happy Make Big Confusing Proposal Day

master
Joey Hess 2009-04-01 13:47:41 -04:00
parent 8f485a813e
commit f7037f4870
1 changed files with 68 additions and 0 deletions

View File

@ -0,0 +1,68 @@
[[!tag wishlist blue-sky]]
In the long term, I have been considering rewriting ikiwiki in haskell.
It's appealing for a lot of reasons, including:
* No need to depend on a C compiler and have wrappers. Instead, ikiwiki
binaries could be built on demand to do the things wrappers are used for
now (cgi, post-commit, etc).
* Potentially much faster. One problem with the now very modular ikiwiki is
that it has to load up dozens of perl modules each time it runs, which
means both opening lots of files and evaluating them. A haskell version
could run from one pre-compiled file. Other speed efficienies are also
likely with haskell. For example, pandoc is apparently an order of
magnitude faster than perl markdown implementations.
* Many plugins could be written in pure functional code, with no side
effects. Not all of them, of course.
* It should be much easier to get ikiwiki to support parallel compilation
on multi-core systems using haskell.
* A rewrite would be an opportunity to utterly break compatability and
redo things based on experience. Since the haskell libraries used for
markdown, templates, etc, are unlikely to be very compatable with the perl
versions, and since perl plugins obviously wouldn't work, and perl setup
files wouldn't be practical to keep, a lot of things would unavoidably
change, and at that point changinge everything else I can think of
probably wouldn't hurt (much).
- Re templates, it would be nice to have a template library that
doesn't use html-ish templating tags, since those are hard for users to
edit in html editors currently.
- This would be a chance to make WikiLinks with link texts read
"the right way round" (ie, vaguely wiki creole compatably).
- The data structures would probably be quite different.
- I might want to drop a lot of the command-line flags, either
requiring a setup file be used for those things, or leaving the
general-purpose `--set var=value` flag.
- Sometimes the current behavior of `--setup` seems confusing; it might
only cause a setup file to be read, and not force rebuild mode.
- Hard to say how the very high level plugin interface design would change,
but at the least some of the names of hooks could stand a rename, and
their parameter passing cleaned up.
We know that a big, break-the-world rewrite like this can be a very
bad thing for a project to attempt. It would be possible to support
external plugins written in haskell today, without any rewrite; and a few
of the benefits could be obtained by, eg, making the mdwn plugin be a
haskell program that uses pandoc. I doubt that wouod be a good first step
to converting ikiwiki to haskell, because such a program would have very
different data structures and intercommuniucation than a pure haskell
version.
Some other things to be scared about:
* By picking perl, I made a lot of people annoyed (and probably turned
several people away from using ikiwiki). But over time there turned out
to be a lot of folks who knew perl already (even if rustily), and made
some *very* useful contributions. I doubt there's as large a pool of haskell
programmers, and it's probably harder for a python user to learn haskell
than perl if they want to contribute to ikiwiki.
* It might be harder for users of hosting services to install a haskell based
ikiwiki than the perl version. Such systems probably don't have ghc and
a bunch of haskell libraries. OTOH, it might be possible to build a
static binary at home and upload it, thus avoiding a messy installation
procedure entirely.
* I can barely code in haskell yet. I'm probably about 100x faster at
programming in perl. I need to get some more practical experience before
I´m fast and seasoned enough in haskell to attempt such a project.
(And so far, progress at learning has been slow and I have not managed
to write anything serious in haskell.) --[[Joey]]