[Libguestfs] [PATCH v2] inspector: recognize ppc64 and ppc64le archs (RHBZ#1211996)

[Pino Toscano]

On Friday 17 April 2015 15:15:26 Maros Zatko wrote:
> On 04/17/2015 10:56 AM, Pino Toscano wrote:
> > On Thursday 16 April 2015 16:51:48 Maros Zatko wrote:
> >> ---
> >>   src/filearch.c | 18 +++++++++++++-----
> >>   1 file changed, 13 insertions(+), 5 deletions(-)
> >>
> >> diff --git a/src/filearch.c b/src/filearch.c
> >> index df65c98..5de2959 100644
> >> --- a/src/filearch.c
> >> +++ b/src/filearch.c
> >> @@ -59,8 +59,9 @@ cleanup_magic_t_free (void *ptr)
> >>   # endif
> >>   
> >>   COMPILE_REGEXP (re_file_elf,
> >> -                "ELF.*(?:executable|shared object|relocatable), (.+?),", 0)
> >> -COMPILE_REGEXP (re_elf_ppc64, "64.*PowerPC", 0)
> >> +                "ELF.*(MSB|LSB).*(?:executable|shared object|relocatable), (.+?),", 0)
> >> +COMPILE_REGEXP (re_elf_ppc64, "MSB.*64.*PowerPC", 0)
> >> +COMPILE_REGEXP (re_elf_ppc64le, "LSB.*64.*PowerPC", 0)
> >>   
> >>   /* Convert output from 'file' command on ELF files to the canonical
> >>    * architecture string.  Caller must free the result.
> >> @@ -87,6 +88,8 @@ canonical_elf_arch (guestfs_h *g, const char *elf_arch)
> >>       r = "ia64";
> >>     else if (match (g, elf_arch, re_elf_ppc64))
> >>       r = "ppc64";
> >> +  else if (match (g, elf_arch, re_elf_ppc64le))
> >> +    r = "ppc64le";
> >>     else if (strstr (elf_arch, "PowerPC"))
> >>       r = "ppc";
> >>     else if (strstr (elf_arch, "ARM aarch64"))
> >> @@ -114,6 +117,7 @@ magic_for_file (guestfs_h *g, const char *filename, bool *loading_ok,
> >>   {
> >>     int flags;
> >>     CLEANUP_MAGIC_T_FREE magic_t m = NULL;
> >> +  CLEANUP_FREE char *tmp = NULL;
> >>     const char *line;
> >>     char *elf_arch;
> >>   
> >> @@ -145,7 +149,7 @@ magic_for_file (guestfs_h *g, const char *filename, bool *loading_ok,
> >>     if (loading_ok)
> >>       *loading_ok = true;
> >>   
> >> -  elf_arch = match1 (g, line, re_file_elf);
> >> +  match2 (g, line, re_file_elf, &tmp, &elf_arch);
> >>     if (elf_arch == NULL) {
> > Better switch the checking from elf_arch (which is not touched if
> > match2 fails, so being an uninitialized pointer) to the return value
> > of match2.
> >
> >>       error (g, "no re_file_elf match in '%s'", line);
> >>       return NULL;
> >> @@ -315,6 +319,8 @@ guestfs_impl_file_architecture (guestfs_h *g, const char *path)
> >>   {
> >>     CLEANUP_FREE char *file = NULL;
> >>     CLEANUP_FREE char *elf_arch = NULL;
> >> +  CLEANUP_FREE char *endianness = NULL;
> >> +  CLEANUP_FREE char *end_elf_arch = NULL;
> >>     char *ret = NULL;
> >>   
> >>     /* Get the output of the "file" command.  Note that because this
> >> @@ -324,8 +330,10 @@ guestfs_impl_file_architecture (guestfs_h *g, const char *path)
> >>     if (file == NULL)
> >>       return NULL;
> >>   
> >> -  if ((elf_arch = match1 (g, file, re_file_elf)) != NULL)
> >> -    ret = canonical_elf_arch (g, elf_arch);
> >> +  if ((match2 (g, file, re_file_elf, &endianness, &elf_arch)) != 0) {
> >> +    end_elf_arch = guestfs_int_safe_asprintf (g, "%s %s", endianness, elf_arch);
> >> +    ret = canonical_elf_arch (g, end_elf_arch);
> >> +  }
> > I still do think this approach, i.e. match MSB|LSB, create a new string
> > with it and have it matched again, is doing the same thing twice for no
> > reason.
> >
> > As I suggested in the v1 review, please do pass the result of the
> > MSB|LSB match as new parameter for canonical_elf_arch, checking it
> > when the architecture is ppc64. This will also avoid having a new regex,
> > re_elf_ppc64le, and leave re_elf_ppc64 as it is currently:
> >
> > char *
> > canonical_elf_arch (guestfs_h *g, const char *endianness,
> >                      const char *elf_arch)
> > {
> >    ...
> >      else if (match (g, elf_arch, re_elf_ppc64)) {
> >        if (STREQ (endianness, "LSB")
> >          r = "ppc64le";
> >        else
> >          r = "ppc64";
> >      }
> > }
> >
> > No need for a new regex, and reuses the new MSB|LSB match as it is.
> I don't like this because it creates more structured if/elseif that 
> needed and becomes more cluttered.

Considering the proposed changes to re_file_elf would already give
a MSB|LSB match, what would be the usefulness of forging a new
architecture string, different than either the file/magic output and
where you need to look for MSB|LSB *again*?

Passing the MSB|LSB match would allow also to check for different
architectures with different endianness (e.g. mips/mipsel), without
adding new regexps.

-- 
Pino Toscano