SRFI implementations as libraries in standard Scheme.
Taylan Ulrich B 0e2843b3b9 srfi-tests/54.sld: Add missing `else' to `cond-expand' usage | 10 tahun lalu | |
---|---|---|
packages | 10 tahun lalu | |
srfi | 10 tahun lalu | |
srfi-tests | 10 tahun lalu | |
.dir-locals.el | 10 tahun lalu | |
.gitignore | 10 tahun lalu | |
COPYING | 10 tahun lalu | |
COPYING.LESSER | 10 tahun lalu | |
README.md | 10 tahun lalu | |
test-all.scm | 10 tahun lalu |
Our intention is to bridge gaps between credible Scheme platforms, as well as explore what is possible to implement sanely in R7RS. Unlike a lot of legacy "portable" Scheme code, this is not a freak show. Therefore:
Feel free to assume sane platform properties like a full numeric
tower (excluding exact complex numbers), Unicode, etc. and keep
cond-expand
usage to a minimum. (Moving to an R7RS-large base
will make this point moot.)
Feel free to use other SRFIs in your SRFI implementation, even if they're not in this repository, if the alternative is using horrid hacks.
Test suites are written in a strict format using SRFI-64 for human and machine consumption; see below.
For those SRFIs which cannot be implemented in pure R7RS, it's fine
to write libraries that just wrap features of specific platforms via
cond-expand
and thus at least work across several platforms even
if not all R7RS implementations. However, don't try to implement
functionality via library code when it's clearly intended to be
implemented at the platform level.
Withdrawn SRFIs are no priority since we assume they've been withdrawn for good reasons, but if there's any which you find useful then feel free to make a request or (better) contribute it.
Don't use the pure GPL for your libraries; people will be likely to just reimplement them with a different license if they don't like the GPL. I like copyleft, so would use the LGPL, but do as you wish.
As explained above, we expect some basic maturity and feature richness from platforms. So far this is not documented in any more detail, but it should be something along the lines of:
http://trac.sacrideo.us/wg/wiki/MandatoryDocket
If a library in this repository doesn't work on your preferred Scheme platform, file a bug report. If it contains a working patch it will most likely be accepted; otherwise it might get rejected on the grounds that the mentioned Scheme platform is not suitable.
The author of this repository uses Chibi to test the libraries. See the next section for instructions.
Other than Chibi, the following platforms are explicitly intended to be supported once they implement R7RS:
The following do not intend to support R7RS, for various reasons:
Source: http://lists.scheme-reports.org/pipermail/scheme-reports/2011-October/001595.html
Chibi Scheme is a clean, small, portable, and pretty much fully R7RS compliant Scheme platform, so it's a good choice for testing R7RS libraries.
Version 0.7 has some bugs and limitations; compiling from the upstream
repository is recommended at least until there is a newer release.
One can run Chibi from its source directory after compilation, but it
requires exporting LD_LIBRARY_PATH=$chibi_dir
, and unless the
executable chibi-scheme
is run from $chibi_dir
, one must add
$chibi_dir/lib
explicitly to the load path via the -I
(capital
'i') or -A
switches.
Prepending the path to this repository to the load-path of Chibi via
the -I
(capital 'i') flag will make the implementations here
override any native SRFI implementations of Chibi. However, that may
break Chibi's initialization process.
Instead the -A
switch may be used to append to the load-path, in
which case Chibi's SRFI implementations will take priority. That can
be worked around by visiting the directory $chibi_dir/lib/srfi
and
moving away the following files: 1.sld
, 2.sld
, 26.sld
, 27.sld
,
8.sld
, 95.sld
, 99.sld
. (This list may grow as Chibi adds more
SRFIs which clash with those in this repository.)
Most other SRFIs in the directory are ones that aren't in the scope of
this repository. (Specifically, many implement R7RS features, so
don't remove them!) The only remaining clash is 33.sld
, which seems
to be needed for Chibi's initialization, so be careful to import the
correct (srfi 33)
when testing with Chibi, for example by renaming
the srfi/33.sld
in this repository, as well as the library declared
in it, to some name that won't clash.
The SRFI libraries are offered through Snow packages as well:
http://taylanub-r7rs-srfis.s3-website-us-east-1.amazonaws.com/
(Thanks to Seth Alves for the hosting and help in making Snow packages for the libraries.)
To install an SRFI, you can install snow2 and run:
snow2 -r http://taylanub-r7rs-srfis.s3-website-us-east-1.amazonaws.com/index.scm install '(srfi n)'
for any SRFI-n
.
The repository is turned into a Snow repository by running snow2
package
in the top-level directory.
Test suites are also available as (srfi-tests n)
for any SRFI-n
.
All implementations should be valid R7RS-small libraries for now. We will move to an R7RS-large core/base when possible.
Libraries go into a file named n.sld
under the srfi
directory,
where n
is the SRFI number. The library is correspondingly named
(srfi n)
.
Test suites are libraries too, and go into n.sld
under the
srfi-tests
directory, and are named accordingly. They must export
a procedure named run-tests
defined via define-tests
from
(srfi-tests aux)
. The first argument to this macro must be the
identifier run-tests
, and the second must be the string "SRFI-n"
for every SRFI-n
. The rest of the arguments make up the body of
an SRFI-64 test suite; the initial test-begin
and final test-end
forms must not appear in this body, since define-tests
handles
that.
The following rules about libraries apply to all libraries in this project, including SRFI implementations and test suites.
The order of things in a library declaration is:
begin
if the main body is in an include
include
or begin
If an export or import list doesn't fit in one line, then put a
newline directly after the export
or import
keyword, i.e., don't
put any export or import specs on the same line as the keyword.
