I'm a little confused about what the problem is. Could you give specific
examples?

I'm also unsure about why having a file-system cache will help. Other than
persistence, what advantage does having the cache in the file system
provide? I can see an argument that the file system forces a particular
organization to the cache, but the downside is that the implementation is
more complex than one in memory.

I've been assuming that it's OK to have more than one library in a single
file. The file-system approach doesn't handle that naturally, though
multiple links to a single file is reasonable.

On Sat, Nov 24, 2018 at 8:27 AM Taylor R Campbell <campb...@mumble.net>
wrote:

> Having worked on various different packaging and library installation
> systems, I've found it's always been a pain when there is either
>
> (a) a loading process that takes time proportional to the number of
> installed libraries (which either costs linear program startup time or
> quadratic library installation time like TeX, and can make the mere
> installation of a library cause a program that doesn't even use it to
> behave differently), or
>
> (b) a central cache file to reduce that cost, which has to be
> maintained by a special-purpose program to install and deinstall
> things (which doesn't have a natural merge operation that programs
> like tar or system package installer tools already know about).
>
> If we want libraries whose names are not tied to the file system, can
> we at least use the file system itself as a cache and install them at
> a predictable location for the loader?  E.g., when installing an sld
> file that has a library named (frob/nozzle möller), can we just make a
> file $libdir/mit-scheme-aarch64/sld/frob%2fnozzle/m%c3%b6ller with a
> pointer (maybe a symlink) to the real library file?
>
> _______________________________________________
> MIT-Scheme-devel mailing list
> MIT-Scheme-devel@gnu.org
> https://lists.gnu.org/mailman/listinfo/mit-scheme-devel
>
_______________________________________________
MIT-Scheme-devel mailing list
MIT-Scheme-devel@gnu.org
https://lists.gnu.org/mailman/listinfo/mit-scheme-devel

Reply via email to