Re: [PATCH v1 1/8] lib/string: introduce match_string() helper

[Rasmus Villemoes]

On Mon, Jan 11 2016, Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx> wrote:

> On Sat, 2016-01-09 at 13:57 +0200, Andy Shevchenko wrote:
>> On Sat, Jan 9, 2016 at 3:12 AM, Sergey Senozhatsky
>> <sergey.senozhatsky@xxxxxxxxx> wrote:
>> > Andy Shevchenko wrote:
>> > [..]
>> > > > 
>> > > > strncmp() case seems to be quite common.
>> > > 
>> > > Like I answered to Rasmus, please, provide real examples.
>> > 
>> > [..]
>> > > > int nmatch_string(array, array_size, string, string_len)
>> > > > {
>> > > > ÂÂÂÂÂÂdo {
>> > > > ÂÂÂÂÂÂÂÂÂÂÂÂÂÂstrncmp();
>> > > > ÂÂÂÂÂÂ} while ();
>> > > > }
>> > > > 
>> > > > int match_string(array, array_size, string)
>> > > > {
>> > > > ÂÂÂÂÂÂreturn nmatch_string(array, array_size, string,
>> > > > strlen(string));
>> > > > }
>> > > 
>> > > See above.
>> > 
>> > after some quick and inaccurate grepping, well, probably you're
>> > right - not worth it.
>> 
>> Good grep anyway, it clearly shows that there is hard to generalize
>> which limit to use: a) length of a first argument / item from a list,
>> b) length of a second argument or a constant.
>> 
>> > arch/mips/bcm63xx/boards/board_bcm963xx.cÂÂvoid __init
>> > board_prom_init(void)
>> > net/irda/irnet/irnet_irda.cÂÂÂirnet_dname_to_daddr()
>> > arch/powerpc/sysdev/ppc4xx_cpm.c static ssize_t cpm_idle_store()
>> > arch/x86/ras/mce_amd_inj.c static int __set_inj
>> > drivers/hwtracing/intel_th/msu.c mode_store
>> > drivers/pci/pcie/aer/ecrc.c void pcie_ecrc_get_policy
>> > drivers/pci/pcie/aspm.c pcie_aspm_set_policy
>> > drivers/scsi/aic7xxx/aic7xxx_osm.c aic7xxx_setup
>> > drivers/scsi/aic7xxx/aic79xx_osm.c aic79xx_setup
>> > drivers/scsi/scsi_transport_fc.c static int get_fc_##title##_match
>> > drivers/staging/android/ion/hisilicon/hi6220_ion.c get_type_by_name
>> > drivers/staging/lustre/lustre/lmv/lproc_lmv.c placement_name2policy
>> > drivers/xen/sys-hypervisor.c pmu_mode_store
>
> Thought more about those cases.
>
> If you would like you may introduce something like
>
> int nmatch_string(array, array_size, string, int len)
> {
>   if (len < 0)
>     return match_string();
>
>   for (...) {
>     size_t itemlen = (len > 0) ? len : strlen(array[index]);
> ...
>     if (!strncmp(array[index], string, itemlen))
>       return index;
>   }
>   return -EINVAL;
> }

Yeah, a separate function is probably better. But why not a more
explicit name, match_prefix, match_string_prefix, match_string_starts?

I like the idea of passing the string length if one wants the "is this a
prefix of some array element" semantics, and a sentinel otherwise. But I
don't see any case where one would want match_string() semantics (why
not call match_string directly instead?), so why not let len < 0 mean
"is some array element a prefix of this string" and "len >= 0" be the
other case. I don't see why one shouldn't be able to ask "is the empty
string a prefix of some array element" (that is, are there any elements
in the array); both the array and the string might be run-time things,
so this could occur. And it's not up to a generic library routine like
this to impose restrictions like "the empty string makes no sense, go
away".

Rasmus