December 2000 - Cvs-hugs

RE: cvs commit: hugs98/src GreenCard.h connect.h hugs.c input.c p arser.y static.c
by Simon Marlow 14 Dec '00

14 Dec '00

> reid 2000/12/13 01:36:05 PST > > Modified files: > src GreenCard.h connect.h hugs.c input.c > parser.y static.c > Log: > Final commit in the series. Alastair - this is great stuff. Nice one. We should work together on getting the FFI libraries into Hugs too, and sync up as much as possible. BTW, we're using (Ptr a) instead of Addr these days. Cheers, Simon

1 0

cvs commit: hugs98/src GreenCard.h connect.h hugs.c input.c parser.y static.c
by Alastair Reid 13 Dec '00

13 Dec '00

reid 2000/12/13 01:36:05 PST Modified files: src GreenCard.h connect.h hugs.c input.c parser.y static.c Log: Final commit in the series. Added +G flag to options in hugs.c. Lexer, parser and static checker additions. Note that foreign export (static) is not implemented. My problem was how to handle the following: plus x y = x + y foreign export "plusInt" plus :: Int -> Int -> Int foreign export "plusFloat" plus :: Float -> Float -> Float What I want to do is to define: __plusInt :: Int -> Int -> Int __plusInt = plus __plusFloat :: Float -> Float -> Float __plusFloat = plus then invoke the typechecker which will insert dictionaries, check that the instances make sense, etc. And now I can foreign export each individual instance with no bother at all. One problem is that ffi declarations are processed fairly early in the checking process - it's a bad time to go invoking the typechecker. Another problem is that I'm not entirely sure how to actually do the above. I think it should be like checking the bindings in an instance declaration but I don't quite grok that code. [This is a subtly disguised plea for help from Mark, Jeff or someone else who feels happy in the typechecker. (I'm assuming someone suitably qualified is on the commit mailing list.)] Anyway, enough about what if can't do. Here's what it can do: 1) Copy the following files into Test.hs and test.c 2) hugs +G Test.hs This generates Test.c 3) cc -shared Test.c test.c -o Test.so This generates Test.so 4) hugs Test.hs This loads Test.hs and Test.so 5) test (+) 1 1 This should print 2! I know it's a lot of work just to add two numbers... but there's a certain satisfaction in knowing that the addition was done in C isn't there? ---------------------------------------------------------------- -- Start of Haskell part of test file Test.hs ---------------------------------------------------------------- -- Doesn't support: -- o foreign export static -- o Int64, Word64, Int32, Word32, Int16, Word16, Int8, Word8 -- o The named C types HsChar, HsInt16, etc module Test where foreign import x :: Int -> Char -> IO Addr foreign import z :: Int -> Int -> Int foreign import dynamic x' :: Addr -> (Int -> Char -> IO Addr) -- foreign export static isn't implemented (typechecking problems) -- foreign export y :: Int -> Char -> IO Float -- y _ _ = return 1.0 foreign export dynamic toC :: (Int -> Int -> Int) -> IO Addr -- You can use either of these definitions - both work -- foreign import "fromC" fromC :: Addr -> (Int -> Int -> Int) fromC = fromfromC fromC_addr foreign import dynamic fromfromC :: Addr -> (Addr -> (Int -> Int -> Int)) foreign label "fromC" fromC_addr :: Addr test f x y = do{ f' <- toC f; print (fromC f' x y) } ---------------------------------------------------------------- -- Start of Haskell part of test file Test.hs ---------------------------------------------------------------- /**************************************************************** * Start of C part of test file test.c ****************************************************************/ extern void* y(int a, char b); void* x(int a, char b) { // return y(a,b); } int z(int a, int b) { return a+b; } void* from_dyn(void* (*f)(int i, char c),int i, char c) { return (*f)(i,c); } void* fromC(void* (*f)(int a, int b),int a, int b) { return (*f)(a,b); } /**************************************************************** * Start of C part of test file test.c ****************************************************************/ Revision Changes Path 1.4 +12 -5 hugs98/src/GreenCard.h 1.14 +9 -2 hugs98/src/connect.h 1.26 +8 -3 hugs98/src/hugs.c 1.16 +29 -2 hugs98/src/input.c 1.11 +44 -4 hugs98/src/parser.y 1.26 +357 -2 hugs98/src/static.c

1 0

cvs commit: hugs98/src ffi.c Makefile.in connect.h hugs.c machdep.c storage.c storage.h
by Alastair Reid 13 Dec '00

