Re: [PATCH] OCI8 link failure (Tru64/AIX) (#22324)

Jani Taskinen <sniper@iki.fi> wrote: > > I still don't think we should be linking with libocijdbc8 in any case. > Can you ask some Oracle support why their libs differ with same version > but in different OSes?? I can try to do so [1], although I don't see what that will buy us. From "Oracle9i Administrator's Reference" <http://download-east.oracle.com/docs/html/A97297_01/ch4_comp.htm#i26133>, chapter 4, "Using Oracle Precompilers and the Oracle Call Interface": | Custom Make Files | | Oracle Corporation recommends that you use the provided demo_product.mk | make files to create user programs as described in the specific product | sections of this chapter. | [...] | If you choose to use a make utility such as nmake or GNU make, be aware | of how macro and suffix processing differs from the make utility | provided with the platform. Oracle make files are tested and are | supported with the make utility for your platform. The "demo_product.mk" is in reality called demo_rdbms.mk - and of course it's virtually undocumented. | Oracle library names and the contents of Oracle libraries are subject to | change between releases. Always use the demo_product.mk make file that | ships with the current release as a guide to determine the required | libraries. Then they go on to explain their "symfind" utility, which is a shell script around "find", "grep" and "nm" - with platform-specific output [2]. | Correcting Undefined Symbols (Solaris Only) | | Oracle provides the symfind utility to assist you in locating a library | or object file where a symbol is defined. When linking a program, | undefined symbols are a common error that produce an error message | similar to the following: IMHO this is totally unacceptable. If they can't provide a well-defined set of interface libraries, they should at least offer some kind of "oci-config" or a few text files with -lsomelib -lanotherlib lines. But no matter how hard we wish they had some such thing, they don't have it at the moment - and even if they would suddenly start to clean up that whole mess, it would probably take a long time until their patches would be installed everywhere. I still think that the best solution would be to check if linking with -lcntsh alone is enough to get OCILobIsTemporary, and if not, check if it works with -locijdbc8. Regards... Michael [1]: My question in the "Oracle Technology Network" forum has not been answered for two weeks now, ditto in comp.databases.oracle.misc. [2]: Symfind (or nm) finds the OCILobIsTemporary definition (T) not only in libocijdbc8.so, but also in libclntsh.so and a few other libs: OCILobIsTemporary | 0000000000000000 | U | 0000000000000008 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ./lib/libsql8.a OCILobIsTemporary | 0004396969863776 | T | 0000000000000008 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ./lib/libnjni8.so OCILobIsTemporary | 0000000000005408 | T | 0000000000000008 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ./lib/libclient8.a OCILobIsTemporary | 0000000000005408 | T | 0000000000000008 OCILobIsTemporary | 0000000000000000 | U | 0000000000000008 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ./lib/libclntst8.a OCILobIsTemporary | 0004396969227952 | T | 0000000000000008 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ./lib/libocijdbc8.so OCILobIsTemporary | 0004396962545744 | T | 0000000000000008 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ./lib/libocijdbc8_g.so OCILobIsTemporary | 0000000000000000 | U | 0000000000000008 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ./lib/libsqlplus.a OCILobIsTemporary | 0000000000000000 | U | 0000000000000008 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ./lib/libordim8.a OCILobIsTemporary | 0004396966909952 | T | 0000000000000008 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ./lib/libclntsh.so This is on the Compaq/HP test drive with Tru64 5.1 and Oracle 8.1.7.