Add icudt.dll for Windows

Add icudt.dll for Windows

1. Generate and add windows/icudt.dll with the procedure outlined
in README.chromium.  It uses a out-of-tree copy of the upstream ICU
along with our custom-built icudtl.dat and a locally modified version
of makedata.mak.

  We used to have a separate build/ directory for VS solution/project files
to build icudtl.dll. Maintaining them is rather cumbersom now that we
want to update our ICU (major version changes) more frequently. 

  Note that icudt.dll is not used by default (icu_use_data_file_flag=1). 
The GN build still uses it by default and we should not break that build.

2. Add scripts/make_mac_assembly.sh to simplify the generation of the icu
data assembly source file for Mac.

3. Update README.chromium accordingly.

BUG=132145
TEST=None until icu is rolled to this version.

[jungshik at Google, 2014-04-25 20:39:04]

Can you take a look? 

This will be the last CL before rolling DEPS to icu52, I hope. :-)

[Mark Mentovai, 2014-04-25 21:00:32]

https://codereview.chromium.org/255943004/diff/1/scripts/make_mac_assembly.sh
File scripts/make_mac_assembly.sh (right):

https://codereview.chromium.org/255943004/diff/1/scripts/make_mac_assembly.sh...
scripts/make_mac_assembly.sh:11: .globl _icudt52_dat
I thought we were going to get rid of the version number from here too.

https://codereview.chromium.org/255943004/diff/1/scripts/make_mac_assembly.sh...
scripts/make_mac_assembly.sh:21: tail -n +10 ${LINUX_SOURCE} >> ${MAC_SOURCE}
The fixed “10” is weird, there’s no check to see whether too much or too little
got chopped off. You could just organize this as a patch, then at least if the
input data changes, there’d be a signal that manual attention is needed.

[jungshik at Google, 2014-04-26 00:20:02]

Thanks for taking a look. Please, take another look.

https://codereview.chromium.org/255943004/diff/1/scripts/make_mac_assembly.sh
File scripts/make_mac_assembly.sh (right):

https://codereview.chromium.org/255943004/diff/1/scripts/make_mac_assembly.sh...
scripts/make_mac_assembly.sh:11: .globl _icudt52_dat
On 2014/04/25 21:00:32, Mark Mentovai wrote:
> I thought we were going to get rid of the version number from here too.

That does not earn us anything. We wanna get rid of the version number from the
filenames so that we don't have to track down every single reference of icudtXX
in our build/source files when updating icu.(e.g. http://goo.gl/Dhtdxm in skia's
v8.gyp) 

OTOH, all the icu symbols carry version number anyway (to get rid of them will
require an additional patch and will add a maintenance burden [2] ) and
icu_util.cc (in chrome tree and v8 tree) has the following (no need to change
manually):

#elif ICU_UTIL_DATA_IMPL == ICU_UTIL_DATA_SHARED
#define ICU_UTIL_DATA_SYMBOL "icudt" U_ICU_VERSION_SHORT "_dat"
#if defined(OS_WIN)
#define ICU_UTIL_DATA_SHARED_MODULE_NAME "icudt.dll"
#endif


So, I'd just leave it alone here.

[2] To get rid of the version number from icudt.dll may require more 'hackery'
on Windows. (getting rid of that in the assembly source is easy)

https://codereview.chromium.org/255943004/diff/1/scripts/make_mac_assembly.sh...
scripts/make_mac_assembly.sh:21: tail -n +10 ${LINUX_SOURCE} >> ${MAC_SOURCE}
On 2014/04/25 21:00:32, Mark Mentovai wrote:
> The fixed “10” is weird, there’s no check to see whether too much or too
little
> got chopped off. You could just organize this as a patch, then at least if the
> input data changes, there’d be a signal that manual attention is needed.

How about patch set 3? I detect '^icudt' in the Linux assembly (which is at the
end of the header) and count # of lines up to that point. 

As you said, with a patch, I have to manually fix it when the first few lines of
the body change (which happens often with the data update).

[Mark Mentovai, 2014-04-26 00:25:53]

That’s less fragile, LGTM.

[jungshik at Google, 2014-04-28 18:01:01]

On 2014/04/26 00:25:53, Mark Mentovai wrote:
> That’s less fragile, LGTM.

Thank you. Checked in manually in two parts because codereview/appengine are
'misbhehaving'.