X-Git-Url: http://git.megacz.com/?a=blobdiff_plain;f=utils%2Fext-core%2FREADME;h=84f213d7953334456d8a540cca11284f2a108d0a;hb=d107207d57f6102f580578e7c168b7317b04b9c4;hp=0f6c16b93237cc4ea64937533a7c2b998df3f100;hpb=6e93da5e0a775b2bfb9c9f2bd31a36cc828521cb;p=ghc-hetmet.git

diff --git a/utils/ext-core/README b/utils/ext-core/README
index 0f6c16b..84f213d 100644
--- a/utils/ext-core/README
+++ b/utils/ext-core/README
@@ -1,15 +1,65 @@
 A set of example programs for handling external core format.
 
 In particular, typechecker and interpreter give a precise semantics.
+---------------------
+tjc April/May 2008:
 
-To build, run "make".
+==== Documentation ====
+
+Documentation for the External Core format lives under docs/ext-core in the
+GHC tree. If you are building from HEAD, do not rely on the version of the
+External Core documentation that lives in haskell.org -- it is obsolete!
+
+==== Notes ====
+
+The checker should work on most programs. Bugs (and infelicities) 
+I'm aware of:
+   
+1. There's some weirdness involving funny character literals. This can
+   be fixed by writing a new lexer for chars rather than using Parsec's
+   built-in charLiteral lexer. But I haven't done that.
+
+2. The test driver attempts to find module dependencies automatically,
+   but it's slow. You can turn it off with the "-n" flag to the driver,
+   and specify all dependencies on the command line (you have to include
+   standard library dependencies too.)
+     * It would help to cache dependency info for standard libraries
+       in a file, or something, but that's future work.
+
+3. To avoid implementing some of the I/O primitives and foreign calls,
+   I use a gross hack involving replacing certain standard library
+   modules with simplified versions (found under lib/) that depend on
+   fake "primops" that the Core interpreter implements. This makes it
+   difficult (if not impossible) to load optimized versions of standard
+   libraries right now. Fixing this is future work too.
+
+Typechecking all the GHC libraries eats about a gig of heap and takes a
+long time. I blame Parsec. (Someone who was bored, or understood happy
+better than I do, could update the old happy parser, which is still in the
+repo.)
+
+The interpreter is also memory-hungry, but works for small programs
+that only do simple I/O (e.g., putStrLn is okay; not much more than that)
+and don't use Doubles or arrays. For example: exp3_8, gen_regexps, queens,
+primes, rfib, tak, wheel-sieve1, and wheel-sieve2, if modified so as not
+to take input or arguments.
+
+==== Building ====
+
+To run the checker and interpreter, you need to generate External Core
+for all the base, integer and ghc-prim libraries. This can be done by
+adding "-fext-core" to the GhcLibHcOpts in your build.mk file, then
+running "make" under libraries/.
 
-To run the checker and interpreter (which currently aren't working anyway),
-you need to generate External Core for all the base, integer and ghc-prim
-libraries. This can be done by adding "-fext-core" to the GhcLibHcOpts in
-your build.mk file, then running "make" under libraries/.
 Then you need to edit Driver.hs and change "baseDir" to point to your GHC
 libraries directory.
 
-Most recently tested with GHC 6.8.2. I make no claims of portability. --tjc
+Once you've done that:
+1. make prims (to generate the primops file)
+2. make
+3. make nofibtest (to run the parser/checker on all nofib programs...
+   for example.)
+
+Tested with GHC 6.8.2. I make no claims of portability.
+