targz_sizes
===========
targz_sizes is a small and simple tool to find out how much space do specific
files take inside a .tar.gz file, after compression. It does that by reading and
decompressing the archive file in memory, then dumping raw information on the
standard output. Archive file is processed in small chunks, so that memory
usage is minimal.
This program has been written to check which directories take the most space in
my `duplicity` backups. This is different than checking space usage on a live
file system because of two factors: compression (text files often take very
little space in archives compared to live file system) and incremental backups
(files that change often are stored multiple times in backups). targz_sizes
performs only basic data gathering, you will need something to provide proper
statistics though.
Usage
-----
No options are implemented. The archive file is taken on standard input,
file sizes are printed on standard output. E.g.:
targz_sizes <my_archive.tar.gz >file_sizes.txt
Output consists of lines, where each line shows size in bytes and name of file,
e.g.:
145 targz_sizes-1.0/
63 targz_sizes-1.0/AUTHORS
3525 targz_sizes-1.0/missing
42103 targz_sizes-1.0/configure
93 targz_sizes-1.0/NEWS
12936 targz_sizes-1.0/config.guess
10025 targz_sizes-1.0/config.sub
Note that directory entries also take up space in the archive! Also note that
technically a file can be encoded by any number of bits, not necessarily being
round number of bytes. In such cases, amount of space shown in the output is
approximate.
Known problems
--------------
To compile you need to pass `--with-zlib` to ./configure. `zlib` is required,
even if the configure script goes well without this switch. E.g.:
./configure --with-zlib
Ustar-formatted file sizes are not supported, the output will be garbage. Ustar
format allows storing files bigger than 8GB in tar archives. GNU tar does not
use this format unless it is necessary, so if you used GNU tar and you know
there are no files bigger than 8GB, you're safe.
Standard input is not reopened in binary mode, as technically there is no way
to perform this operation with standard C primitives. This will be a problem on
platforms that by default provide standard input as a text stream (e.g.
Windows). In such cases, the output will be garbage.
The above problems should be easily fixable. You are free to fork the project
if you want to fix them.
Archive file names are not re-encoded to current locale. You can use `iconv`
for that.
Archive file names are truncated to 100 characters (the limit for original tar
specification). There are format extensions which can cope with longer names,
but those are not supported by targz_sizes. The output will contain the names
as they appear in the 100-byte field in the header, most probably simply
truncated.