[tulip-bug] Errors on macro definitions on kern_compat.h header

Maciej_Samsel@notes.amdahl.com Maciej_Samsel@notes.amdahl.com
Tue, 5 Dec 2000 09:59:04 -0800

I have a problem with compilation of the tulip.c and pci-scan.c files. The
GCC compiler reports the same problem on the macro definitions at line 173
and 177 of kern_compat.h

I use Caldera Systems  eDesktop 2.4 that is one of major distributions (I
guess second after Red Hat). It seems that I have all header files. Here is
the listing of what I get on the screen:

gcc -DMODULE -D__KERNEL__ -o6 -c tulip.o

In file included from tulip.c:162:
kern_compat.h:173: parse error before `0'
kern_compat.h:177: warning: parameter names (without types) in function
kern_compat.h:177: conflicting types for `mark_bh'
/usr/include/asm/softirq.h:101: previous declaration of `mark_bh'
kern_compat.h:177: warning: data definition has no type or storage class
kern_compat.h:177: parse error before `}'
gcc: Internal compiler error: program cc1 got fatal signal 11

By the way, I noticed that the kern_compat.h was modified (as the only one
file) several days ago (November 28, 2000). Could you please provide at
least one level of backup (old version) files so people have chance not
meet only new problems, but also revert to older, perhaps compilable
version of source code?

This is a bit annoying as some major networking hardware manufacturers
(Linksys and my card is LNK100TX) and major Linux distros like Caldera
Systems recommend your modules in tech support. This does not seem really
serious platform if at least testing is not in right place before release
(well, I did not expect tech. support from Open Source software). This is
enough frustrating that after 8-9 hours of developing commercial software
on freak'n Windows instead of having pleasure of working with Linux at home
I have to work as a developer again and playing with compiler instead for
example using great graphic applications or Internet browsing (OS is not
intended to be just developed but used for some productive work). I believe
that this is already not acceptable to many users that do not have
programming background. But at least if they need to recompile  some
software (which I believe should be provided as binaries to avoid whole
hussle - this is part of real modularization I guess) then lets make simple
source code packages and all header files (definitions inside) tested on
major disros of Linux. The compile script would be good too if possible.

Could you help me and I guess many other to come soon? Give me a tip what
is wrong with my compileation, fix definititions in the header file and/or
put older header files in some backup of previos versions online.