There was a (in)famous incident where the implementation for rerunning
the Semantic Index Builder did the same thing - it (formerly) erased all
the files in the target directory.
What happened was a senior level manager (Dave Ferrucci) was just trying
out this code, he was busy and didn't read all the fine print, and when
it asked him where to put the index, he said - well - hmm - just put it
on my desk top. Sure enough, the code erased his entire desktop... He
was not happy about this, to say the least.
So - if you want to do this, please be sure this kind of user scenario
won't occur by accident.
-Marshall
Michael Baessler wrote:
Hi,
when installing a PEAR file, currently the InstallationController API
overrides all available data in the target installation directory. The
directory content isn't removed before.
So it can happen that old class files or descriptor files that are
never valid are still in the installation directory since they will
not be removed when the new stuff is installed. Existing files with
the same name will be overridden.
So my plan was to clean the target installation directory before the
new PEAR file is installed to it. If the target install directory
should be cleaned can be specified with an additional parameter when
using the PEAR installer API.
An known issue with that approach is, that after installing a PEAR
file to a directory, the same PEAR file cannot be installed again to
the same directory with the same JVM since the JVM has locked the jars
so they cannot be deleted when the directory for the next installation
should be cleaned. To do that, the JVM must be restarted, or another
installation directory must be used. The JVM locks the jar files when
the installation verification is executed.
What do others think about this approach?
-- Michael
