Because the linker for HP-UX itself must use the following shared library file！ And this file libu2comp.so is just bundled with the HP-UX aCC commercial software！
# gcc scsi_test.c -o scsi_test ld: Unable to load shared library "/opt/langtools/lib/hpux64/libu2comp.so" Fatal error. collect2: ld returned 1 exit status
Notes: There is also no GNU linker within the binutils for either IA/PA64 or PA32！
2) I think that the GNU linker should not use this shared library file that bundled with HP-UX aCC！ Notes: It seems that the GNU linker integrates with the binutils for the most UNIXes except HP-UX！
> 2) I think that the GNU linker should not use this shared library file that bundled with HP-UX aCC
That's not the problem. The problem is that ld(1) is not giving you an error when you're passing in the wrong architecture objects, archives or shlibs.
Eventually I found that I had made an embarrassing low-level mistake！ It seems that the original soruce filename(scsi_test.c) was wrong with some invisible charactors！ Now I just delete this original soruce file and then recreate it, so at last al l right！
But I still do not understand why gcc would report such ld libu2comp.so error under this situation！
gcc got misled by the unexpected content in the file and considered it an object file, so passed it to the linker; the linker in turn took it for an instrumented object file to be (re-)processed by the compiler backend - that is the case when it tried to load libucomp.so .
> It seems that the original source filename (scsi_test.c) was wrong with some invisible characters.
Yes, that would confused the driver if it didn't end in ".c". cc(1)/c99(1) work the same way.
> I still do not understand why gcc would report such ld libu2comp.so error under this situation
Because ld(1) is broken and doesn't make sure it is an ELF file.
> the linker in turn took it for an instrumented object file to be (re-)processed by the compiler backend - that is the case when it tried to load libucomp.so .
Wasn't this fixed years ago?
> how did you locate the problem?
Well, the source file was listed on the verbose ld output. :-)
> It seems that the original source filename (scsi_test.c) was wrong with some invisible characters.
Yes, that would confused the driver if it didn't end in ".c". cc(1)/c99(1) work the same way.
> I still do not understand why gcc would report such ld libu2comp.so error under this situation
Because ld(1) is broken and doesn't make sure it is an ELF file.
> the linker in turn took it for an instrumented object file to be (re-)processed by the compiler backend - that is the case when it tried to load libucomp.so .
Wasn't this fixed years ago?
At least now.
[ cathh02 tmp ] $ gcc -Wl,-V test.cx
ld: 92453-07 linker ld HP Itanium(R) B.12.63 IPF/IPF
ld: Mismatched ABI (not an ELF file) for test.cx
Fatal error.
collect2: error: ld returned 1 exit status
I got the same results with 12.58, couldn't find 12.61 though.
Now I wonder what the other error messages would be returned via ld with the same problematic source file if the system has installed with the shared library of libu2comp.so ？
Now I wonder what the other error messages would be returned via ld with the same problematic source file if the system has installed with the shared library of libu2comp.so ？
As long as the file is properly named, any errors that can be detected with the installed compilter would be reported already.
If the file is not named correctly, error reports may not be reliable. So don't bother about the errors when the file is incorrectly named.
From the compiler manual page cc_bundled(1):
Other Suffixes
All other arguments, such as those names ending with .o,
.a, or .so are taken to be relocatable object files and
are passed to ld (see ld(1)) to be included in the link
operation.
The behaviour of the linker for unrecognized input formats seems to be undocumented.
These days I have searched the knowledges for the instrumented object/code/execution！ But I just only found few information about them including the following reference link:
So could you tell me the significance of the "instrumented object file" concept in my this issue？
In your case there is no instrumented object file. Because of the bad filename, the compiler passed the source file to the linker. The linker mistakenly assumed it to be an instrumented object file. If your original problem goes away after renaming the source file, you don't have to bother about it any more.
