ld big-file bug
utzoo!henry
utzoo!henry
Wed Feb 17 00:01:55 AEST 1982
To see an interesting explosion, try the following. Compile all the
pieces of f77pass1. Either work in a copy of the makefile or set up
a shell variable (say $all) containing the list of files in $(OBJECTS)
in the makefile. Do this:
ld -r -X $all -o junk.o
This has created one a.out with all the object modules in it. The -r
is to tell it we're not through yet, the -X is to strip trivia out of
the symbol table so later steps won't blow up on too many symbols. Run
"size" on junk.o, noticing that text+data (forgetting bss) is over 16 bits.
Now:
ld -X /lib/crt0.o -i junk.o -lc
This ought to yield f77pass1, right? Wrong, it lists _main as undefined
and then explodes, giving an "internal error" message. The fixed-up
loader supplied with VU Pascal explodes the same way. The fixes don't go
far enough.
The problem is almost certainly a 16-bit overflow in a file offset, but
I've tried some obvious fixes and they don't help (at least, not enough),
and I lack the time to explore ld.c in detail just now. Note that it shows
up only for monstrous INPUT files: ld has no trouble building f77pass1
if the input is broken up into the usual pieces. This makes the problem
a bit less important than it would be otherwise.
More information about the Net.bugs.v7
mailing list