- Why this library is written
- Quick start
This is a feature-rich, memory-efficient, and type-safe library to
manipulate Zip archives. The library is the most complete and efficient
implementation of .ZIP specification in Haskell (at least from open-sourced
ones). In particular, it’s created with large multimedia data in mind and
provides all features users might expect, comparable in terms of feature-set
with libraries like
libzip in C.
Why this library is written
There are a few libraries to work with Zip archives, yet every one of them provides only a subset of all the functionality a user may need (obviously the libraries provide functionality that their authors needed) and otherwise is flawed in some way so it cannot be easily used in some situations. Let’s examine all the libraries available on Hackage to understand motivation for this package.
zip-archive is a widely used library. It’s quite old, well-known and
simple to use. However it creates Zip archives purely, as
memory that you can then write to the file system. This is not acceptable if
you work with more-or-less big data. For example, if you have collection of
files with total size of 500 MB and you want to pack them into archive, you
can easily consume up to 1 GB of memory (the files plus resulting archive).
Not always you can afford to do this or do this at scale. Even if you want
just to look at list of archive entries it will read it into memory in all
This is a binding to C library
libzip. There is always a certain
kind of trouble with bindings. For example, you need to take care that
target library is installed and its version is compatible with the version
of your binding. Yes, this means additional headaches. It should be just
“plug and play”, but now you need to watch out for compatibility.
It’s not that bad with libraries that do not break their API for years, but
it’s not the case with
libzip. As the maintainer of
LibZip puts it:
libzip 0.10, 0.11, and 1.0 are not binary compatible. If your C library is 0.11.x, then you should use LibZip 0.11. If your C library is 1.0, then you should use LibZip master branch (not yet released to Hackage).
Now, on my machine I have version 1.0. To put the package on Stackage we had to use version 0.10, because Stackage uses Ubuntu to build packages and libraries on Ubuntu are always ancient. This means that I cannot use the version of the library from Stackage, and I don’t yet know what will be on the server.
After much frustration with all these things I decided to avoid using
LibZip, because after all, this is not that sort of a project that
shouldn’t be done completely in Haskell. By rewriting this in Haskell, I
also can make it safer to use.
This one uses the right approach: leverage a good streaming library
conduit) for memory-efficient processing. This is however is not
feature-rich and has certain problems (including programming style, it uses
error if an entry is missing in archive, among other things), some of them
are reported on its issue tracker. It also does not appear to be maintained
(last sign of activity was on December 23, 2014).
The library supports all features specified in the modern .ZIP specification except for encryption and multi-disk archives. See more about this below.
For reference, here is a copy of the specification.
zip supports the following compression methods:
The best way to add a new compression method to the library is to write a
conduit that will do the compression and publish it as a library.
then depend on it and add it to the list of supported compression methods.
The current list of compression methods reflects what is available on
Hackage at the moment.
Encryption is currently not supported. Encryption system described in the .ZIP specification is known to be seriously flawed, so it’s probably not the best way to protect your data anyway. The encryption method seems to be a proprietary technology of PKWARE (at least that’s what stated about it in the .ZIP specification), so to hell with it.
Sources of file data
The library gives you many options how to get file contents to compress and how to get extracted data. The following methods are supported:
- File name. This is an efficient method to perform compression or decompression. You just specify where to get data or where to save it and the rest is handled by the library.
- Conduit source or sink.
- ByteString. Use it only with small data.
- Copy file from another archive. An efficient operation, file is copied “as is”—no re-compression is performed.
When necessary, the
ZIP64 extension is automatically used. It’s necessary
when anything from this list takes place:
- Total size of archive is greater than 4 GB.
- Size of a single compressed/uncompressed file in archive is greater than 4 GB.
- There are more than 65535 entries in archive.
The library is particularly suited for processing of large files. For example, I’ve been able to easily create 6.5 GB archive with reasonable speed and without significant memory consumption.
The library has API that makes it impossible to create archive with non-portable or invalid file names in it.
As of .ZIP specification 6.3.2, files with Unicode symbols in their names can be put into Zip archives. The library supports mechanisms for this and uses them automatically when needed. Besides UTF-8, CP437 is also supported as it’s required in the specification.
Meta-information about files
The library allows to attach comments to entire archive or individual files, and also gives its user full control over extra fields that are written to file headers, so the user can store arbitrary information about file in the archive.
Codec.Archive.Zip provides everything you may need to
manipulate Zip archives. There are three things that should be clarified
right away, to avoid confusion in the future.
First, we use the
EntrySelector type that can be obtained from relative
FilePaths (paths to directories are not allowed). This method may seem
awkward at first, but it will protect you from the problems with portability
when your archive is unpacked on a different platform.
The second thing, that is rather a consequence of the first, is that there is no way to add directories, or to be precise, empty directories to your archive. This approach is used in Git, and I find it quite sane.
