elvy@harvard.ARPA (Marc Elvy) (07/04/84)
Subject: Bug in "asm" statement in 4.X BSD C compiler
Index: /lib/ccom and its neigbors
Description:
The "asm" statement is handled incorrectly in both the 4.1BSD and
4.2BSD C compilers. An "asm" after an "if" is placed within the
range of the "if"; braces make no difference.
Repeat-By:
Compile the following C program. This was prepared explicitly for
illustration purposes (which is why R11 is assumed), so please do
not waste your time hassling me about the philosophical problems
with "asm" statements.
main ()
{
register int flag = 0;
if (flag == 1)
flag = 0;
asm ("movl $666, r11");
}
The relevant portion of the assembly language produced follows.
clrl r11
cmpl r11,$1
jneq L16
clrl r11
movl $666, r11 /* Here is the "asm" statement. */
L16:
ret
Note that the "movl" should be AFTER the label, not before.
Unfortunately, I have not discovered why this is done, but I
suspect that the asm statement is dumped before the if statement
is completely processed.
Marc
Marc A. Elvy ( elvy@harvard.{arpa,uucp} )
Aiken Computation Laboratory
Harvard Universitychris@umcp-cs.UUCP (07/05/84)
This is what's called a ``known bug'' (or sometimes a ``feature'').
Everyone who uses asm() is supposed to know that extra null statements
are needed after if's. asm()s sneak through the back door of the
compiler and don't count as a statement. I suspect it's not worth
fixing. By the way, if you write
if (foo())
asm("hi there");
bar();
else
baz();
the call to ``bar'' is inside the ``if'', not outside. The compiler
won't complain about the ``else'' either.
I say, leave the ``feature'' in there as a warning to anyone who
wants to use asm()!
Not entirely serious,
--
In-Real-Life: Chris Torek, Univ of MD Comp Sci (301) 454-7690
UUCP: {seismo,allegra,brl-bmd}!umcp-cs!chris
CSNet: chris@umcp-cs ARPA: chris@marylandgreg@sdcsvax.UUCP (07/17/84)
The reason that this happens is that the "asm" statement is effectivly processed at the lexical level and just dumped into the output file. It would work correctly if the "asm" statment were turned into a semi-colon, have that processed, and THEN drop it into the output. This is the reason for the common hack to always preceed the "asm" statement by a semi-colon. In fact, if pre-processor substitution worked the way the C standard will have it (where a substituted macro loses its meaning during the evaluation; see the parallel discussion in net.unix-wizards), using #define asm ;asm would cause it to work in all cases -- at least I can't think of any where it wouldn't. -- -- Greg Noel, NCR Torrey Pines Greg@sdcsvax.UUCP or Greg@nosc.ARPA