Re: [PATCH v9 3/3] Add tests for strfrom functions

[Joseph Myers <joseph@codesourcery.com>]

On Tue, 18 Oct 2016, Gabriel F. T. Gomes wrote:

> +/* Hexadecimal tests.  */
> +static const struct htests htest[] = {
> +  {"%a", { "0x1,ffp+6", "0x3,fep+5", "0x7,fcp+4", "0xf,f8p+3" },
> +   {0x1.ffp+6, 0x1.ffp+6, 0x1.ffp+6}},
> +  {"%a", { "0x1,88p+4", "0x3,1p+3", "0x6,2p+2", "0xc,4p+1" },
> +   {0x1.88p+4, 0x1.88p+4, 0x1.88p+4}},
> +  {"%A", { "-0X1,88P+5", "-0X3,1P+4", "-0X6,2P+3", "-0XC,4P+2" },
> +   {-0x1.88p+5, -0x1.88p+5, -0x1.88p+5}},
> +  {"%a", { "0x1,44p+10", "0x2,88p+9", "0x5,1p+8", "0xa,2p+7"},
> +   {0x1.44p+10, 0x1.44p+10, 0x1.44p+10}},
> +  {"%a", { "0x1p-10", "0x2p-11", "0x4p-12", "0x8p-13"},
> +   {0x0.0040p+0, 0x0.0040p+0, 0x0.0040p+0}},
> +  {"%a", { "0x1,4p+3", "0x2,8p+2", "0x5p+1", "0xap+0"},
> +   {10.0, 10.0, 10.0}}
> +};

These still aren't type-generic.

> +/* Hexadecimal tests.  */
> +static const struct htests htest[] = {
> +  {"%a", { "0x1.ffp+6", "0x3.fep+5", "0x7.fcp+4", "0xf.f8p+3" },
> +   {0x1.ffp+6, 0x1.ffp+6, 0x1.ffp+6}},
> +  {"%a", { "0x1.88p+4", "0x3.1p+3", "0x6.2p+2", "0xc.4p+1" },
> +   {0x1.88p+4, 0x1.88p+4, 0x1.88p+4}},
> +  {"%A", { "-0X1.88P+5", "-0X3.1P+4", "-0X6.2P+3", "-0XC.4P+2" },
> +   {-0x1.88p+5, -0x1.88p+5, -0x1.88p+5}},
> +  {"%a", { "0x1.44p+10", "0x2.88p+9", "0x5.1p+8", "0xa.2p+7"},
> +   {0x1.44p+10, 0x1.44p+10, 0x1.44p+10}},
> +  {"%a", { "0x1p-10", "0x2p-11", "0x4p-12", "0x8p-13"},
> +   {0x0.0040p+0, 0x0.0040p+0, 0x0.0040p+0}},
> +  {"%a", { "0x1.4p+3", "0x2.8p+2", "0x5p+1", "0xap+0"},
> +   {10.0, 10.0, 10.0}}

Likewise.

> +#define TEST(s, fmt, size, rc, f, d, ld)	\
> +  {						\
> +    s, fmt, size, rc,  { ENTRY (f, d, ld) }	\
> +  }

You should have one argument to TEST that is the constant used for all 
types (and that then gets the appropriate suffix added to it for each 
type).  It should not be necessary to update the tables of tests at all 
when adding a new floating-point type (in general, the idea of making such 
tests type-generic is so that as little as possible needs to change to 
make them test a new type; ideally only the definitions of 
GEN_TEST_STRTOD_FOREACH and STRTOD_TEST_FOREACH).  The only obvious 
differences I see between the constants in your tests for different types 
is that some have more precision in the constants for double and long 
double, but it's fine to have the extra digits in the float constant even 
if they aren't significant to its value.

That is, the definition should look something like

{ s, fmt, size, rc, { GEN_TEST_STRTOD_FOREACH (SOMETHING, f) } }

where SOMETHING expands to concatenate its 7th argument (the 
floating-point number) with its 5th (the constant suffix) (followed by a 
comma).  Then I think you can get rid of the ENTRY and CHOOSE macros.

(tst-strtod-round is different because it involves data for every 
floating-point *format*, which then needs mapping to the supported types.)

-- 
Joseph S. Myers
joseph@codesourcery.com