Konstantin Litvinenko writes:

tom fogal пишет:

...

This is a minor part of the application we're writing, so I can't really justify the size. Has anybody out there written a `decoupled' filesystem, with less dependencies on other boost libraries? Even a trimmed down version would probably suit or needs for quite some time.

As I understand you, your problem not a size of boost.filesystem or how many dependencies it has. The problem is that boost.filesystem can't be integrated into your build system with one simple action. Am I right?

No, adding external code to our tree is quite easy and I don't mind spending the 20 minutes it takes to do so. The size is really the root issue. I asked for a decoupled version because boost.filesystem is actually pretty small: tf@shigeru boost_1_38_0 du -hcs libs/filesystem/src/ boost/filesystem boost/filesystem.hpp 72K libs/filesystem/src/ 148K boost/filesystem 4.0K boost/filesystem.hpp 224K total tf@shigeru boost_1_38_0 so if I didn't need to pull in e.g. MPL to get filesystem, it wouldn't be as big a deal. -tom

On Wed, Apr 8, 2009 at 1:51 AM, Anton Daneyko wrote:

On Tue, Apr 7, 2009 at 8:27 PM, tom fogal wrote:

...

Upon build, it seemed to require some additional boost headers, so I added them; goto 10.

There's an utility called bcp to extract particluar pieces of boost (http://www.boost.org/doc/libs/1_38_0/tools/bcp/bcp.html). boost-root-dir$ bcp filesystem /destination_directory will copy all the dependencies you need.

...

This is a minor part of the application we're writing, so I can't really justify the size.

I won't give you any answer, it's just my blah-blah: I had a similar problem when I wanted to use boost program options -- the guys I was working with told me that it's just rediculous to pull such a load of code that exceeds or own code base by an order of magnitude. My point here that couple of dozens megs of boost code in your repository does not harm anyone. Okay, it's big, but who cares? So what? We now have plenty of space -- the space is not a problem per se -- our perception is the problem. What is the real show stoppage here, except someone's dim unconscious feelings? Anyway I was never persuasive enough -- they forced me to reinvent the wheel and do stuff by our own hands.

So in your case, like Konstantin Litvinenko (I suspect) tried to point out, there is no reason to stop yourself from using boost::filesystem.

Why not pull in the entire Boost then? It certainly wouldn't hurt, would it? :) I think that the request to keep coupling in boost::filesystem in check is valid. The fact that bcp exists means that many people are interested in acquiring just a part of Boost, which also implies that they don't want to depend on a lot of code. Your might be correct in your analysis, but without knowing enough details about each use case, you are only speculating. Emil Dotchevski Reverge Studios, Inc. http://www.revergestudios.com/reblog/index.php?n=ReCode