On Sun, 10 Jun 2001, Brian Ingerson wrote:
> With interpreted languages you technically don't need a build() because
> everything can be done on the fly. With AUTOLOAD or whatever. But all
> the current interpreted ILSMs like Python, Tcl, and Java *do* preparse
> their code in build() and save the bindable methods in the "object", in
> some proprietary format (like, say, Data::Dumper). Then load() just
> reads that object file, and binds to those methods.
I considered something like that but I don't think it can work very well
in the face of a dynamic language like Scheme. The same code could very
well produce different bind-able symbols on each successive evaluation.
Of course, the same could be said about Tcl (and Python? I don't know
enough to say). Perhaps it just seems more likely given schemers
penchant for self-generating code.
> Anyway, if you really do want to shun/circumvent the build/load
> processing, just use something like this to create an empty object. That
> will satisfy Inline.
Will do.
> If writing something to disk really bugs you, we can address that. But
> Inline::Guile will still create a .Inline/config file, so I wouldn't
> worry about it too much for now.
Ok. I'd say that it does bug me - it's a waste of resources to be
creating directories and writing files if you're not going to use them.
Perhaps we could add a register key to tell Inline to skip the filesystem
access? Hmmm.. no_disk, diskless, no_files, use_directory=>0...
Something like that.
-sam
