am, as the former is generated from the latter.

[barracuda156]

Yeah, it’s just the port uses configure when it exists to avoid extra build dependency on autotools.

[hpjansson]

Ok. I'll accept a reworked patch or PR, otherwise I will rework it sometime next week, probably. In either case I'd appreciate help testing the fix, since I don't have Mac hardware.

[hpjansson]

Is this something that requires versioning, i.e. would the fix make older MacOS installs fail to register?

[barracuda156]

Is this something that requires versioning, i.e. would the fix make older MacOS installs fail to register?

I am not sure I understand the question correctly.
The current version compiles fine, it just disables CG whenever configure test fails, so this will be a feature change, not build fix, and thus justifies a (micro?) version update.

[barracuda156]

Ok. I'll accept a reworked patch or PR, otherwise I will rework it sometime next week, probably. In either case I'd appreciate help testing the fix, since I don't have Mac hardware.

Let me ask someone to try this on 10.4/10.5 first, we need to be sure that whenever configure enables CG the code compiles (I won’t mind if on 10.4, for example, CG is never enabled, but we should not break then build at least).

[hpjansson]

Is this something that requires versioning, i.e. would the fix make older MacOS installs fail to register?

I am not sure I understand the question correctly.
The current version compiles fine, it just disables CG whenever configure test fails, so this will be a feature change, not build fix, and thus justifies a (micro?) version update.

Oh, I meant would it work on older MacOS versions too, or does the check need to be different there. Sorry for being unclear.