ant-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From David Rees <d.ree...@usa.net>
Subject [SUBMIT] Cullers (finer grained selection in FileSets)
Date Sat, 10 Mar 2001 00:28:03 GMT
Attached is my first cut at the culler functionality I mentioned last
week. It doesn't include HTML updates yet because I think text
will better facillitate  discussion. Below I describe using cullers
and how to write additional cullers (which is very easy). I think its
ready to go in its current state, but I have a second round on
enhancements that I have planned to be ready in about a week
(these are described below). I also wanted to get this cut out to
receive comments for changes in the next round.

Its about 5 days late due the incredible pain I endured and hoops I 
had to jump through trying to get VisualAge to play nice with the
source and not scramble it. Happily I now have hacks and scripts to
deal with this for future work (now the hoops are lower to the ground
and are no longer flaming ;).

My thoughts for how this "should" be done in Ant2 will be in different
post (later), but the generally idea is the same.

dave


** Usage

The changes extend FileSet to support "FileCullers". A FileCuller is
an abstract type for a class that determines if a file should
selected. Concrete sub-types are based the baseDir and relative path
and do whatever they like to decide about selecting the file. The
current implementation includes a atrribute-based culler and a culler
set that supports grouping and simple logic. However, core to the
design is that new sub-types of FileCuller are easy to add. 

An example says it best:

<copy toDir="\tmp\toDir">
  <fileset dir="\tmp\fromDir">
    <include name="*.txt" />
    <AttributeFileCuller canWrite="no" />
  </fileset>
</copy>

Selects all files that end in ".txt" and are read only.


* AttributeFileCuller

AttributeFileCuller current supports the following attributes. Their
function is pretty clear and they match the corresponding method on
File. The default for all of them is ignore so a blank
AttributeFileCuller
will select all files. Also planned is lastModified and length which
will
have a greater than, less than syntax of some sort.
  canRead	- true/yes, false/no, *ignore
  canWrite	- true/yes, false/no, *ignore
  exists	- true/yes, false/no, *ignore
  fileDir	- file, dir, *ignore


* FileCullerSet

Collection of FileCullers that is also a FileCuller itself
(they can be nested). Supports the following attributes (* is
default):

  logic - and*, or
  Used to indicate what logic is used to combine my nested cullers.

  not - true/yes, false/no*
  If "true/yes" the FileSetCuller's value is reversed.

Cullers are applied in the order they are listed. Short circuiting is
used (e.g. the first false nested culler with
"and" logic will result in the FileCullerSet returning false).

For example:

<copy toDir="\tmp\toDir">
  <fileset dir="\tmp\fromDir">
    <filecullerset>
       <attributefileculler fileDir="file" canWrite="yes" />
       <attributefileculler fileDir="dir" />
    </filecullerset>
  </fileset>
</copy>

Selects all directories and only writable files. 


* Caveats

Unlike patterns, cullers do not effect what directories are scanned,
they only effect what directories are included. This is by design
since the reverse makes them much less useful. Patterns still do
effect what directories are scanned and are applied before a cullers
are applied. One thought is to somehow support the option of
indicating a separate set of scan cullers.

Some subclasses of FileSet/DirectoryScanner may not support certain
FileCullers because they assume files (e.g. ZipFileSet). In this case
the user will receive a build exception that the file (which in this
case isn't really a file) does not exist. The possible extension below
of having the culler ask its scanner for the file will result in the
same limitation, but a nicer message.


* Extending

Additional FileCullers can be added easily by creating a subtype that
implements shouldCull() and then adding the add<ElementName> methods
to FileSet and FileCullerSet.


** Plans

Add lastModified, length to AttributeFileCuller.

Add xor to FileCullerSet.

Create CompareFileCuller which will select files based on comparison
with another directory. Will support culling files that exist, don't
exist or that are different. Can be used with delete/copy for
rsync/robocopy functionality. Possibly will use Mappers?

Create ContainsFileCuller which will select files based on their
contents using a RegExp.

PatternFileCuller - for completeness and to support fine grained
control than available using just patterns.

Make FileCullerSet a DataType so it supports id/refid.

Use CullerSet in FileSet and DirectoryScanner rather than a Vector.

Cleanup FilesetCopyTest to support easy extension for new Cullers.


** Possible Plans

Possibly introduce PathCuller as super-type to recognize that not all
sources are file based.

Possibly pass DirectoryScanner to cullers so it can be used to access
the file. Cullers would then request their file (if they use one) from
the scanner. Allows a variety of different scanners to be used with
cullers. A build exception would be raised for requests that the
scanner does not support (Vaj). Also, lends itself to later supporting
streams for stream-based cullers so that any variety of non-File
sources can be used transparently (FTP, VAJ, Zip).




Mime
View raw message