Re: [PATCH 3/7] vcs-svn: fix signedness warnings

[David Michael Barr <davidbarr@google.com>]

On Fri, May 25, 2012 at 12:48 AM, Jonathan Nieder <jrnieder@gmail.com> wrote:
> David Barr wrote:
>
>> --- a/vcs-svn/fast_export.c
>> +++ b/vcs-svn/fast_export.c
>> @@ -259,7 +259,7 @@ static int parse_ls_response(const char *response, uint32_t *mode,
>>       }
>>
>>       /* Mode. */
>> -     if (response_end - response < strlen("100644") ||
>> +     if (response_end - response < (off_t) strlen("100644") ||
>
> I wish the static analyzer could notice that "response_end - response"
> is always nonnegative and stop worrying.  If we want to appease it,
> I guess I'd mildly prefer something like
>
>        if (response_end - response < (signed) strlen("100644") ||
>
> which expresses the intent more directly.

Noted.

> [...]
>> --- a/vcs-svn/line_buffer.c
>> +++ b/vcs-svn/line_buffer.c
>> @@ -91,8 +91,7 @@ char *buffer_read_line(struct line_buffer *buf)
>>       return buf->line_buffer;
>>  }
>>
>> -size_t buffer_read_binary(struct line_buffer *buf,
>> -                             struct strbuf *sb, size_t size)
>> +off_t buffer_read_binary(struct line_buffer *buf, struct strbuf *sb, off_t size)
>>  {
>>       return strbuf_fread(sb, size, buf->infile);
>>  }
>
> On systems with larger off_t than size_t (think "typical 32-bit PC,
> since file offsets tend to be 64 bits"), this silently throws away
> bits.  I think the cure is worse than the disease.

Agreed, I'll implement a better approach when I update the series.

> [...]
>> --- a/vcs-svn/sliding_window.c
>> +++ b/vcs-svn/sliding_window.c
>> @@ -43,11 +43,11 @@ static int check_offset_overflow(off_t offset, uintmax_t len)
>>       return 0;
>>  }
>>
>> -int move_window(struct sliding_view *view, off_t off, size_t width)
>> +int move_window(struct sliding_view *view, off_t off, off_t width)
>>  {
>
> Likewise.  I'd rather the caller know that the window has to fit in an
> address space which can be smaller than the maximum file size.
>
> Is this to avoid having two different functions that parse a
> variable-length integer, or is there some other reason?

Nope, just me taking the wrong approach.

I'll submit an alternate patch, which much shorter and less worrying.

--
David Barr