[U-Boot] [PATCH 05/15] lib: vsprintf: correct printing of Unicode strings

[Alexander Graf]

On 26.08.18 20:34, Heinrich Schuchardt wrote:
> On 08/26/2018 08:05 PM, Alexander Graf wrote:
>>
>>
>> On 11.08.18 17:28, Heinrich Schuchardt wrote:
>>> The width and precision of the printf() function refer to the number of
>>> characters not to the number of bytes printed.
>>>
>>> Signed-off-by: Heinrich Schuchardt <xypron.glpk at gmx.de>
>>> ---
>>>  lib/vsprintf.c | 22 +++++++++++++---------
>>>  1 file changed, 13 insertions(+), 9 deletions(-)
>>>
>>> diff --git a/lib/vsprintf.c b/lib/vsprintf.c
>>> index a07128ad96..b7eb9d5f5e 100644
>>> --- a/lib/vsprintf.c
>>> +++ b/lib/vsprintf.c
>>> @@ -280,18 +280,22 @@ static char *string16(char *buf, char *end, u16 *s, int field_width,
>>>  		int precision, int flags)
>>>  {
>>>  	u16 *str = s ? s : L"<NULL>";
>>> -	int utf16_len = u16_strnlen(str, precision);
>>> -	u8 utf8[utf16_len * MAX_UTF8_PER_UTF16];
>>> -	int utf8_len, i;
>>> -
>>> -	utf8_len = utf16_to_utf8(utf8, str, utf16_len) - utf8;
>>> +	ssize_t i, len = utf16_strnlen(str, precision);
>>>  
>>>  	if (!(flags & LEFT))
>>> -		while (utf8_len < field_width--)
>>> +		for (; len < field_width; --field_width)
>>>  			ADDCH(buf, ' ');
>>> -	for (i = 0; i < utf8_len; ++i)
>>> -		ADDCH(buf, utf8[i]);
>>> -	while (utf8_len < field_width--)
>>> +	for (i = 0; i < len; ++i) {
>>> +		s32 code = utf16_get((const u16 **)&str);
>>> +
>>> +		if (code < 0) {
>>> +			code = '?';
>>> +			if (*str)
>>> +				++str;
>>> +		}
>>> +		utf8_put(code, &buf);
>>
>> Can you introduce or reuse a strcpy() helper in charset.c for this? That
>> way the compiler has the chance to inline utf16_get() and utf8_put() and
>> make the function fast.
> 
> strcpy() works on bytes not on multi-byte utf-8 characters. So it is
> unclear to me how I should make use of strcpy() here.

What I was trying to imply is that what you're doing here is very
similar to utf8_utf16_strncpy(). Maybe we can reuse the same function or
at least something very similar.

> Of cause we could define utf8_put() and utf8_get() as inline function.
> But that would increase code size. Is this what you would prefer? I
> would guess that the serial interface is always the slowest part of text
> output anyway.

Real serial output is definitely orders of magnitude slower, I agree.
But if we can make the code easier to read along the way I'm all for it ;).

I think what it boils down to is that I'd prefer if we keep
utf{8,16}_{get,put}() as local to charset.c as we can and instead put
slightly higher level wrappers around them, like you did for pretty much
everything else.


Alex