In message <200909091217.n89chdr4018...@d06av03.portsmouth.uk.ibm.com>, Mark Hindess writes: > > In message <4aa0cfae.7070...@gmail.com>, Tim Ellison writes: > > > > On 03/Sep/2009 22:48, Mark Hindess wrote: > > > In release mode we currently discard windows .pdb files. This is > > > good because it keeps the size of the jre down but perhaps we should > > > place the .pdb files for jre/bin/*.dll in the hdk/ lib directory and > > > package them in our release HDKs? > > > > Or we could follow the other distros that (IIRC) include the source > > jars and native debug info in JDKs but leave them out of JREs. > > You are right. The JDK is a better place for them. (I suggested > HDK/lib as that is where the corresponding .lib files for windows dlls > are placed.) > > > I can imagine that having source JARs and .debug/.pdb files is useful > > as you develop your Java programs, but they are an unnecessary > > overhead when you are simply running them. > > Exactly. > > > > On linux we could also create smaller .so files (quite a lot smaller > > > in some cases!) by stripping them but maintain debug capability by > > > creating associated .debug files. See: > > > > > > http://sources.redhat.com/gdb/current/onlinedocs/gdb_17.html#SEC166 > > > > > > Something like: > > > > > > objcopy --only-keep-debug $dll $hdk/lib/$dll.debug > > > strip --strip-debug --strip-unneeded $dll > > > objcopy --add-gnu-debuglink=$hdk/lib/$dll.debug $dll > > > > > > so gdb can pick them up[0]. > > > > > > Do people think this is a reasonable idea? > > > > Yes, though as I say, I'd s/HDK/JDK/g > > I agree. > > > The next logical step would be to compile at a different optimizaion > > level for the JRE, but I think it is helpful to have the same debug > > info around so that it can be brought alongside a JRE to help debug > > it. > > I definitely intended that the debug files would match and would > automatically used by tools (like gdb) if they were (downloaded and) > placed on the library path, so having them produced from the same > original library is essential. > > I agree that providing non-optimized libraries might also serve a useful > purpose (essentially the native equivalent of running with -Xint) but that > is outside the scope for what I was trying to do at the moment. I was > concentrating on the first failure capture scenario at the moment. > > > We could even host the .debug/.pdb files on the website for debuggers to > > pick-up, ... > > Hmm... my gdb skills don't extend to that I'm afraid. > > -Mark.
I've committed this change in r818225. However, since committing I notice that some (very) old linux distros might not have the requisite support for this. If you happen to be unlucky enough to have a broken build because of this, then please use: ant -Dmake.arg="HY_CAN_LINK_DEBUG=no" ... to workaround this (and let me know what "objdump --version" says on your machine). I plan to support non-linux platforms by just copying the whole .so file with debug symbols to jdk/lib and stripping the version in the jre/bin directory. I'll probably handle old linux distros in a similar way. -Mark.
