On 01/20/11 14:38, Matt Olson wrote: > When using compression, it is unknowable in advance at what point in tarring > the source data > "tape-length" will be reached. In other words, it is not possible to predict > what the compression ratio will be in advance.
Fine, but all this suggests is that 'tar' is implementing --use-compress-program incorrectly. When combined with -M, 'tar' should not give the compression program direct access to the output devices. Instead, it should cause the compression program to read from one pipe and write to another, and 'tar' itself should deal with the output devices, switching among them as needed. This may require 'tar' to fork one more time, but that's fine. That would allow 'tar' to work in the above scenario, and it should also work in the encryption scenario.
