You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Dangling IFD pointer tags persisting after deleting their content #3459
When all tags within a SubIFD (such as GPSInfo or Exif) are deleted from ExifData, the corresponding pointer tag (e.g., Exif.Image.GPSTag, Exif.Image.ExifTag) remains in the metadata. This leads to the creation of empty or invalid IFDs when writing the image, causing inconsistency across platforms and failures in tests like geotag_test which fails quite frequently especially on macos.
To Reproduce
This is a non-deterministic bug, so it does not fail every time, however it fails quite frequently on geotag_test on macos on ci.
Unrelated (i.e. still saw the GPS tag failure once or twice on macOS after that was merged), so looks like this is an actual issue in the code somewhere?
I've been trying reproduce that test failure, but haven't been able to. I tried to copy the build instructions exactly, using clang-21 and meson. Then I ran the tests repeatedly but they never failed. @serghov: do you know how to reproduce the bug?
Activity
kevinbackhouse commentedon Feb 17, 2026
@kmilos: was this fixed by #3470?
kmilos commentedon Feb 17, 2026
Unrelated (i.e. still saw the GPS tag failure once or twice on macOS after that was merged), so looks like this is an actual issue in the code somewhere?
kmilos commentedon Feb 17, 2026
And the currently repeatable failure on macOS Intel only is I think unrelated as well, always fails on the network test...
kmilos commentedon Feb 17, 2026
@kevinbackhouse Yep, still happening intermittently:
https://github.com/Exiv2/exiv2/actions/runs/22096664410/job/63855411317#step:6:2040
and different to the known failure:
https://github.com/Exiv2/exiv2/actions/runs/22096660679/job/63860861082#step:5:152
kevinbackhouse commentedon Feb 17, 2026
I've been trying reproduce that test failure, but haven't been able to. I tried to copy the build instructions exactly, using clang-21 and meson. Then I ran the tests repeatedly but they never failed. @serghov: do you know how to reproduce the bug?
serghov commentedon Feb 18, 2026
@kevinbackhouse unfortunately I don't know how to reproduce it. And the fix isn't trivial either, I took a shot at it here, but don't like it