Windows builds in Chinese environment create generated files with Chinese characters

It appears that somewhere in the jdk8 builds, the system encoding is leaking into the build process. This makes builds unpredictable.

-kto

From Frank Ding:


On Jan 7, 2013, at 9:59 PM, Frank Ding wrote:

Hi Kelly,
 I have filed a bug whose internal review id is 2421470. It was filed with "Product/Category" being "JDK/JRE" and "Subcategory" being "Problems common to more than one tool". I am wondering since it only happens when building OpenJDK, is it eligible for a Java bug?
 In addition, do you have any idea on how to force java programs such as idlj to use ascii?

Best regards,
Frank

On 1/8/2013 3:42 AM, Kelly O'Hair wrote:
Did a bug report get filed for this issue?

-kto

On Jan 4, 2013, at 9:37 PM, Frank Ding wrote:

Hi Volker,
 Yes, I think so. The comment is pasted below.
/**
* org/omg/PortableServer/Current.java .
* 由IDL-to-Java 编译器 (可移植), 版本 "3.2"生成
* 从../../../../src/share/classes/org/omg/PortableServer/poa.idl
* 2013年1月4日 星期五 下午01时21分01秒 CST
*/

It's in Chinese, and it says when translated to English that "Generated by IDL-to-Java compiler(portable) version 3.2 and the date. You can also view a "normal" English one under openjdk folder "corba\gensrc\" after performing a build.

An relevant question is that in my env (export LANG=C), java output without any param always gives me Chinese help. Even though we may find out which env var or value jvm reads on startup, it could be impossible to change them. Any insight or idea on the mechanism is welcome.

Best regards,
Frank

On 1/4/2013 4:59 PM, Volker Simonis wrote:
This is just a wild guess, but perhaps idlj uses the value of some environment variables (or values derived from them - check System.getProperties()) which contain non ASCII characters? This could be something like PATH, HOSTNAME, USER. What exact characters are there in the comment and what kind of comment is it? How does this comment look on a "normal" system?

Regards,
Volker


On Fri, Jan 4, 2013 at 6:29 AM, Frank Ding <dingxmin@linux.vnet.ibm.com <mailto:dingxmin@linux.vnet.ibm.com>> wrote:

   Hi Kelly
     I investigated how local specific characters get into generated
   sources in corba module. Those classes are generated by following
   command idlj
   c:/openjdk/dep/jdk1.7.0_02/bin/idlj -J-XX:-PrintVMOptions
   -J-XX:+UnlockDiagnosticVMOptions -J-XX:-LogVMOutput -J-Xmx512m
   -J-Xms512m -J-XX:PermSize=32m -J-XX:MaxPermSize=160m -td
   "c:/openjdk/ojdk8_ojdk_739/../ojdk8_ojdk_739-debug/corba/gensrc"
   -i "../../../../src/share/classes/org/omg/PortableServer" -i
   "../../../../src/share/classes/org/omg/PortableInterceptor" -corba
   3.0 -fall -pkgPrefix PortableServer org.omg
   ../../../../src/share/classes/org/omg/PortableServer/poa.idl

     I checked idlj help but there is no encoding specific option. My
   locale environment vars are listed below

   $ locale
   LANG=C
   LC_CTYPE="C"
   LC_NUMERIC="C"
   LC_TIME="C"
   LC_COLLATE="C"
   LC_MONETARY="C"
   LC_MESSAGES="C"
   LC_ALL=

     Could you give me any hint about how to force idlj to generate
   ascii chars only?

   Best regards,
   Frank


   On 1/1/2013 12:47 AM, Kelly O'Hair wrote:

       In the past, the "-encoding ascii" was important, all the
       reasons I can't completely list right now. But it is important
       that regardless of the locale, the bits created during the
       build should be the same for everyone.
       The definition of "same" might not be bit for bit, but by
       minimizing the potential differences we have a fighting
       chance of measuring "the same".

       But my question is, how are any locale specific characters
       getting into generated sources? That's what we need to find out.

       Removing "-encoding ascii" is probably not the right answer,
       and if it is, will require some debate.

       -kto

       On Dec 30, 2012, at 9:25 PM, Frank Ding wrote:

           Hi
             I have an encoding problem when building openjdk 8 on my
           Windows 7. My windows is Chinese environment but I
           exported LANG=C in cygwin bash before building. The issue
           is that in module corba and jdk, some java classes are
           generated by program. They happen to contain Chinese
           characters in comment. However, they are compiled with
           explicit option "-encoding ascii" in makefile. This
           results in unrecognizable chars complained by javac
           (Error: encoding ascii unmappable chars) . I have a patch
           that removes all unnecessary "-encoding ascii" but I am
           not sure all its side effect. Shall I submit a bug?

             Best regards,
             Frank