Re: [Bug-tar] Detecting compressed archives from stdin

[Tim Kientzle]

Michael B. Trausch asked:
> When piping a compressed archive to tar, it does not automatically
> detect and subsequently take appropriate decompression action.
> ... However, if the archive is stored on disk and specified on the
> command-line, it works without this error message.

FYI, both Gunnar Ritter's Heirloom Tar
<http://heirloom.sourceforge.net/tools.html> and my own
bsdtar <http://people.freebsd.org/~kientzle/libarchive/>
handle this as you would expect.  (Last I checked, star
did not, although that may have changed.)

Eric Blake responded:
> ... when piping data in, the data is not seekable ...
> It is technically possible to create a filter thread or
> process, ... create a new output pipe, replay the  first
> few bytes ... then read the remainder ... and play it out
> to the child process...

Libarchive (the archiving library that bsdtar is built
on) handles this transparently without creating an extra
thread or process, although the general principle is still
essentially the same.  (For gzip/bzip2, of course, libarchive
uses zlib/bzlib directly, but Joerg Sonnenberger recently
contributed some code to interact with an external program
without creating an extra filter process.)

I'm not sure how Heirloom Tar implements this, but I
suspect it uses the multi-process approach, exactly
as outlined by Eric Blake.

It would be tricky, I expect, to adapt the code from
libarchive into GNU tar, though you're welcome to try.

Cheers,

Tim Kientzle