Removed default '#' from %p

[stefano-zanotti-88]

%p is implementation-defined, so the current implementation is fine.
However, it introduces a default '#'. This is present in other implementations as well, though not all of them.
%p respects width and precision specifiers, as well as flags. [edit: currently, %p explicitly ignores the precision: out_spec->prec_opt = NPF_FMT_SPEC_OPT_NONE]
So, it seems inconsistent that it should not respect the presence/absence of '#'.

Before vs after this change:

printf("%p"     , NULL);     // "0x0"          vs "0"
printf("%#p"    , NULL);     // "0x0"          vs "0"
printf("%10p"   , NULL);     // "       0x0"   vs "         0"
printf("%#10p"  , NULL);     // "       0x0"   vs "         0"
printf("%010p"  , NULL);     // "0x00000000"   vs "0000000000"
printf("%#010p" , NULL);     // "0x00000000"   vs "0000000000"
printf("%.10p"  , NULL);     // "0x0000000000" vs "0000000000"
printf("%#.10p" , NULL);     // "0x0000000000" vs "0000000000"
printf("%0.10p" , NULL);     // "0x0000000000" vs "0000000000"
printf("%#0.10p", NULL);     // "0x0000000000" vs "0000000000"
printf("%p"     , (void*)1); // "0x1"          vs "1"
printf("%#p"    , (void*)1); // "0x1"          vs "0x1"
printf("%10p"   , (void*)1); // "       0x1"   vs "         1"
printf("%#10p"  , (void*)1); // "       0x1"   vs "       0x1"
printf("%010p"  , (void*)1); // "0x00000001"   vs "0000000001"
printf("%#010p" , (void*)1); // "0x00000001"   vs "0x00000001"
printf("%.10p"  , (void*)1); // "0x0000000001" vs "0000000001"
printf("%#.10p" , (void*)1); // "0x0000000001" vs "0x0000000001"
printf("%0.10p" , (void*)1); // "0x0000000001" vs "0000000001"
printf("%#0.10p", (void*)1); // "0x0000000001" vs "0x0000000001"

You could argue that with this change, %p is mostly useless, as it is equivalent to %x with appropriate parameters; however, that's mostly true in any case, and the major benefit of %p is that you don't need a cast to the appropriate integer type (like casting to uintptr_t and using %tx, or casting to (unsigned long) and using %lx, or something else, depending on your platform).

Indeed, it could be easier just to change line 914 from:

need_0x = 'x';

to

if (fs.alt_form) { need_0x = 'x'; }

Note that this, just like the current behavior, always adds "0x", whereas the "%x" equivalent proposed in this change omits it when the number is 0 (a very weird corner case required by the standard).

I'd say it could be more logical, and also result in smaller code, to apply the one-line change instead.
Also, this would allow for an easier removal of %p support, which would reduce the code footprint. I'm sure that not everybody has a need for %p.

Also note: if '#' is made optional here, and you accept #288 (making '#'-support optional for NPF), then further work would be needed to completely remove the 'need_0x' logic when '#'-support is disabled.

[charlesnicholson]

This is very insightful and IMO a nice improvement to the code size- if users really want %p to add the 0x, they can add the # themselves.

I'm happy to merge this but the tests need some updating. I also think your other proposal in #288 is good so I think there's now an ordering issue. What order would you like to proceed in for these PRs?

[stefano-zanotti-88]

I'd go with #288 first.
I'll fix the tests.
Let me know which route you like best: the single-line change, or the actual content of the PR, or something else.

[charlesnicholson]

Implemented as #305 - thanks for the PR!