[Swan-dev] Build fails with -Werrror on i686 after commit 72636a3df4ecac7eca85bb317b3c3d57b01d6c31

[D. Hugh Redelmeier]

| From: Tuomo Soini <tis at foobar.fi>

| On Fri, 18 Nov 2016 10:29:20 +0200
| Tuomo Soini <tis at foobar.fi> wrote:
| 
| > Error message from build:
| > 
| > /builddir/build/BUILD/libreswan-3.18/programs/pluto/rcv_whack.c: In
| > function
| > 'writewhackrecord': /builddir/build/BUILD/libreswan-3.18/programs/pluto/rcv_whack.c:210:2:
| > error: right shift count >= width of type [-Werror] header[1] = now >>
| > 32; ^ cc1: all warnings being treated as errors

I'm sorry about adding that unportable code.  I'm so used to
degenerate cases working as they should.  But C does not require
32-bit shifts to work on 32-bit systems.

[I once found a bug in the 8086 processor because I counted on a
particular degenerate case working.  Intel fixed the bug (at my
prompting) by changing the specifications of x86 hardware.]

| DHR: are you be ok with easy fix?

-	header[1] = now >> 32;
+	header[1] = ((long long)now) >> 32;

I would prefer the following if it doesn't generate warnings (it is
perfectly legal):

	header[1] = (now >> 16) >> 16; /* >> 32 not legal on some systems */

(The parentheses are redundant but may help the reader in this
somewhat peculiar case.)

What warning might be generated?  On a system with 32-bit time_t and
32-bit int, the two shifts together will always produce a 0 result.
That's what I want but the compiler might think that the complex way
of generating 0 is a sign of insanity.

Tuomo: can you test this on a 32-bit system?  I don't have any at
hand.

If that causes the compiler to whine, I would prefer:

	header[1] = ((u_int64_t)now) >> 32;