|Perl: the Markov chain saw|
Archive::Tarby graff (Chancellor)
|on Jun 01, 2004 at 04:06 UTC||Need Help??|
Item Description: Module for manipulating tar file contents
Review Synopsis: Despite some limitations in design, the module provides functions that are difficult or impossible to do by other means
Requires IO::Zlib in order to handle compressed tar files.
For just about any operating system you'll ever use, you should be able to find some command-line utility or GUI application that knows how to read unix tar files (even those that have been compressed), list the names and sizes of directories and files that they contain, and extract their contents. Many of these utilities also know how to create tar files from a chosen directory tree or list of files. So why does anyone need a Perl module that can do this?
Well what if you're on a system where the available tools don't support the creation of a tar file? In this case, Archive::Tar will allow you to create such a tool yourself, so when some remote colleague says "just send me a tar file with the stuff you're working on", you can do that -- rather than embarrass yourself by replying with something like "I only have pkzip / WinZIP / Stuff-it / Toys-R-Us Archiver / ... Will that format be okay?"
(To be honest, that's not really much of a problem these days -- most archiving file formats have supporting utilities ported to most OS's, and the definitive command-line "tar" utility is available and fully functional for every version of Microsoft OS and MacOSX, as well as being intrinsic to every type of unix system. There must be one for Amiga as well...)
But there is one feature of the definitive "tar" tool that can be a bit limiting at times: whether creating or extracting tar file contents, it is completely, inescapably faithful to its source. In creation, whatever is on disk goes directly into the tar set, sticking to the path and file names provided; on extraction, whatever is in the tar set goes right into directories and files on disk, again, sticking to the names provided. There are ways to (de)select particular paths/files, but that's about all the flexibility you get. Usually, this is exactly what everyone wants, but sometimes you might just wish you could do something a little different when you create or extract from a tar file, like:
There's also a situation not envisioned when "tar" was first conceived decades ago, but common today: you may want to accumulate a set of resources from the web and save them all in one tar file, without ever bothering to write each one to a separate local disk file first -- tar is just a really handy format for this sort of thing.
Unfortunately, as of this writing (Archive::Tar v1.08), the flexibility you get with this module is limited by one major design issue: the entire content of a tar set (all the uncompressed data files contained in the tar file) must reside in memory. Presumably, this approach was chosen so that both compressed and uncompressed tar files would be handled in the same way.
If people only dealt with uncompressed tar files, then the module could be designed to scan a tar image and get the path names, sizes, other metadata, and the byte offsets to each individual data file in the set, so that only the indexing info would be memory resident. But you can't do that very well when the tar file is compressed, and tar files tend to be compressed more often than not. And since there is no inherent upper bound on the size of tar files -- and tar is often used to package really big data sets -- users of Archive::Tar need to be careful not to use this module on the really big ones. (This can be awkward when a compressed tar file contains stuff that "compresses really well", like XML data with overly-verbose tagging -- I've seen compression ratios of 10:1 on such data.)
When you install Archive::Tar, you also get Archive::Tar::File, which is the object used to contain each individual data file in a tar set. When you do:
this creates an array of Archive::Tar::File objects. Called in a list context, "read" returns the list of these objects (and as you would expect, when called in a scalar context, it returns the number of File objects). When you are creating a tar set, each call to the "add_data" or "add_files" method will append to the list of File objects currently in memory. When you call the "write" method, all the File objects currently in the set are written as a tar stream to a file name that you provide (or the stream is returned to the caller as a scalar, if you do not provide an output file name).
There are also three class methods that provide just the basic operations of listing and extracting tar-set contents, and creating a tar set from a list of data file names. These methods avoid the memory load, because they don't bother holding the data files in memory as objects. This also means that you give up a lot of detailed control on individual data files.
The Archive::Tar::File objects provide a lot of handy features, including accessors for file name, mode, linkname, user-name/uid, group-name/gid, size, mtime, cksum, type, etc. You can rename a file or replace its data content, check for empty files using "has_content", and choose between getting a copy of the file content or a reference to the content ("get_content" or "get_content_by_ref").
Getting back to the use of the Archive::Tar object itself, I did come across one potential trap when trying to do a "controlled" extraction of data from an input tar file. Most of the object methods related to reading/extracting are presented in terms of using one or more individual file names as the means for specifying which data file to handle -- in fact, after the "new" and "read" methods, the next three methods described are "contains_file( $filename )", "extract( [@filenames] )" and "list_files()". Most of the remaining methods also need or allow file names as args, which leads one to assume that the "easiest" way to use the object is to do everything in terms of file names.
But the problem is that each time you specify a file name for one of these object methods, the Archive::Tar module needs to search through its list of Archive::Tar::File objects to find the file with that name. This gets painfully slow when you're dealing with a set of many files, and doing something with each of them.
Obviously, there are bound to be situations where you will need to do things by specifying particular data file names in a tar set, but more often, you'll want to work directly with the File objects. A couple of examples will suffice to show the difference:
The problem there is that each call to "tar->get_content( $filename )" invokes a linear search through the set of data files, in order to locate the File object having that name. The following approach goes much faster, because it just iterates through the list of File objects directly:
And of course, given the list of File objects, you have much better control over the selection and handling of files -- here are a few examples:
To conclude: on the whole, if fine-grained control of tar file i/o is something that would be helpful to you, and if you can limit yourself to dealing with tar files that fit in available memory, then you really should be using this module. It's good.
(updated to fix some typos and simple mistakes.)