Bug 30620 – [GTK] Failing test fast/xsl/sort-locale.xml

• Status: UNCONFIRMED
• Component: WebKitGTK

[Philippe Normand]

That test fails on the debug slaves which have recent libxml2, libicu and libxslt versions.

[Martin Robinson]

Confirmed this is still failing. The sort order is incorrect in the output.

[Zan Dobersek]

Looking at this failure. At the moment, this is the expected collation per locale:
en:    peach, péché, pêche, sin
fr:    peach, pêche, péché, sin
fr-CA: peach, pêche, péché, sin

Based on the collation demo on icu-project.org[1], here's what the collations should really look like:
en:    peach, péché, pêche, sin
fr:

[1] http://demo.icu-project.org/icu-bin/locexp?_=fr_CA&d_=en&x=col

[Zan Dobersek]

(In reply to comment #2)
Comment #2, but complete.

Looking at this failure. At the moment, this is the expected collation per locale:
en:    peach, péché, pêche, sin
fr:    peach, pêche, péché, sin
fr-CA: peach, pêche, péché, sin

Based on the collation demo on icu-project.org[1], here's what the collations should really look like:
en:    peach, péché, pêche, sin
fr:    peach, péché, pêche, sin
fr-CA: peach, pêche, péché, sin

So, these seem to be the correct collations and should be used as the baseline. These are also the results I'm getting on Ubuntu 12.10.

Here's the collations from various builders though:
64-bit Release:
en:    sin, péché, pêche, peach
fr:    peach, pêche, péché, sin
fr-CA: peach, pêche, péché, sin

64-bit Debug:
en:    sin, péché, pêche, peach
fr:    peach, péché, pêche, sin
fr-CA: peach, pêche, péché, sin

And my local Ubuntu 12.04.1 setup:
en:    sin, pêche, péché, peach
fr:    sin, pêche, péché, peach
fr-CA: pêche, sin, péché, peach

Lots of different results. I'll see if adding a specific ICU version to jhbuild dependecies helps and come up with a patch. 

Overall, GTK, EFL and Qt mark this test as a failure (Qt skips it). There are baselines with correct collations for mac and chromium ports and a baseline with wrong collation for mac-snowleopard port. These baselines also need attention.

[1] http://demo.icu-project.org/icu-bin/locexp?_=fr_CA&d_=en&x=col

[Zan Dobersek]

Adding any specific ICU version to jhbuild dependencies didn't help.

r79912 introduced the correct baseline that the Chromium port uses and the changelog there details well that switching to ICU 4.6 caused the changes.
http://trac.webkit.org/changeset/79912

However, in both Ubuntu 12.04[1] and 12.10[2] the ICU versions are already 4.8.1.1, so I'm not sure what the exact problem with the test failing miserably on 12.04 might be.

The situation is similar on Debian Sid[3] (what the bots are mostly using), 4.8.1.1 is in use there as well.


[1] http://packages.ubuntu.com/precise/libicu48
[2] http://packages.ubuntu.com/quantal/libicu48
[3] http://packages.debian.org/sid/libicu48