13 Dec '00

reid 2000/12/13 01:01:54 PST Modified files: src Makefile.in connect.h hugs.c machdep.c storage.c storage.h Added files: src ffi.c Log: ffi.c is the code generator. Very, very roughly, it is a translation of GreenCard's backend into C. Added mkFFIFilename to machdep.c This turns a Haskell filename like foo/bar/BBD.hs into BBD.c. This is where the generated code gets written. The file is written into the current directory because code generation interacts really badly with import chasing. There's no obvious way of turning code generation on/off on a module-by-module basis and it would be a really bad idea to try to scatter all those writes across the directory hierarchy (if we wrote into the same directory that the Haskell code was in). Added extFun and callconv fields to Name structure. Revision Changes Path 1.6 +3 -2 hugs98/src/Makefile.in 1.13 +11 -2 hugs98/src/connect.h 1.25 +3 -2 hugs98/src/hugs.c 1.9 +28 -2 hugs98/src/machdep.c 1.12 +4 -2 hugs98/src/storage.c 1.13 +4 -2 hugs98/src/storage.h

1 0

cvs commit: hugs98/lib Prelude.hs hugs98/lib/exts Foreign.hs hugs98/src connect.h type.c
by Alastair Reid 13 Dec '00

13 Dec '00

reid 2000/12/13 00:44:31 PST Modified files: lib Prelude.hs lib/exts Foreign.hs src connect.h type.c Log: Changes to make it easy for ffi implementation to refer to the ffi-able types: typeInt and friends used to be static (local to type.c). Now they are global (defined in connect.h). Also, ForeignObj, StablePtr and Word are now defined in the Prelude. Revision Changes Path 1.5 +5 -1 hugs98/lib/Prelude.hs 1.2 +2 -2 hugs98/lib/exts/Foreign.hs 1.12 +13 -2 hugs98/src/connect.h 1.27 +27 -14 hugs98/src/type.c

1 0

cvs commit: hugs98/src builtin.c storage.c storage.h
by Alastair Reid 13 Dec '00

13 Dec '00

reid 2000/12/13 00:28:57 PST Modified files: src builtin.c storage.c storage.h Log: Started adding some of the necessary infrastructure. Starting with the less invasive changes. Added concatText :: (String,String) -> Text to storage.c Added findQualFun :: (Text,Text) -> Name to storage.c Added HugsAPI3 struct to storage.h. This has a few extra operations required to implement the full FFI: void getUnit(void); int getBool(void); void putBool(int); HugsStablePtr lookupName(char*,char*); // look up qualified name void apMany(int); // apply function to m arguments These new functions are added at the end of the struct decl so that you could also use it as a HugsAPI2 struct if you wanted. (But we don't actually do that.) Added tag checks to the get functions. (Use --enable-tag-checks when configuring to turn these on.) Added code for building the foreign export dynamic thunk. Currently x86 specific. Added code to build a HugsAPI3 struct. Revision Changes Path 1.7 +198 -9 hugs98/src/builtin.c 1.11 +34 -2 hugs98/src/storage.c 1.12 +54 -2 hugs98/src/storage.h

1 0

cvs commit: hugs98/src type.c
by Alastair Reid 13 Dec '00

13 Dec '00

reid 2000/12/12 23:55:34 PST Modified files: src type.c Log: [Copied a trick I used in STG-Hugs back into Classic Hugs.] Added link{Name,Class,Tycon} which looks up a string and reports an error if it doesn't find it. Used for finding builtin constructors, classes, etc. in the typechecker. This change makes the code more readable, more robust, shorter and gives better error messages. Win, win, win, win. Revision Changes Path 1.26 +87 -112 hugs98/src/type.c

1 0

cvs commit: hugs98/src preds.c storage.c hugs98/src/unix configure configure.in
by Alastair Reid 13 Dec '00

13 Dec '00

reid 2000/12/12 23:43:37 PST Modified files: src preds.c storage.c src/unix configure configure.in Log: [This is the first of a series of commits from merging the ffi code into the Hugs repository. Some of the changes (such as this one) are really independent of having ffi support so I'm going to do them first.] Tweaked which flags you get if you configure with --enable-lint. It turns out that storage.h contains some function pointer declarations that won't compile (or maybe give a warning) if you write them one way and compile with Visual C++ and give a warning if you write them the other way and compile with gcc. Since any warnings from storage.h affect every file, it's best to just turn the warning off so that we can at least see the other warnings. Also fixed two easily fixed sources of warnings in preds.c and storage.c. Seems like it's been a while since anyone did a cleanup pass. This leaves a warning in input.c, output.c and runhugs.o (hugs.c) - these have been here a long, long time (i.e., I won't try to fix them now.) Revision Changes Path 1.27 +6 -2 hugs98/src/preds.c 1.10 +16 -16 hugs98/src/storage.c 1.5 +3 -3 hugs98/src/unix/configure 1.5 +6 -3 hugs98/src/unix/configure.in

1 0

cvs commit: hugs98/src version.h
by nordland＠glass.cse.ogi.edu 12 Dec '00

12 Dec '00

nordland 2000/12/12 12:47:39 PST Modified files: src version.h Log: Updated version string, just to check that commit works. Revision Changes Path 1.11 +3 -3 hugs98/src/version.h

1 0