From: Gert D. <ger...@te...> - 2005-05-06 10:03:49
|
> -----Original Message----- > From: nan...@li... > [mailto:nan...@li...] On > Behalf Of Martin Aliger > Sent: vrijdag 6 mei 2005 11:54 > To: 'Gary Feldman'; nan...@li... > Subject: RE: [nant-dev] patch for FileSet extension > > >Interesting. I think there is an opportunity for optimizing this, to > eliminate the asis by changing the synchronization point. > >For example, if the contents of the foo1 fileset were not > computed until > actually needed, then you could do away with the asis and > apply the exclude. > > I'm not sure I want change something in such fundamental > thing as fileset. > Simple extension should be ok, but change some > functionality... I dont want > change whole NAnt, I just want to contribute small > enhancement which I find > very useful. Since I dont use asis myself I dont know what > for hacks it > could be good. > > To your second note: filesets are computed lazily - that > means when they > used for first time. There are some issues about it (like > creating fileset > then deleting some files and wondering why their names are > not stored in > fileset!). Merge operation is use of fileset, so I > explicitelly scan it. Its > real need becouse of merging two filesets with distinct basedirs. > > > What happens if you remove the implicit asis that you > > generate, and instead insert explicit asis attributes into > > the foo1 fileset? > > It workes well (except doubling bug 1195736), excludes > working. Only "issue" > is that every file from merged fileset is scanned twice. Its not big > performance issue, imo, since scan pathed file is just test for file > existence. I'll make some more testing with this config and > let you know. > > > (Regardless of what the decision is, the test cases need to > > include all combinations of asis, include, exclude, fromdir, > > relative basedir, absolute basedir, and omitted basedir.) > > That would be really nice. Unfortunatelly possible > combinations grow with > factorial - noone have time to write all those testcases. But > I'll try to > write a little more of them. I think you should await consensus that this feature is indeed necessary, and as such will end up in NAnt cvs. Gert |