ant-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Antoine Levy-Lambert" <>
Subject Re: cvs commit: ant/src/main/org/apache/tools/ant/types/selectors
Date Tue, 15 Jul 2003 15:06:35 GMT
On Jul 15 2003, wrote :

>I'm not sure I understand what you've done, I'll try to come up with
>some unit tests if I think something is wrong 8-)

Please do the unit tests. I also added 3 unit tests to DirectoryScannerTest
before checking in my new code.

What I did is the following :
   - create a partition of patterns without wildcards from the include
patterns of the fileset,
   - only start scanning from the patterns in this partition which are not
   - mark the patterns which are files and are included directly as selected
   - mark the patterns which are directories and are included as selected

> * You go a long way to avoid redundantly adding patterns in
> DirectoryScanner#checkIncludePatterns (the parent check).  If the
> patterns come in reversed, you'll still have duplicate roots, no?  Say
> a/b/c is added before a/b is.  You may avoid this by sorting the
> include patterns first.

True, I will change this tonight

> * If all patterns are plain files, i.e. no "pattern" contains any
> wildcards, things could still be improved by not performing any
> pattern matching at all.
In the current solution, if all patterns are plain files, then there is no
scanning, simply the patterns get added to the array includedFiles.

> * SelectorUtils#rtrimWildcardTokens should probably short circuit and
> return the original input if the input doesn't contain any wildcards
> at all.
Yes, but I am not too concerned about this.
I did a test of scanning speed on filesets containing 2 or 3 patterns nested
5 or 6 directories deep, and the new solution is much faster.
But scanning time is just unfortunately a small fraction of the time needed
to make a jar or a zip file.


To unsubscribe, e-mail:
For additional commands, e-mail:

View raw message