[
https://issues.apache.org/jira/browse/COMPRESS-414?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16069474#comment-16069474
]
Simon Spero commented on COMPRESS-414:
--------------------------------------
I think that this could use a fair chunk of the prototyping in the version 2
branch.
It ought to be possible to evolve an API along the lines discussed above and in
the 2 branch, with the existing api becoming deprecated for removal (less
removal than becoming the implementation classes becoming not exported to OSGI
and Jigsaw users)
Much of the implementation code could slide into new, private packages already;
the current API could mostly be delegates, and the most of the test suite
that use the public API could become kind of integration tests (with the other
tests following the implementation.)
Things would work cleaner with a multiproject build, with projects for the
general apis, with a separate project for each archiver / compressor.
Implementations of formats would be exposed as services (OSGI declarative and
module-info / META-INF/services)
There should also be a combined single dependency thing (a shadowed / shaded
jar, and a Karaf feature.xml?)
The artifacts for each format could be released on separate schedules (e.g. a
bug-fix or performance tweak for 7z could be released without requiring new
releases of everything else (so people just wanting tar don't have to be
affected).
> Provide API for Channel(-ish) Seekable, Possibly Read/Write, Possibly
> Transactional Compressors & Archivers,
> --------------------------------------------------------------------------------------------------------------
>
> Key: COMPRESS-414
> URL: https://issues.apache.org/jira/browse/COMPRESS-414
> Project: Commons Compress
> Issue Type: Improvement
> Reporter: Simon Spero
>
> It would be very useful to have APIs to support random-access, possibly
> read-write archives & compressors.
> These interfaces should probably be based around Seekable, Scatter, and
> Gather ByteChannels.
> It might be necessary or desirable to have these classes extend FileChannel,
> since there are some important methods that first make their appearance as
> abstract methods on that class.
> The methods of interest are the absolute positioned read and write methods,
> plus force/sync.
> The use case I'm thinking of for the the non-relative read/write would be
> for a random-access compressed "file", where it may be useful to keep a cache
> of decompressed blocks, and where a task may require access to several parts
> of the file simultaneously. Nesting this deeper, there might be (well,
> there are) a tar.xz file containing a sequence of versions of a maven
> artifact jar, and with an analysis process that needs access to two or
> more of the files simultaneously.
> A use case requiring read/write + transactional + sync support would be to
> support in-place appending new items to a tar.xz file. It can be quite
> useful to be able to perform such updates "in-place"; for safety this
> requires a log the parts of the existing file that are going to be changed,
> so that the contents can be recovered in the event of failure.
> If there is related metadata in a separate transactional resource, it is good
> to have both stores remain in sync. The complicated parts are the job of a
> transaction manager (e.g. Aries); basic XAResource support is relatively
> simple. There are other bits of glue that are useful for handling simple
> transactions with no transaction manager around, but those can be optional
> imports.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)