re: Installation from a tape

[matthew green <mrg%eterna23.net@localhost>]

it occured to me that gzip could handle this better so now i think
this patch would make this problem go away.

can someone try?  it seems to work for me - depending on what fstat(2)
tells it as st_blksize, it uses that for IO for that first chunk:

    https://www.netbsd.org/~mrg/gzip.blksize.diff

thanks.  i didn't test beyond gzip output, and it could be cleaned up
to have the backends free the input, or at least have them to re-use
it for their IO, though it might be nicer to also pass both the
buffer size and read size (ie, preferred IO size, and, amount that is
currently present in the buffer.)

if this works, i'll commit either as-is, or advance as above..

thanks.


.mrg.