| 1 |
On 2022-02-22, Grant Edwards wrote: |
| 2 |
> That doesn't work. It shows the size of the drive as the |
| 3 |
> "uncompressed" size and 0 as compressed: |
| 4 |
> |
| 5 |
> # gzip -clt </dev/sdd |
| 6 |
> compressed uncompressed ratio uncompressed_name |
| 7 |
> 31658606592 0 0.0% stdout |
| 8 |
> |
| 9 |
> The actual size of the compressed data is about 1/3 the value shown |
| 10 |
> above. |
| 11 |
> |
| 12 |
> It's not reading through the stream. It's seeking to the end and |
| 13 |
> looking at what it thinks is the trailer info. I thought that maybe |
| 14 |
> using a pipe instead of a file would make it read through the data, |
| 15 |
> but that doesn't work either: |
| 16 |
> |
| 17 |
> $ ls > foo |
| 18 |
> $ ls -l foo |
| 19 |
> -rw-r--r-- 1 grante users 12923 Feb 22 07:51 foo |
| 20 |
> |
| 21 |
> $ gzip foo |
| 22 |
> $ ls -l foo.gz |
| 23 |
> -rw-r--r-- 1 grante users 6083 Feb 22 07:51 foo.gz |
| 24 |
> |
| 25 |
> $ gzip -clt <foo.gz |
| 26 |
> compressed uncompressed ratio uncompressed_name |
| 27 |
> 6083 12923 53.1% stdout |
| 28 |
> |
| 29 |
> $ echo asdf >> foo.gz |
| 30 |
> |
| 31 |
> $ gzip -clt <foo.gz |
| 32 |
> compressed uncompressed ratio uncompressed_name |
| 33 |
> 6088 174482547 100.0% stdout |
| 34 |
> |
| 35 |
> $ cat foo.gz | gzip -clt |
| 36 |
> compressed uncompressed ratio uncompressed_name |
| 37 |
> -1 -1 0.0% stdout |
| 38 |
> |
| 39 |
> |
| 40 |
> |
| 41 |
> Here's relevent portion of the strace for the 'gzip -clt <foo.gz' |
| 42 |
> where it seeks to end-8 and reads what it thinks is the uncompressed |
| 43 |
> length and the CRC: |
| 44 |
> |
| 45 |
> lseek(0, -8, SEEK_END) = 6080 |
| 46 |
> read(0, "2\0\0asdf\n", 8) = 8 |
| 47 |
> write(1, " 6088 17"..., 54) = 54 |
| 48 |
> close(0) = 0 |
| 49 |
> close(1) = 0 |
| 50 |
> exit_group(0) = ? |
| 51 |
|
| 52 |
Hi Grant, |
| 53 |
|
| 54 |
you're right it doesn't work with the trailing garbage. I wasn't aware |
| 55 |
it actually seeks even on pipes. |
| 56 |
|
| 57 |
By coincidence it seems the next release will even change this behavior: |
| 58 |
|
| 59 |
https://git.savannah.gnu.org/cgit/gzip.git/commit/?id=cf26200380585019e927fe3cf5c0ecb7c8b3ef14 |
| 60 |
|
| 61 |
But this actually still doesn't solve your problem, since this only |
| 62 |
adjust the calculation of the uncompressed size, but the compressed size |
| 63 |
is still derived from stat. |
Archives