Build Failure on OpenSuSE 10.2 AMD 64 and Solution

[Kelly O'Hair]

Keep in mind that there are some java sources that are templates for
generated code, and there are several other ways java source is generated
during the build process. So a simple ant script just isn't going to be
possible. Look in the build/*/gensrc directory after a build for all the
generated java sources.

-kto

Andreas Schaefer wrote:
> Kelly O'Hair wrote:
>>
>>
>> Andreas Schaefer wrote:
>>> Hi
>>>
>>> Petteri Räty wrote:
>>>> Andreas Schaefer kirjoitti:
>>>>  
>>>>> /usr/lib64/gcc/x86_64-suse-linux/4.1.2/../../../../x86_64-suse-linux/bin/ld: 
>>>>>
>>>>> cannot find -lX11
>>>>>     
>>>> See that?
>>>>   
>>> Yes, I saw that and that what made me go installing X11. The strange
>>> thing is that there was only one X11.so (in ghostscript) and that after
>>> installing the additional RPMs I did not found any new X11.* files.
>>>
>>> It also seems only to happen on AMD64 because with OpenSuSE on Intel
>>> 32-bit I did not have to install these packages.
>>>>  
>>>>> After installing X11-devel and X11-Server I could go with the build 
>>>>> even
>>>>> though I have no clue what fixed it.
>>>>>
>>>>>     
>>>> You were missing the libraries so installing them fixed it. Of 
>>>> course it
>>>> would be better to check for them in the sanity check and fail with a
>>>> nice error message.
>>>>   
>>> Well, maybe but I am not quite sure if you can check for all
>>> dependencies. It is also difficult to keep up as you see with the
>>> AccColorT.h file that is not used to build but the sanity check looks
>>> for that file.
>>>
>>> All of that is a strong incentive for me to have a Java-source-only
>>> build using Ant. It is way to complicated to build the Java source with
>>> the make file.
>>>
>>
>> Not sure what ant vs. make has to do with sanity checks and native 
>> libraries
>> missing, seems like they are a pain no matter what you use to build.
>>
>> Just building the java source is worth investigating, but keep in mind
>> that we need to clean up the javac/recompile build ordering issue we 
>> have.
>> Right now, we use the boot jdk (stage 1) to build a javac from the jdk 
>> sources plus
>> enough runtime for this new first stage2 javac to run. These classes are
>> built by the boot javac, and usually aren't the final ones we want.
>> Then we take the stage2 javac we built and recompile everything javac
>> needs again, this gives us a new javac built from itself so to speak,
>> then we compile all the rest of the java source of the jdk.
>> The fact that we run what we build during the build process means that
>> any java sources that are backed up with JNI native libraries must
>> get their native code compiled after the java classes are built.
>> Since you need the javah headers to compile the native code, you need
>> the classes files first, then run javah, then run the native C/C++ 
>> compiler.
>> It's pretty convoluted.
>>
>> But we plan on changing this soon, and in such a way that the entire
>> jdk is built completely, and running the build at all isn't required 
>> to get
>> a build. In other words, we remove the javac recompile stage2 compiler
>> and use a prebuilt javac from the javac team.
>> These changes are in the works, and should make things simplier to
>> build, and faster too.
>> The recompile process will be removed and delegated to the javac
>> team.
>>
>> -kto
> My primary goal is to create a build script that only builds the Java 
> part to make life easier for "Java-only" developers. That said I agree 
> that it would not make much sense to have a build in Make and Ant but 
> for now that is were I will start just to find the pitfalls and traps. 
> Right now it takes way to long for my little change in the URLConnection 
> to build before I can fire up a test.
> 
> Maybe Java is called so because I can get a few cups of Joe when Java is 
> building - just kidding.
> 
> -Andy