Re: [Arpix-devel] New version of tarlib

[Giovanni C. <val...@li...>]

Simo Sorce wrote:
> 
> >   /* support functions */
> > int check_file (char* file)
> > {
> >   union block block;
> >   char* magic;
> >   int retval = 0;
> >   FILE* fp = fopen(file, "rb");
> 
> I think you should use standard file descriptors (open, write, read,
> seek) and not file streams as it is more portable across systems.

I prefer open() calls too, but I found that POSIX-style calls have
nothing like ftell() which I think may be useful, esp. in extracting or
deleting files.
Should my guess happen to be wrong, I'd happily switch to open() calls.


> 
> >
> >   if (!fp)
> >     return -1;
> >
> >   if (fread(&block, sizeof(union block), 1, fp) != 1)
> >   {
> >     fclose(fp);
> >     return -1;
> >   }
> >
> >   magic = block.header.magic;
> >
> >   if (memcmp(magic, TMAGIC, sizeof(char)*TMAGLEN))
> >     retval = 1; /* posix archive */
> >   if (memcmp(magic, OLDGNU_MAGIC, sizeof(char)* 8))
> >     retval = 2; /* GNU archive */
> 
> Using sizeof(char) in this contest is a bug, file format declares magic
> beeing long 8 bytes!
> This is indipendent from how many bytes a char use on the platform.
> On platforms with chars of greater than 8bit size this would break as
> they will copy more bytes causing possible memory leaks.

I could have used strcmp() as well, but AFAIK memcmp() is more efficient
than strcmp(), even when comparing strings. So I prefered to force a
string comparison by use of memcmp(), but I had to be sure that it would
have compared 8 chars, not 8 bytes.
Do you think I should use strcmp(), instead of this hack?