brian@uw-june.UUCP (11/23/86)
Has anybody managed to do this? (is anybody even running
SUN 1.1 anymore?)
After creating a scratch directory on a VAX (where things run
fine), and using the size alignment files (as generated by
szal.c), I am having the following (seemingly unrelated) problems.
1. Two files in the scratch directory cause the sun compiler
to croak when the ..c files from the scratch directory are
transferred to the sun from the vax. These are dcl..c and
simpl..c. Sun's 1.1 C compiler complains (something like)
line xxx: compiler error: no table entry for op REG
The problem stems from this expression:
(strcuct expr *)(_type_tsizeof( sdkhfk ) / someInteger );
which, on the face of it, seems to be coercing the result of
an integer division into a pointer. While this is allowed on the
vax (don't ask me why), Sun's 1.1 cc won't swallow it.
You can test your compiler with something like:
struct A { int x };
main() {
struct A* ap;
int XX;
ap = (struct A*)(XX/32);
}
If it compiles... your compiler works.
The "fix" was the following in the makefile
sed 's/(struct expr \*)\((_type_tsizeof (.*\/\)/\1/' src/dcl..c > t..c
mv t.c src/dcl..c
for both dcl..c and simpl..c
This got things past cc and got make scratch working, giving me a cfront.
Ambitiously, I decided to see if I could make cfront from the c++/src
directory directly. Everything hummed along nicely until cfront was handed
main.c:
CC main.c:
"main.c", line 451: internal <<cfront 05/20/86>> error: bus error (or something nasty like that)
***
Error code 1
Stop
This had me a little worried. I tried my (from scratch) cfront on
the usual test suite (#include <stream.h> main(){cout << "hello world\n");}
and it worked just fine. No core dumps, no segv's... nothing.
I then tried cfront on a medium size c++ program (~2000 lines) and
everything compiled and ran great. I then figured... so what
if I can't get it to make itself... I just won't ever erase
the binary. This is not a good thing to assume since the
next thing I ported (about 10K lines) (which runs great
on a VAX) gave me a bus error on the last line of some function
(line 451 in c++/src/main.c is also the last line of a function which
caused cfront to croak (but at least catch itself))
Since I am not sure why dcl..c and simpl..c would want to
coerce results of integer divisions into pointers, I can't
be sure if my sed expression "broke" things. I know that this
stuff runs on sun 3's and I know that the sun 3 compiler
will accept (struct x*)(int/int)... but that could just
be a coincidence.
Anybody else have any problems porting this thing to sun 2's
running BSD UNIX 4.2 release 1.1.
--
brian@june.cs.washington.edu Brian Bershad
{ihnp4,decvax,ucbvax}!uw-beaver!brian Dept. of Computer Science, FR-35
Seattle, WA 98195wescott@sauron.UUCP (Mike Wescott) (11/25/86)
In article <1618@uw-june.UUCP> brian@uw-june.UUCP (Brian Bershad) writes: > Has anybody managed to do this? (is anybody even running > SUN 1.1 anymore?) ... > This got things past cc and got make scratch working, giving me a cfront. > Ambitiously, I decided to see if I could make cfront from the c++/src > directory directly. Everything hummed along nicely until cfront was handed > main.c: > > CC main.c: > "main.c", line 451: internal <<cfront 05/20/86>> error: bus error (or something nasty like that) > *** > Error code 1 > Stop This is due to a bug in print.c and shows up on some 68000 machines that do not force long alignments (68000's are generally happy with 2-byte alignments for long words, 68020's perform better with 4-byte alignments but will work even with odd-byte address for longs). The fix was posted by me just yesterday. Send me e-mail if it doesn't show up at your site soon. -Mike Wescott ncrcae!wescott -- -Mike Wescott ncrcae!wescott