Re: [PATCH] wip: change ISSPACE to use ascii whitespace check.

Alejandro Colomar via Mutt-dev <[email protected]> Thu, 19 Feb 2026 12:51:57 +0100
Newsgroups	gmane.mail.mutt.devel
Message-ID	<aZbvktbVG_lJndcz@devuan>
Hi Ian,

On 2026-02-19T10:30:09+0000, Ian Collier via Mutt-dev wrote:
> On Thu, Feb 19, 2026 at 01:46:34AM +0100, Alejandro Colomar via Mutt-dev wrote:
> > On 2026-02-18T23:42:28+0000, Ian Collier via Mutt-dev wrote:
> > > I do not understand why strchrnul and strcmp are even being invoked here.
> 
> > Because they are necessary.  (Well, strcmp(3) is not, as you could do
> > manual byte operations, but it's simple, more readable, and optimized
> > out.)  strchrnul(3) is necessary.
> 
> > 	strchr(s, '\0') != NULL
> 
> > would evaluate to true, which is a misbehavior: it would treat '\0' as
> > white space, while it is not. 
> 
> The nul character is a bit of an edge case when considering strxxx functions,
> of course.  But that's easily fixed:
> 
>  c && strchr(s, c) != NULL
> 
> No doubt you'll now be telling me that you don't want to evaluate c twice
> in the context of a macro.  Well then, use an inline function. :-)

An inline function with that would be an alternative.  A macro is more
compact, though, and can be implemented as a one-liner.  When
implementing several of these, it can help readability of the entire set
of APIs if they fit in a compact space and they are very similar:

	// isascii_c - is [:ascii:] C-locale
	#define isascii_c(c)   (!!strchr(CTYPE_ASCII_C, c))
	#define iscntrl_c(c)   (!!strchr(CTYPE_CNTRL_C, c))
	#define islower_c(c)   (!streq(strchrnul(CTYPE_LOWER_C, c), ""))
	#define isupper_c(c)   (!streq(strchrnul(CTYPE_UPPER_C, c), ""))
	#define isdigit_c(c)   (!streq(strchrnul(CTYPE_DIGIT_C, c), ""))
	#define ispunct_c(c)   (!streq(strchrnul(CTYPE_PUNCT_C, c), ""))
	#define isspace_c(c)   (!streq(strchrnul(CTYPE_SPACE_C, c), ""))
	#define isalpha_c(c)   (!streq(strchrnul(CTYPE_ALPHA_C, c), ""))
	#define isalnum_c(c)   (!streq(strchrnul(CTYPE_ALNUM_C, c), ""))
	#define isgraph_c(c)   (!streq(strchrnul(CTYPE_GRAPH_C, c), ""))
	#define isprint_c(c)   (!streq(strchrnul(CTYPE_PRINT_C, c), ""))
	#define isxdigit_c(c)  (!streq(strchrnul(CTYPE_XDIGIT_C, c), ""))

Here's a comparison of the assembly they produce:

	alx@devuan:~/tmp$ diff -u isspace?.c
	--- isspace1.c	2026-02-19 12:43:01.016632233 +0100
	+++ isspace2.c	2026-02-19 12:39:40.987744755 +0100
	@@ -1,8 +1,7 @@
	 #include <string.h>
	 #define CTYPE_SPACE_C   " \t\n\v\f\r"
	-#define streq(a,b)  (strcmp(a,b) == 0)
	 bool
	 isspace_c(int c)
	 {
	-	return !streq(strchrnul(CTYPE_SPACE_C, c), "");
	+	return c && strchr(CTYPE_SPACE_C, c) != NULL;
	 }
	alx@devuan:~/tmp$ gcc -S -O2 isspace?.c
	alx@devuan:~/tmp$ diff -u isspace?.s
	--- isspace1.s	2026-02-19 12:44:24.264776257 +0100
	+++ isspace2.s	2026-02-19 12:44:24.276776354 +0100
	@@ -1,4 +1,4 @@
	-	.file	"isspace1.c"
	+	.file	"isspace2.c"
		.text
		.section	.rodata.str1.1,"aMS",@progbits,1
	 .LC0:
	@@ -10,16 +10,23 @@
	 isspace_c:
	 .LFB0:
		.cfi_startproc
	+	testl	%edi, %edi
	+	je	.L5
		subq	$8, %rsp
		.cfi_def_cfa_offset 16
		movl	%edi, %esi
		leaq	.LC0(%rip), %rdi
	-	call	strchrnul@PLT
	-	cmpb	$0, (%rax)
	+	call	strchr@PLT
	+	testq	%rax, %rax
		setne	%al
		addq	$8, %rsp
		.cfi_def_cfa_offset 8
		ret
	+	.p2align 4,,10
	+	.p2align 3
	+.L5:
	+	xorl	%eax, %eax
	+	ret
		.cfi_endproc
	 .LFE0:
		.size	isspace_c, .-isspace_c

The strchrnul(3) version seems to have significantly less instructions.

And strchrnul(3) is internally simpler than strchr(3).  In musl, for
example, strchrnul(3) is used to implement strchr(3).


Have a lovely day!
Alex

-- 
<https://www.alejandro-colomar.es>
signature.asc (application/pgp-signature, 833 B)