saibotrash wrote:

Question: how exactly does WM extract? And: I think it should be ignored in such cases.

saibotrash wrote:

Current last 500 uploads by Rotatebot: 1 such a case (0.2 %)

How do we know which one is the 'correct' one?

As an aside, this is an upstream issue, since we don't extract the exif ourselves.

saibotrash wrote:

(In reply to comment #3)

How do we know which one is the 'correct' one?

We do not know - we can just guess or ask the user if we would be an interactive program. ;-) But it seems to be a bad idea to rotate. Do you cross the street if someone tells you "there is no car coming" and another person "there comes a car, stop!"?

(In reply to comment #4)

As an aside, this is an upstream issue, since we don't extract the exif
ourselves.

Which tool and which settings do we use? Is it reported?
At exiftool you need to add parameter -a to get also duplicate tags. If you leave it away it does not output both. Just one - I do not know which one or if it ignores both then.

(In reply to comment #4)

As an aside, this is an upstream issue, since we don't extract the exif
ourselves.

Which tool and which settings do we use? Is it reported?
At exiftool you need to add parameter -a to get also duplicate tags. If you
leave it away it does not output both. Just one - I do not know which one or if
it ignores both then.

We use PHP's exif module (which is a piece of crap for a variety of reasons, and I've contemplated just re-writing its functionality in php more than once).

I think exiftool goes with the first one, and we use the last one, but that's just a guess.

saibotrash wrote:

(In reply to comment #6)
I do now use this:
exiftool -IFD0:Orientation -b -a
Outputs all (-a) tags and therefore provides the full control over interpretation. ;-)
So for Commons the priority is maybe not more "high major".

saibotrash wrote:

In case it is interesting for someone: Current rate in Rotatebot's uploads: 0.5 % (out of 840 images). Until now all images had "Quicktime" in their EXIF info.