Gexp news

[Ludovic Courtès]

Hello Guix!

Last week I fiddled with gexps:

  • I rebased ‘wip-build-systems-gexp’ on ‘core-updates’ and
    improved/optimized a bunch of things.  It’s now close to ‘master’
    performance-wise, which was the main blocker for this branch.
    Hopefully I’ll look into it again next week or so.

    There’s more work to do, such as adapting some of the newer build
    systems, but the goal is to make it our next ‘core-updates’.

  • The ‘wip-gexp-hygiene’ branch makes gexps “hygienic”—i.e., gexps
    cannot capture lexically-bound identifiers introduced by other
    gexps, as in:

      (let* ((inner (lambda (x)
                      #~(let ((x 40)) (+ x #$x))))
             (outer  #~(let ((x 2))
                         #$(inner #~x))))
        (primitive-eval (gexp->sexp* outer)))

    This returns 42 because the ‘x’ in ‘inner’ does not shadow the ‘x’
    in ‘outer’ (and neither shadows the ‘x’ that is the formal parameter
    of ‘inner’.)  In current ‘master’ it returns 80.

    It does so through a custom alpha-renaming phase inspired by
    Kiselyov’s MetaScheme¹, similar to but much simpler than what the
    macro expander does.  That needs a bit more testing, but I think we
    can make probably try and add it on top of the above branch.

Feedback welcome!

Ludo’.

¹ http://okmij.org/ftp/meta-programming/#meta-scheme