Re: [PATCH 2/2] commands: test: add based support for bash-test style

[Marco Felsch <m.felsch@pengutronix.de>]

Hi Ahmad,

On 23-07-04, Ahmad Fatoum wrote:
> On 04.07.23 00:58, Marco Felsch wrote:
> > Add [[ expression ]] support which allow pattern matching like:
> > 
> >   - [[ "foo1" == "foo*" ]]
> >   - [[ "foo" != "bar" ]]
> > 
> > Signed-off-by: Marco Felsch <m.felsch@pengutronix.de>
> > ---
> >  commands/Kconfig | 13 +++++++++++++
> >  commands/test.c  | 39 ++++++++++++++++++++++++++++++++++-----
> >  2 files changed, 47 insertions(+), 5 deletions(-)
> > 
> > diff --git a/commands/Kconfig b/commands/Kconfig
> > index 4d3ff631a8..615e96aa9d 100644
> > --- a/commands/Kconfig
> > +++ b/commands/Kconfig
> > @@ -1219,6 +1219,19 @@ config CMD_TEST
> >  		  !, =, !=, -eq, -ne, -ge, -gt, -le, -lt, -o, -a, -z, -n, -d, -e,
> >  		  -f, -L; see 'man test' on your PC for more information.
> >  
> > +if CMD_TEST
> > +
> > +config CMD_TEST_BASH_COMP
> > +	tristate
> > +	select CONFIG_FNMATCH
> 
> select FNMATCH, no CONFIG_

Right.

> Does this really need its own symbol? Can't you just use CONFIG_GLOB
> and amend the help text?

The implementation was inspired by busybox and they use fnmatch.
Checking glob.c I don't see any difference, therefore I would keep the
fnmatch.

I wasn't sure if every user need the bash(ism) support and sometimes
bootloader partitions are really small due to legacy reasons. Therefore
I went this way to let the user the choice to enable/disable the
support.

> > +	prompt "test bash-compatibility"
> > +	help
> > +	  Enable minimal support for bash compatible tests: [[ ]] to allow
> > +	  pattern matching via: == and !=
> > +
> > +# end CMD_TEST
> > +endif
> > +
> >  config CMD_TRUE
> >  	tristate
> >  	default y
> > diff --git a/commands/test.c b/commands/test.c
> > index c1b84c42ef..77e15aeb03 100644
> > --- a/commands/test.c
> > +++ b/commands/test.c
> > @@ -11,9 +11,13 @@
> >  #include <command.h>
> >  #include <fs.h>
> >  #include <linux/stat.h>
> > +#ifdef CONFIG_CMD_TEST_BASH_COMP
> > +#include <fnmatch.h>
> > +#endif
> >  
> >  typedef enum {
> >  	OPT_EQUAL,
> > +	OPT_EQUAL_BASH,
> >  	OPT_NOT_EQUAL,
> >  	OPT_ARITH_EQUAL,
> >  	OPT_ARITH_NOT_EQUAL,
> > @@ -36,6 +40,7 @@ typedef enum {
> >  
> >  static char *test_options[] = {
> >  	[OPT_EQUAL]			= "=",
> > +	[OPT_EQUAL_BASH]		= "==",
> >  	[OPT_NOT_EQUAL]			= "!=",
> >  	[OPT_ARITH_EQUAL]		= "-eq",
> >  	[OPT_ARITH_NOT_EQUAL]		= "-ne",
> > @@ -67,18 +72,37 @@ static int parse_opt(const char *opt)
> >  	return -1;
> >  }
> >  
> > +static int string_comp(const char *left_op, const char *right_op, bool bash_test)
> > +{
> > +#ifdef CONFIG_CMD_TEST_BASH_COMP
> > +	if (bash_test)
> > +		return fnmatch(right_op, left_op, 0);
> > +#endif
> 
> Scripts that execute differently depending on a barebox config option
> is not good user experience. I think it would be better if [[ is an
> error if support is missing.

I was thinking about a warning which is printed once you use [[ and
didn't enabled it before. Then I read man bash again and [[ supports
'=' as well which equals the [ function. Therefore I dropped the
warning.

> Also can't the #ifdef be replaced with a IS_ENABLED()

I don't think so, since the fnmatch is added conditional to keep the
code base the same if not enabled.

> > +	return strcmp(left_op, right_op);
> > +}
> > +
> >  static int do_test(int argc, char *argv[])
> >  {
> >  	char **ap;
> >  	int left, adv, expr, last_expr, neg, last_cmp, opt, zero;
> >  	ulong a, b;
> >  	struct stat statbuf;
> > +	bool bash_test = false;
> >  
> >  	if (*argv[0] == '[') {
> >  		argc--;
> > -		if (*argv[argc] != ']') {
> > -			printf("[: missing `]'\n");
> > -			return 1;
> > +		if (!strncmp(argv[0], "[[", 2)) {
> 
> This is equivalent to argv[0][1] == '['

I would keep the strncmp.

Regards,
  Marco

> > +			if (strncmp(argv[argc], "]]", 2) != 0) {
> > +				printf("[[: missing `]]'\n");
> > +				return 1;
> > +			}
> > +			if (IS_ENABLED(CONFIG_CMD_TEST_BASH_COMP))
> > +				bash_test = true;
> > +		} else {
> > +			if (*argv[argc] != ']') {
> > +				printf("[: missing `]'\n");
> > +				return 1;
> > +			}
> >  		}
> >  	}
> >  
> > @@ -183,10 +207,11 @@ static int do_test(int argc, char *argv[])
> >  			b = simple_strtol(ap[2], NULL, 0);
> >  			switch (parse_opt(ap[1])) {
> >  			case OPT_EQUAL:
> > -				expr = strcmp(ap[0], ap[2]) == 0;
> > +			case OPT_EQUAL_BASH:
> > +				expr = string_comp(ap[0], ap[2], bash_test) == 0;
> >  				break;
> >  			case OPT_NOT_EQUAL:
> > -				expr = strcmp(ap[0], ap[2]) != 0;
> > +				expr = string_comp(ap[0], ap[2], bash_test) != 0;
> >  				break;
> >  			case OPT_ARITH_EQUAL:
> >  				expr = a == b;
> > @@ -233,7 +258,11 @@ static int do_test(int argc, char *argv[])
> >  	return expr;
> >  }
> >  
> > +#ifdef CONFIG_CMD_TEST_BASH_COMP
> > +static const char * const test_aliases[] = { "[", "[[", NULL};
> > +#else
> >  static const char * const test_aliases[] = { "[", NULL};
> > +#endif
> >  
> >  BAREBOX_CMD_HELP_START(test)
> >  BAREBOX_CMD_HELP_TEXT("Options:")
> 
> -- 
> Pengutronix e.K.                           |                             |
> Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
> 31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
> Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |
> 
>