[pups] extract old archive format?

Johnny Billquist bqt at softjar.se
Fri Apr 9 20:15:03 AEST 2010

Jeremy C. Reed wrote:
> Well I found the ar specification (in ar.5 not ar.1).
>              struct ar_hdr {
>                      char      ar_name[14];
>                      long      ar_date;
>                      char      ar_uid;
>                      char      ar_gid;
>                      int       ar_mode;
>                      long      ar_size;
>              };

Simple enough... :-)

> This is same as the old ar.c source.
> (plus more in the manual page.)
> Now my problem is I don't know what "long" or "int" is on the old PDP-11 
> / system 5 this was made on.

An int on the pdp11 is 16 bits, and a long is 32.
Remember? int is whatever size is most convenient for the architecture? :-)

> And I read about PDP-11 "middle endianess" (first time I heard of 
> "middle").

A mess, but we have to live with it.
In short, the bytes of a long on a PDP11 is likely laid out like this:

So, the 16-bit values are each little-endian, but the 16 bit values as 
such, in the 32-bit view, is laid out as big-endian.
Thus middle-endian... :-)

> So I had (wrong but gets ar_name and ar_size correct for my few tests 
> for the first header but chops two characters into the data section).
> struct {
>         char    ar_name[14];
>         int32_t ar_date;
>         char    ar_uid;
>         char    ar_gid;
>         uint16_t        ar_mode;
>         uint16_t        ar_size;
> } ar_buf;
> Well I know above is wrong because ar_size and ar_date should be the 
> same. But I get ar_size correct each time. But it also loses the next 
> two bytes from the data. So I am guessing I have some endian issue where 
> I am getting some things reversed.
> Any ideas?

As others already said, it's your compiler trying to optimize the 
One solution (already presented) is to just play with 16-bit values.
You could also explain to the compiler that it shouldn't try to optimize 
the alignments, but since you have to deal with the middle-endianess 
anyway, you are probably better off just looking at 16-bit values and 
combine them into 32-bit values as needed yourself.


More information about the TUHS mailing list