Pthread and misc header bugs on Solaris 2.7 Beta

[Wen-Jaw Jonah Tsai <jonah@cs.cmu.edu>]

Hi,

I got a test machine, a SparcEngine AXi (sun4u2) with Solaris 2.7 Beta
rev. 50. I sucessfully (probably) compiled gcc v. 2.8.1 on it, with some
arm twisting. Although I have not run test-suite on it yet, it seems to
work fine, so far.

However, when I tried to compile gdb v.4.17 on it, a few of the header
files clashes due to the resturcturing of Solaris 2.7's
/usr/include/sys/*.h. I solved those problems but pthreads presents a
major problem. Since this OS is a Beta, you may not be interested in
what the problems are. They may just go away in the official release of
Solaris 2.7. But I don't think your implementation which conforms to
Answerbook2 is quite portable and 2.7 is correcting that problem, with a
growing pain. So, I think you may have to deal with this problem when
the official release does come out. Here is why:

In sol-thread.c, you have provided implementation for functions like
ps_pdread(), ps_pdwrite() and pals. I suppose you intended to override
the default implementation of solaris pthread libarary. Or maybe it was
not provided and you are implementing it??? I don't know, I have another
machine w/ Solaris 2.4 instead of Solaris 2.6. Since you included
<proc_service.h>, I guess you are overriding it. But, the signature
clash is non-trivial to correct. In these functions, one of the
arguments "int size" (in your implementaiton) is declared as "size_t
size" in Solaris 2.7 Beta. These functions, in turns, calls rw_common()
and passes the "int size" along, which in turns calls a function table,
target_ops. Unfortunately, rw_common() and target_ops both uses "int
size" which is smaller than 64bit Solaris's size_t (typedef ulong_t
size_t). Solaris 2.7Beta does allow installation of 32-bit addressing
scheme instead of 64-bit addressing, but that's a pain.

The target_ops function table, I suppose, is used by other platforms as
well, which makes it almost impossible for me to do any trivial
modification and expect that it will run just fine. I did managed to
compile it by modifing these ps_pdread() & pals functions to conform to
Solaris 2.7 Beta's signature, compiled gdb-4.17 through without
compile-error, and, of course, expect it to fail on pthreads. It did!
All tests on pthreds failed.

However, since this is a "Beta" version of true 64-bit Solaris, the
thing should do these weird stuff, but at least it doesn't do any MS
beta quality, like Win98 Beta which wiped several of my friends'
harddrives. In the man page, it states exactly as proc_service.h says
"size_t size", but in AnswerBook2, it states "int size."

In the spirit of POSIX, shouldn't that be size_t, avoiding any
hard-wired data type like int? Anyway, these might not be the only
problems the pthreads failed.


Jonah Tsai