Convert base dependencies to use sys_string_conversions instead of the ICU...

Copy the relevant parts of ICU to a new file base/third_party/icu/icu_utf.*
so we can do basic UTF8/16/32 conversions without linking all of ICU.

Change callers who used to call SysUTF8ToWide/SysWideToUTF8 in base to using
these new functions. I will remove the Sys versions of these functions in a
later patch.

TEST=none
BUG=none

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=28219

[agl, 2009-10-05 19:52:29]

LGTM

http://codereview.chromium.org/243102/diff/1/18
File base/string16.cc (right):

http://codereview.chromium.org/243102/diff/1/18#newcode82
Line 82: utf8.resize(str.length() * 4);  // Maximum length of the string in
UTF-8.
This isn't strictly true since 5 and 6 byte UTF-8 sequences are defined, but not
currently used. However, if an attacker could inject UTF-16 which decoded to
these invalid ranges, they might be able to trigger an overflow.

Later: I note that the function never actually emits > 4 bytes, so this is safe.

http://codereview.chromium.org/243102/diff/1/4
File base/sys_info_posix.cc (right):

http://codereview.chromium.org/243102/diff/1/4#newcode27
Line 27: //#include "base/string_util.h"
Remove commented line before committing?

http://codereview.chromium.org/243102/diff/1/17
File base/utf.cc (right):

http://codereview.chromium.org/243102/diff/1/17#newcode56
Line 56: if (utf8_length <= 0)
size_t is unsigned, so s/<=/==/

[brettw, 2009-10-06 23:18:02]

This is an almost completely new patch.

[brettw, 2009-10-07 00:28:19]

I added a check in the download manager UI test for a locale problem I ran into
testing this.

[agl, 2009-10-07 00:31:57]

LGTM. I still wonder if there are some files missing from the review, but what I
can see is fine.

[brettw, 2009-10-07 00:34:00]

Since this file is included from string_util, most files don't know it's changed
at all, which is why there are so few changes. There will be a number of
subsequent passes to break these dependencies.