Josh Triplett wrote:
> On Tue, May 21, 2013 at 08:17:37PM +0100, Ramsay Jones wrote:
>>
>> Signed-off-by: Ramsay Jones <ramsay@xxxxxxxxxxxxxxxxxxx>
>> ---
>
> %Ld for a long long int is actually broken on all architectures; L only
> works for long double. Thanks for fixing that.
Yep, I had forgotten about that change. I should have sent that four
years ago. Sorry about that.
> Formatting nit: I think this reads better when written as:
>
> "... blah blah " PRIx64 " blah blah ..."
>
> , leaving spaces between the macro and the close/open doublequotes.
OK, will do.
>
> Further comments below.
>
>> @@ -336,10 +337,14 @@ const char *show_instruction(struct instruction *insn)
>>
>> switch (expr->type) {
>> case EXPR_VALUE:
>> - buf += sprintf(buf, "%lld", expr->value);
>> + buf += sprintf(buf, "%"PRId64, expr->value);
>> break;
>> case EXPR_FVALUE:
>> +#if !defined(__MINGW32__)
>> buf += sprintf(buf, "%Lf", expr->fvalue);
>> +#else
>> + buf += sprintf(buf, "%f", (double)expr->fvalue);
>> +#endif
>
> This seems really sad; does MinGW really have long double but no way to
> print it? Can we at least emit something here to indicate possible
> truncation or loss of precision, if no means exists to print a long
> double?
I couldn't find any means to do so, just from experimentation (I didn't
have any MinGW specific gcc documentation). Thus, I tried:
$ cat -n junk.c
1 #include <stdio.h>
2
3 int main(int argc, char *argv[])
4 {
5 long double f = 128.821L;
6
7 printf("sizeof(float) = %d\n", sizeof(float));
8 printf("sizeof(double) = %d\n", sizeof(double));
9 printf("sizeof(long double) = %d\n", sizeof(long double));
10 printf("f = %Lf\n", f);
11 return 0;
12 }
$ gcc -Wall -o junk junk.c
junk.c: In function 'main':
junk.c:10: warning: unknown conversion type character 'L' in format
junk.c:10: warning: too many arguments for format
$ ./junk.exe
sizeof(float) = 4
sizeof(double) = 8
sizeof(long double) = 12
f = -0.000000
$ cl -W4 junk.c
Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 15.00.30729.01 for 80x86
Copyright (C) Microsoft Corporation. All rights reserved.
junk.c
junk.c(3) : warning C4100: 'argv' : unreferenced formal parameter
junk.c(3) : warning C4100: 'argc' : unreferenced formal parameter
Microsoft (R) Incremental Linker Version 9.00.30729.01
Copyright (C) Microsoft Corporation. All rights reserved.
/out:junk.exe
junk.obj
$ ./junk.exe
sizeof(float) = 4
sizeof(double) = 8
sizeof(long double) = 8
f = 128.821000
$
So, msvc knows about the L size modifier, but it treats a 'long double'
the same as a 'double'.
$ vim junk.c # change format specifier %Lf => %lf
$ gcc -Wall -o junk junk.c
junk.c: In function 'main':
junk.c:10: warning: format '%lf' expects type 'double', but argument 2 has type
'long double'
$ ./junk.exe
sizeof(float) = 4
sizeof(double) = 8
sizeof(long double) = 12
f = -0.000000
$
I'm sure you can guess what happens if you try "%f" instead.
The current "compat" layer has an string_to_ld() function, so maybe
there should be an ld_to_string()? dunno.
[A 64-bit MinGW system probably doesn't have this problem, of course ...]
>
>> @@ -463,7 +468,7 @@ const char *show_instruction(struct instruction *insn)
>> }
>>
>> if (buf >= buffer + sizeof(buffer))
>> - die("instruction buffer overflowed %td\n", buf - buffer);
>> + die("instruction buffer overflowed %d\n", (int)(buf - buffer));
>
> No, ptrdiff_t does not portably fit in int; it generally has the same
> size as size_t (64-bit on 64-bit platforms). Cast to "long long" and
> use PRId64 if you must.
Yes, for the same reason, git does:
printf("...%" PRIuMAX "...", (uintmax_t)(buf - buffer));
so I'll do that in the re-roll.
>
>> --- a/pre-process.c
>> +++ b/pre-process.c
>> @@ -158,12 +158,17 @@ static int expand_one_symbol(struct token **list)
>> } else if (token->ident == &__DATE___ident) {
>> if (!t)
>> time(&t);
>> +#if !defined(__MINGW32__)
>> strftime(buffer, 12, "%b %e %Y", localtime(&t));
>> +#else
>> + strftime(buffer, 12, "%b %d %Y", localtime(&t));
>> + if (buffer[4] == '0') buffer[4] = ' ';
>> +#endif
> To the best of my knowledge, nothing guarantees the length of %b, so the
> [4] here seems wrong.
Yes, this was just a quick hack to compensate for the lack of the
"%e" format specifier in the msvc strftime(). (elsewhere the lack
of "%T" was easier to replace). I will have to think about this.
Any ideas?
>> @@ -980,7 +981,11 @@ static int show_fvalue(struct expression *expr)
>> int new = new_pseudo();
>> long double value = expr->fvalue;
>>
>> +#if !defined(__MINGW32__)
>> printf("\tmovf.%d\t\tv%d,$%Lf\n", expr->ctype->bit_size, new, value);
>> +#else
>> + printf("\tmovf.%d\t\tv%d,$%f\n", expr->ctype->bit_size, new, (double)value);
>> +#endif
>
> Same comment as above regarding long double.
>
>> --- a/tokenize.c
>> +++ b/tokenize.c
>> @@ -547,8 +547,8 @@ static int get_one_number(int c, int next, stream_t *stream)
>> }
>>
>> if (p == buffer_end) {
>> - sparse_error(stream_pos(stream), "number token exceeds %td characters",
>> - buffer_end - buffer);
>> + sparse_error(stream_pos(stream), "number token exceeds %d characters",
>> + (int)(buffer_end - buffer));
>
> Same comment as above regarding ptrdiff_t.
>
> - Josh Triplett
Thanks Josh. You hit every part of the patch that I wanted to
tidy up! :-D
ATB,
Ramsay Jones
--
To unsubscribe from this list: send the line "unsubscribe linux-sparse" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
