From Gambit wiki
Revision as of 17:38, 28 November 2008 by Pflanze (Initial rant)
Why do I not like the "~~" and "~~foo" path ideas
by Christian Jaeger, please paint over (with your own graffiti or artwork) as you see fit.
- they are a layer on top of the underlying OS; and you can't access a file named '~~' in the current directory anymore
- they are not lispy, in the sense that they are a language by themselves, a sublanguage inside lisp. Just like regex strings are a sublanguage; for the latter a number of lisp/Scheme regex packages provide sexpr based regular expressions.
- I'm missing a good specification for the sublanguage; it feels ad hoc, which is ok for "local" things but when it comes to files put out into the wild, it's bad if the spec changes and you need to adapt the files.
- I understand that "~~" means "use a file delivered with core Gambit". This may seem ok now, but it doesn't allow transparent shadowing by the user. This may not seem to be an issue now, but compare with Perl: there would frequently be a module Foo provided with the default installation, but it might be too old for your application, and thus you install a newer Foo from CPAN and because it's coming in the search path before the core Foo you'll be using the newly installed Foo now. Gambit already delivers some utility code like digest.scm, this is a candidate for overriding.
- Specifying "~~lib/" instead of "~~/lib" to solve the problem where distribution makers are tearing apart the traditional ~~ directory contents, is requiring adaption of the currently written files. If we're going to adapt, it would better be a solution that is meant to stay quite some time. Will this be the case?
- Specifying the "lib" part at all seems to be redundant: the |include| can only deal with library files anyway, right? Or, if include files are put to "~~include" and library files to "~~lib" consistently, then |load| could as well know by itself that a ~~ path is meant to be a ~~lib path and |include| could know that a ~~ path is meant to be a ~~include path.
- (~~ is not understood by other Scheme systems, it's only usable for unportable files; this means if digest.scm is portable, it would better not be delivered with Gambit (if this means it requires ~~ in the path to access it).)
The only thing I wonder is whether the solution at http://scheme.ch/gambit/experimental/paths/paths.scm is too complicated or otherwise has a problem that makes it unsuited.
That solution *should* be usable for non-Gambit-core files, too. Should there be a hard distinction between core and non-core files? Should all code loading except for Gambit core files be handled by a module system which uses such a solution? But why, if the core procedures like |compile-file| and |load| are good enough for many things, also for non-core code. Is the above linked code really too much of a module/user space feature to be included in the core?
Which tools are handling paths how?
- Perl does it pretty much exactly as shown in http://scheme.ch/gambit/experimental/paths/paths.scm
- What about other interpreted languages?
- How does path resolving in gcc for -lfoo etc. exactly work? compare.
- other tools?