59268 - Autodetection should not override charset from cache and bookmark

Status: VERIFIED WORKSFORME

(Core :: Internationalization, defect, P3)

Description

[Eugene Savitsky]

1. Set autodetect to Russian
2. Load the URL.
3. Hit Reload.

See the Estonian - Estoniana, English - aaaEnglish?

Comment 1

[nhottanscp]

Reassign to [e-mail].

Comment 2

[bobj]

The Russian autodetector is detecting this page as Cyrillic/Russian (IBM-866).
The page is tagged with a <META> tag as windows-1250.
  <META HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=windows-1250">

Can someone verify all the text on this page is in one encoding?  It looks
like some of the alt text displays strangely (Latin1 garbage?) when viewed
as Windows-1250, but appears displays better (I cannot read Cyrillic, but
it looks like Cyrillic characters).

Comment 3

[Eugene Savitsky]

Trehe is a alt="" cyrillic text only. The main page is 1250.

Comment 4

[nhottanscp]

This could be related to bug 56689.
Auto detect should not affect pages with META charset.

Comment 5

[Shanjian Li]

I just met reproducible testcase that autodection is overiding meta tag. Hopefully this 
bug is caused by the same problem.\
How it happens: If a page has been loaded in right charset before, charset will be stored
in cache. Next time when you visit the page, meta charset is found, but because it is the 
same charset is the one in cache, so reload did not happen and the charset information 
is still marked as from cache. However, if charset detector detects a different result, it 
will take precedence of charset from cache and thus cause the problem.

Possible solutions:
In nsMetaCharsetObserver.cpp, if meta charset is found, webshell can be notified even though
the charset is the same. But this solution will cause a additional reload unless we did something
in webshell.

Comment 6

[Shanjian Li]

A better solution to this problem might be to lower the priority of charset  
autodectection. If we put kCharsetFromAutoDetection after kCharsetFromCache and 
kCharsetFromBookmarks, the problem will be resolved. Suppose page P is loaded 
using autodetection, and autodetect find charset C1. User found C1 is not the 
right one, and manually choose C2. If we rememver C2 is either cache or 
bookmark, we should not let autodetection override it next time.
Frank, what do you think?

Comment 7

[Frank Tang]

I think shanjian is right.

Comment 8

[Shanjian Li]

Attachment: proposed patch

Comment 9

[Erik van der Poel]

sr=erik

Comment 10

[Shanjian Li]

fix checked in, modified summary to better describe the bug.

Comment 11

[Shanjian Li]

Need to reopen this bug. After I adjust the priority between autodetection and 
cache, reload/autodetection change will not affect page encoding, that is a 
serious problem.

Comment 12

[Teruko Kobayashi]

Changed QA contact to [e-mail].

Comment 13

[Jaime Rodriguez, Jr.]

If this is a "serious problem", can we address it for nsbeta1, and set target
for M0.9.1

Adding nsbeta1 keyword.

Comment 14

[Shanjian Li]

My original fix cause the serious problem, but it has been backed out. 
I will only take care if I have time. For this reason, this might not be
a nsbeta1.

Comment 15

[Jaime Rodriguez, Jr.]

Marking as nsbeta1-.

Comment 16

[Shanjian Li]

I could not reproduce the problem any more. Resolve it as worksforme. 
Reopen if you still see the problem. I would appreciate if you can 
add more information about how to reproduce the problem or create a 
better test case. 

Comment 17

[Teruko Kobayashi]

I could not reproduce this in 2001-05-24-06 build.