142870 - nsIFile should use UCS-2 instead of UTF-8

Status: RESOLVED FIXED

(Core :: XPCOM, defect)

Description

[Darin Fisher]

nsIFile should use UCS-2 instead of UTF-8.  UTF-8 seemed nice, but the
conversions between UTF-8 and the native charset are costly.  moreover, mozilla
is generally built-up on UCS-2 strings, and so we should really provide a
nsIFile interface that is more useful for the majority of the callers (e.g.,
javascript needs UCS-2).

so that's the gist of this bug... this issue was discussed via email/irc/aim,
and we decided that we should make the switch (back) to UCS-2 now before it is
too late.

the resulting patch is very straightforward.

Comment 1

[Darin Fisher]

i've posted an experimental linux build with these changes at

http://unagi.mcom.com/~darinf/builds/bug_142870

i've run through some basic tests with a profile located under a non-ASCII path,
and everything seems ok.

Comment 2

[Darin Fisher]

Attachment: v1 patch

Comment 3

[Darin Fisher]

Comment on attachment 82697 [details] [diff] [review]
v1 patch

got an r=dougt via email

Comment 4

[Darin Fisher]

Comment on attachment 82697 [details] [diff] [review]
v1 patch

sr=alecf via email

Comment 5

[Jim Crumley]

 This fix kills my local Sparc Solaris builds, though last night's nightly
(2002050722) still seems to work fine.

Details of my build environment: Sparc Solaris 7, gcc 2.95.3, and its a debug
build that finished early this morning.

If I start my build normally Iget 10 lines of :
WARNING: wctomb failed: possible charset mismatch, file nsLocalFileUnix.cpp,
line 1581

and then the build locks up without opening a window or dumping core.

If I run under gdb, a window never opens up, and when I hit control-c to
interrupt, I get:
0xff0c9068 in convert_ucs2_to_native (input=@0x1868, 
    result=0xffbeda78 "nxj¸þü¹°", resultLen=1024)
    at ../../dist/include/string/nsStringIterator.h:182
182               while ( n < 0 )

Should I open a separate bug for this?

Comment 6

[Darin Fisher]

jim: yes, please file a separate bug.  thx!

Comment 7

[Jim Crumley]

Ignore my last comment for now. On further review, I think my build is messed up
for other reasons. I'll try building with fresh source, and post a new bug if my
problems continue.

Comment 8

[Jim Crumley]

Comment 5 is not true.  It builds fine for me with fresh source.  Sorry about the 
false alarm.

Comment 9

[Darin Fisher]

mass futuring of untargeted bugs

Comment 10

[Boris Zbarsky [:bzbarsky]]

nsIFile is frozen.  This seems to have gotten fixed as it was being frozen.  No?

Comment 11

[Doug Turner (:dougt)]

old bug. nsFile is frozen and takes native and ucs2 encoded paths.