When there are many exports, put the closing parenthesis of the export form on its own line. Don't do this for imports.
If there is a substantial amount of code such that the indentation
of the library form is an annoyance, split the main body of the
library into a file named n.body.scm
and include
that.
When using a reference implementation with a substantial amount of
code, put the original source code in a file n.upstream.scm
, and
do not edit that file. include
directly that if no changes are
needed; otherwise copy it to n.body.scm
and make changes there.
This is because some reference implementations change ad-hoc without
version control; we want to know what version we forked.
Follow Riastradh's Lisp Style Rules for new code: http://mumble.net/~campbell/scheme/style.txt
You can probably often just wrap a reference implementation in an R7RS library, but do modify them to make good use of R7RS features, to avoid dirty hacky things like redefining type predicates (it's mostly illegal anyway now), liberal use of global variables (use parameters instead), etc.
Be careful about the copyright and licensing. When putting source
code in an n.upstream.scm
file, add suitable legal boilerplate
even if the original code didn't have any (usually the SRFI HTML
page will have a copyright). The same boilerplate goes into
n.body.scm
of course.
For some SRFIs, setting the debug-mode
parameter from the library
(srfi aux)
to true while loading the library will enable improved
diagnostics, usually at the expense of performance. Note that there
is no portable and declarative way to do this; it relies on the
platform supporting an imperative and sequential method for loading
modules and executing code in-between. It is most likely to work in a
REPL, if at all. A library might also be loaded only a single time
for the whole run-time of a platform, so a restart might be necessary
to load a library with a different debug-mode
setting. A number of
SRFIs also support other settings through the (srfi aux)
library;
the same limitations apply to those.
and-let*
: The author of this SRFI thought that let*
is supposed
to error when the same identifier is bound multiple times in
sequential clauses, and specified and-let*
to do the same; it was
in fact an ambiguity prior to R6RS what let*
should do, and has
been clarified since R6RS that it should allow sequential repeated
bindings. Racket's, Guile's, and Chibi's SRFI-2 implementations do
allow it, and it's a lot easier to implement too, hence our
implementation allows it as well.string-titlecase
: The module (srfi aux)
contains the parameters
char-cased?-proc
and char-titlecase-proc
which you can set while
loading this library (and only while loading it, not later) to
correct the behavior of this procedure. The default char-cased?
compares the upcased and downcased versions of a character to decide
whether it's cased, and char-titlecase
does a mere upcase.The SRFI proposes an R6RS library (streams)
with sub-libraries
(streams primitive)
and (streams derived)
. We provide these as
(srfi 41)
, (srfi 41 primitive)
, and (srfi 41 derived)
,
respectively.
All identifiers have been lowercased. If you have source code using
identifiers with uppercase letters, you might want to use the
#!fold-case
directive.
The reference implementation was not conforming to the specification, so I made it conform. I also ripped out source file/line reporting (name your tests instead; and for unnamed tests the expression will be used as a name) and overall rewrote most of the code while keeping the basic logic from the original reference implementation.
test-read-eval-string
: This now takes an optional env
argument
to specify in what environment to evaluate. Passing #f
will
explicitly attempt to call eval
with a single argument, which is
an error in R7RS. Leaving the argument out will try to use a sane
platform-specific default but falls back to #f
.The "u8vector" API in this SRFI is almost equivalent to the bytevector
API in R7RS. The only incompatibilities are that u8vector-copy!
takes a different order of arguments, and that there is no equivalent
of u8vector-compare
in R7RS. I see neither of these reason enough
to implement this SRFI, considering it to be subsumed by R7RS instead.
That being said, I augmented SRFI-67 with bytevector-compare
.
bytevector-compare
, bytevector-compare-as-list
: These are
natural additions to this SRFI since R7RS has bytevectors as an
additional sequence type.This is an odd one. It's not an SRFI in the strict sense, because it requests a change to the standard and not to implementations. In any case I marked it as subsumed by R7RS below because it seems that R7RS improved on the section which this SRFI requests improvements in, although not exactly in the way this SRFI asks for.
Since the "blob" type in this SRFI is obsoleted by bytevectors, we don't define it; we define bytevector equivalents of the procedures in this SRFI which don't already have a bytevector equivalent. Old uses of this SRFI should be simple to convert; a global replace of "blob" to "bytevector" in a body of code will get most of the job done.
endianness
: Given a library system that makes imported bindings
immutable, macros such as this one (which simulate enums) are both
redundant, and harmful, because they either have to match their
input unhygienically, or will be prone to accidental shadowing of
the simple identifiers they expect as input. Therefore, this macro
is not provided, and instead the three bindings endianness-big
,
endianness-little
, and endianness-native
exported.
endianness-native
: So far this is just set to big; you may set the
native-endianness
parameter from the (srfi aux)
module to
endianness-little
before loading this library.
All SRFI are listed here, and marked with one of the following:
UNTESTED and DRAFT are capitalized to emphasize that we're not done with them yet; all other non-empty tags mean we're done with it.
Don't let the UNTESTED tag scare you off; I minimally test any
implementation in Chibi, and some are really trivial, but so long as
an SRFI has no corresponding n.test.scm
, it's UNTESTED, period.
The r7rs tag can be vague. In the simplest case the SRFI is adopted
as-is in R7RS. Sometimes there are minor tweaks. Sometimes the same
essential functionality is provided in a different way in R7RS; a good
example might be SRFI-7, though it's equivalent to define-library
if
you pay attention.