Bug 203250
| Summary: | PNG images produced for some reftests have different hash even when the pixels inside the image are the same. | ||
|---|---|---|---|
| Product: | WebKit | Reporter: | Carlos Alberto Lopez Perez <clopez> |
| Component: | Images | Assignee: | Nobody <webkit-unassigned> |
| Status: | NEW | ||
| Severity: | Normal | CC: | ap, bugs-noreply, robertma |
| Priority: | P2 | ||
| Version: | WebKit Nightly Build | ||
| Hardware: | Unspecified | ||
| OS: | Unspecified | ||
Carlos Alberto Lopez Perez
It happens for example with the test css/css-color/hsl-005.html from WPT.
This test can be run from WPT (via WebDriver) with the command:
./wpt run webkitgtk_minibrowser css/css-color/hsl-005.html
Or it can be run also with the usal layout tests runner, with for example:
Tools/Scripts/run-webkit-tests --release --gtk imported/w3c/web-platform-tests/css/css-color/hsl-005.html
The expected image produced by this test has a different hash than the one of the reference.
However, when you check the pixel differences inside the images you find that the pixels are exactly equal.
It has been suggested this may be related to some timestamp (or something similar) that the screenshooting code writes into the png image,
but I don't think that is the case because:
- The expected PNG image you get, even when different in hash from the original, is the same always
(if you run it at different times)
- If you copy hsl-005.html over hsl-005-expected.html (or viceversa) then the hashes of the images match always.
That indicates IMHO that the issue that causes different hashes is not on the screenshooting code itself.
This is currently not an issue when running the layout tests, because the tooling
fallbacks to do a pixel comparasion when the hash of the images don't match.
But it can be an issue for another test suites like WPT.
Related: https://github.com/web-platform-tests/wpt/pull/19825
It would be nice if the PNG image produced has the same hashes when the contents of the pixels are the same
| Attachments | ||
|---|---|---|
| Add attachment proposed patch, testcase, etc. |
Carlos Alberto Lopez Perez
On https://people.igalia.com/clopez/wkbug/203250/ I uploaded the two images produced for port GTK with this test.
I also have analysed the images with the imagemagick tool "identify -verbose" and then I can see there seems to be some different structures on both images, even some values like green, kurtosis, etc are different.
Pretty diff here: https://people.igalia.com/clopez/wkbug/203250/hsl-005-actual-expected.png.identify_diff.html
I'm not a expert in PNG images, not sure what that means.