Display pathnames correctly in a Right-to-Left context (bsc#1128091) #873
Conversation
mvidner
force-pushed
the
rtl-pathnames
branch
3 times, most recently
from
a3017c2
to
8709057
Compare
| # then it would display as dev/sda1/ in a RTL context. | ||
| # Unicode BiDi Left-to-right embedding (LRE), | ||
| # then Pop Directional Formatting (PDF) | ||
| format("\u{202A}%s\u{202C}", s) |
There was a problem hiding this comment.
This is the actual fix. The rest of the PR 1) works around a libyui-ncurses bug 2) adjusts the tests.
There was a problem hiding this comment.
In the "after" screenshot I see boot/efi/. Is that the expected outcome? Whenever a user tries to use that path for anything, there will probably be an error since the system expects /boot/efi.
I understand this must be difficult for users of such an RtoL language (Arabic or Hebrew); if your natural reading direction is right-to-left, how is a user supposed to read such a technical path specification that is really still entirely Western-style? They might still start reading that part from the right, only then to realize that this is really left-to-right; but boot/efi/ seems wrong in both worlds.
Looks like I had opened the "before" screenshot twice. My fault. Disregard.
mvidner
force-pushed
the
rtl-pathnames
branch
from
8709057
to
6f0c46f
Compare
mvidner
force-pushed
the
rtl-pathnames
branch
from
6f0c46f
to
232240c
Compare
|
The PR rules demand screenshots... and that's a good thing because comparing the screenshots made me realize I have a bug in handling the root |
mvidner
force-pushed
the
rtl-pathnames
branch
from
232240c
to
995df7b
Compare
mvidner
force-pushed
the
rtl-pathnames
branch
from
995df7b
to
e9cd8b1
Compare
Problem: In Arabic, the device table displays /dev/sda1 as dev/sda1/ That's because / is in general a weak character but pathnames should be treated as overall left-to-right while isolated by the slashes. Solution: Use Unicode BiDi formatting characters in the source (creating a machine "translation" of the string).
There was a problem hiding this comment.
LGTM, and AFAICS it fixes this particular bug.
But I fear we will need this in more places, so in the long run, we will probably need those Bidi routines in a more central place (yast-yast2 package?) so they are accessible from other packages as well.
Also, we will probably need this in C++ as well; libstorage-ng comes to mind. We have places where libstorage-ng generates user-visible messages that include both translated text and paths: The action graph and the subvolume actions. They will have the same problem; unless the Arabic and the Hebrew translator inserted those bidi markers manually in each message.
Yup, that's likely, so I named the library so that it is easily movable without needing to change the usage places. I've even checked for collisions with other bidi gems (which are solving different problems, so merging is unlikely). |
|
Actually, most of the other places do not need special handling like this, because they can be solved with the usual special handling, which is using BiDi control characters in the translated messages. The storage table is special in having cells containing just filesystem paths, which noone has thought would need translation. |
| context "when the path is root /" do | ||
| let(:device_name) { "/dev/sda" } | ||
|
|
||
| it "returns the mount path with wrapped with LTR Isolate" do |
| # > | ||
| # > Memory: he said "(RLI)I NEED WATER!(PDI)", and expired. | ||
| # > Display: he said "!RETAW DEEN I", and expired. | ||
| module BidiMarkup |
There was a problem hiding this comment.
Is this yast2-storage-ng specific problem? I think this should be moved to yast2, just in case we need the same solution elsewhere....
There was a problem hiding this comment.
In the end it looks like it might indeed be specific to storage, but the file is designed to be easily movable once we find out we need it.
to make independent of y2partitioner, Also not to collide with https://rubygems.org/gems/bidi (which cares only about visual rendering)
add missing licenses to Ruby and Bash sources skip non-programs skip PDFs
|
✔️ Public Jenkins job #370 successfully finished |
Problem
In Arabic, the device table displays
/dev/sda1asdev/sda1/Solution
Use Unicode BiDi formatting characters in the source (creating a machine "translation" of the string).
Disabled for TUI: libyui-ncurses does not handle the BiDi characters well, truncating the strings. TUI does not have the general RTL direction anyway.
If you start off with a JEOS and only install the bare minimum of packages needed to test this, you may end up seeing dotted box characters in place of the BiDi controls. Installing noto-sans-fonts.rpm helps.
Testing
/Screenshots