rvinterf/etmsync/fsnew.c: cast 0 to (char *) for execl sentinel I generally don't use NULL and use plain 0 instead, based on a "NULL considered harmful" discussion on the classiccmp mailing list many aeons ago (I couldn't find it, and I reason that it must have been 2005 or earlier), but a recent complaint by a packager sent me searching, and I found this: https://ewontfix.com/11/ While I don't give a @#$% about "modern" systems and code-nazi tools, I realized that passing a plain 0 as a pointer sentinel in execl is wrong because it will break on systems where pointers are longer than the plain int type. Again, I don't give a @#$% about the abomination of x86_64 and the like, but if anyone ever manages to port my code to something like a PDP-11 (16-bit int, 32-bit long and pointers), then passing a plain 0 as a function argument where a pointer is expected most definitely won't work: if the most natural stack slot and SP alignment unit is 16 bits, fitting an int, with longs and pointers taking up two such slots, then the call stack will be totally wrong with a plain 0 passed for a pointer. Casting the 0 to (char *) ought to be the most kosher solution for the most retro systems possible.

/*
 * This assembly module provides a replica of OsmocomBB's bogo-millisecond
 * delay_ms() function.  It is literally a copy of what OsmocomBB's delay_ms()
 * compiles to with their gcc version and their optimization settings, as seen
 * by doing arm-elf-objdump on their lib/delay.o.
 *
 * This hack is intended for those cases where we have to copy OsmocomBB's
 * black magic voodoo operations with no ability to understand what is actually
 * needed, such as SPCA552E initialization on the Pirelli DP-L10.
 */

	.text
	.code	32
	.globl	osmo_delay_ms
osmo_delay_ms:
	mov     r3, #0
	sub     sp, sp, #4
	str     r3, [sp]
	ldr     r3, =1300
	mul     r3, r0, r3
	b       2f
1:	ldr     r2, [sp]
	ldr     r2, [sp]
	add     r2, r2, #1
	str     r2, [sp]
2:	ldr     r2, [sp]
	cmp     r2, r3
	bcc     1b
	add     sp, sp, #4
	bx      lr