Test heif/heif_read fails with libheif 1.18.0+

[loqs]

Describe the bug The test heif/heif_read fails with libheif 1.18.0 or newer which contain https://github.com/strukturag/libheif/commit/0b236060d0585d744e26e2f745cb18419a38ccb8. The failing test in gd appears to be triggerd by a new check in libheif that returns an error Image has no 'ispe' property from heif_context_read_from_memory_without_copy for tests/heif/label.heic.

To Reproduce Steps to reproduce the behavior:

 ./bootstrap.sh
 ./configure \
    --prefix=/usr \
    --disable-rpath
make
make check

Expected behavior All checks should pass.

Actual results The test heif/heif_read fails with:

FAIL: heif/heif_read
====================

dlopen: libavcodec.so.61: cannot open shared object file: No such file or directory
GD Warning: gd-heif context creation failed
heif/heif_read.c:23: Assert failed in <heif/heif_read.c:23>
FAIL heif/heif_read (exit status: 2)

Environment (please complete the following information):

• OS: Arch Linux
• Version rolling
• fontconfig 2.15.0
• libxpm 3.5.17
• libwebp 1.4.0
• libavif 1.1.1
• libheif 1.18.2

Additional context test-suite.log https://gitlab.archlinux.org/archlinux/packaging/packages/gd/-/issues/1

[cmb69]

Indeed, all four MinGW CI jobs are failing because of this.

I do not understand, though, what this test is really supposed to be testing. The comment is

https://github.com/libgd/libgd/blob/6e42ee80eb3bcf840c9fc03c6adc0550c1e02c63/tests/heif/heif_read.c#L2-L3

but apparently, the test checks for the opposite. Maybe @YakoYakoYokuYoku can clarify?

[YakoYakoYokuYoku]

Maybe @YakoYakoYokuYoku can clarify?

It was meant to check that libgd indeed supports reading HEIF images, can't remember why the me from the past chose such confusing words.

[YakoYakoYokuYoku]

I'm in doubt, because FFmpeg can read the image but libheif cannot. Said image was produced with a version previous to v1.12.0, possibly v1.10.0, which didn't add the elusive ispe property. This is an upstream issue with libheif, so an issue should be opened there.

[cmb69]

The issue has been resolved upstream, and the MinGW builds (using libheif 1.19.5-2) are green again (one fails for another test case), so I'd close this issue if the test wasn't failing for me locally with a self-built libheif 1.19.5 (using libde265 1.0.15). I'll check that ASAP.

[cmb69]

if the test wasn't failing for me locally

Actually another test fails for me (heif_im2im), so I consider this issue to be resolved.