Finally, the third feature of the library is that it does not modify archive
instantly, because doing so on every manipulation would often be
inefficient. Instead we maintain a collection of pending actions that can be
turned into an optimized procedure that efficiently modifies archive in one
pass. Normally this should be of no concern to you, because all actions are
performed automatically when you leave the realm of
ZipArchive monad. If,
however, you ever need to force an update, the
commit function is your
friend. There are even “undo” functions, by the way.
Let’s take a look at some examples that show how to accomplish most typical tasks with help of the library.
To get full information about archive entries, use
λ> withArchive archivePath (M.keys <$> getEntries)
This will return a list of all entries in the archive at
possible to extract contents of an entry as a strict
λ> withArchive archivePath (getEntry entrySelector)
…to stream them to a given sink:
λ> withArchive archivePath (sourceEntry entrySelector mySink)
…to save a specific entry to a file:
λ> withArchive archivePath (saveEntry entrySelector pathToFile)
…and finally just unpack the entire archive into some directory:
λ> withArchive archivePath (unpackInto destDir)
Modifying is also easy, efficient, and powerful. When you want to create a
new archive use
withArchive will do. To add an
λ> createArchive archivePath (addEntry Store "Hello, World!" entrySelector)
You can stream from
Source as well:
λ> createArchive archivePath (sinkEntry Deflate source entrySelector)
To add contents from a file, use
λ> let toSelector = const (mkEntrySelector "my-entry.txt") λ> createArchive archivePath (loadEntry BZip2 toSelector myFilePath)
Finally, you can copy an entry from another archive without re-compression
(unless you use
recompress, see below):
λ> createArchive archivePath (copyEntry srcArchivePath selector selector)
It’s often desirable to just pack a directory:
λ> createArchive archivePath (packDirRecur Deflate mkEntrySelector dir)
It’s also possible to:
- rename an entry with
- delete an entry with
- change compression method with
- change comment associated with an entry with
- delete comment with
- set modification time with
- manipulate extra fields with
- check if entry is intact with
- undo changes with
- force changes to be written to file system with
This should cover all your needs. Feel free to open an issue if you’re missing something.
You can contact the maintainer via the issue tracker.
Pull requests are also welcome.
Copyright © 2016–present Mark Karpov
Distributed under BSD 3 clause license.
- Set user permissions on linux platform as follows: if an existing file is added, use its permissions; if an entry is generated from a bytestring or a stream, use 0600. This behavior mimics the zip utility.
Added support for Zstandard (zstd) compression
Added a Cabal flag
-fdisable-zstdto remove the zstd C library dependency and hence support for Zstd entries.
Dropped support for GHC 8.4.
- Fixed the build on Mac.
The “version made by” info inside archive now correctly sets Unix as the OS that produced the archive when the library is compiled on Unix. This allows other utilities such as
unzipto read and correctly restore file permissions. Issue 62.
- Fix a bug where removing a temporary file failed in the prescence of async exceptions.
- The test suite is now faster.
Dropped support for GHC 8.2 and older.
Added a Cabal flag
-fdisable-bzip2to remove the bzip2 C library dependency and hence support for BZip2 entries.
- Added the
setExternalFileAttrsfunction and the
edExternalFileAttrsfield in the
unpackIntorestore modification time of files.
Stop depending on
Made the module
Remove derived instances of
EntrySelectornot to expose its inner structure.
Change signature of the
loadEntryfunction, its second argument is now just
EntrySelectorof the entry to add.
The second argument of
packDirRecurnow receives paths that are relative to the root of the directory we pack.
MonadBaseControlinstances for the
ZipArchivemonad. Also exported the
ZipStatetype without revealing its data constructor and records.
Improved documentation and metadata.
getEntrySourcepolymorphic in terms of the
Numerous cosmetic corrections to the docs.
- Fixed a bug with modification time serialization on 32 bit systems.
- Fixed a bug that caused
zipto write incorrect number of entries (instead of
0xffff) in central directory when Zip64 feature is enabled.
- Fix literal overflows on 32 bit systems.
Fixed an issue when empty archives with Zip 64 feature enabled could not be read (the “Parsing of archive structure failed: Cannot locate end of central directory”).
- Switched to using
withFile, because the latter does nasty conversions on Windows, see docs for
Added several simple code examples in
Improved speed of detection of invalid archives.
Relaxed dependency on
Added explicit check of “version needed to extract”, so if archive uses some advanced features that we do not support yet, parsing fails.
Value of “version needed to extract” field is now calculated dynamically with respect to actually used features, e.g. if you just store or deflate a not very big file,
2.0version will be written (previously we wrote
4.6unconditionally). This is needed to avoid scaring tools that can only handle basic Zip archives.
- Make decoding of CP437 faster.
- Initial release.