129736 - Saving of file with a charset different from system locale fails without an error

Status: RESOLVED WORKSFORME

(Core :: Internationalization, defect, P2)

Description

[marina]

*** observed with 2002-03-08 build ***
in the trunk builds when saving a file (image or html) with a name  different
form the system default (say giving a ja name while on english system) your
saving operation fails without a warning or error message. This is not a
behavior of 6.2.1. When such an operartion fails the error message pops up
"saving of file failed" and you are returned to the Save as dlgbox. This makes
sense to the user and he undrestands that this is not a legitimate operation, so
the way the current builds behave is a regression.

Comment 1

[marina]

added keywords

Comment 2

[Roy Yokoyama]

marina: when do we start seeing this bug? 
adding nsbeta1

Comment 3

[nhottanscp]

This might be related to the xpcom native converter change, cc to alecf.

Comment 4

[Alec Flett]

I don't believe we actually changed any actual behavior with the nsLocalFile*
changes. Also, the change hasn't landed on linux yet.. what platform are we
seeing this on? 
more than likely, someone is using the non-unicode API when they should be using
the unicode API. The non-unicode api assumes that the incoming string is in the
system charset. If the non-unicode API is being called across a JavaScript
boundary, the unicode bytes will simply be chopped from a 16-bit character to an
8-bit character without any conversion.

I think more debugging is in order. Find out who is calling setLeafName or
setLeafNameUnicode

Comment 5

[marina]

<I don't believe we actually changed any actual behavior with the nsLocalFile*
changes. Also, the change hasn't landed on linux yet.. what platform are we
seeing this on?
-  am seeing this on Mac OS and Windows

Comment 6

[nhottanscp]

Note this only happens when you save with a filename which is different from the
system's defult. We expect to have a conversion error for that case. I think we
used to fallback to '?' if the conversion fails. 
For WideCharToMultiByte, it has a flag,
"WC_DEFAULTCHAR Replace exceptions with the default character during conversion."
I think we need to set that flag. 

Comment 7

[nhottanscp]

I tried this with my debug build. Saved as a Japanese file name on English
system. In nsLocalFileWin.cpp, WideCharToMultiByte actually mapps the
unconverted characters to '?'. So, the problem is in somewhere else.

Comment 8

[Frank Tang]

nsbeta1+, we should show a warning for file name that we cannot handle. 
Both Excel and WinZip do not handle Unicode file name but they DO show warning 

Comment 9

[Yuying Long]

More info.:
I tried to same a Japanese file name through comopser on SimpChinese WinXP.
N6.2.1: the file name shows fine and can be saved.
Recent trunk build: the file name will show as ".....html" but still can be saved. 

Comment 10

[marina]

in today's build i get a warning while saving an html file with the name
different from system encoding, but i don't get any warning for the image file.

Comment 11

[Kathleen :Brade]

Marina, Roy, or someone else:  Could you give me a more detailed example to test
on my Macintosh build?  What site should I go to?  Thanks!

Comment 12

[marina]

Try those steps:(win2k EN)
- go to netscape.com site;
- select an image and save it as an image with ja or cyrillic or any other
charset different from Western-8859-1;
- note: there was no warning, the file is not saved and it just fails.

Comment 13

[Frank Tang]

adt3 P2

Comment 14

[Roy Yokoyama]

marina: I tried to saving img file; but it displays an error.
(saving korean.filename in W2K-Japanaese)
Please try again.

Comment 15

[marina]

yes,saving the image file is working now ( i mean you do get an error). But try
to save a web page: try this:
http://www.time.com/time/covers/1101020401/fallen.html. First save it as
"fallen"- you'll see the fallen.html in the dir you specified. Then try "יייי"
on the english system; it works too, you'll get 'יייי.html". But if you'll try
to save with "cyrillicname.html" you won't get any error message and no file
with the given name will appear (even with ????) in the directory.

Comment 16

[Frank Tang]

Impact Summery
Impact Platform: windows only
Impact language users: ALL 560M 100% of internet users
Probability of hitting the problem:  LOW, users less seldom to use characters
not in their locale.
Severity if hit the problem in the worst case: File not saved. data lost without
warning.
Way of recover after hit the problem: save again with different file name.
Risk of the fix: unknown
Potential benefit of fix this problem: unknown

Comment 17

[Frank Tang]

I am considering nsbeta1- this one.

Comment 18

[Frank Tang]

this will only happen when the users use some characters that are not part of
the system locale charset. nsbeta1- 

Comment 19

[Roy Yokoyama]

making off my radar. -> 1.0.1 milestone

Comment 20

[Jungshik Shin]

This doesn't fail. Instead, characters outside the current locale is replaced with '_'. That's bug 359148.