Refactor rnd functions

.github/workflows/lint_cc.yml

@@ -15,5 +15,5 @@ jobs:
       - uses: DoozyX/[email protected]
         with:
           source: win/rl src/nle.c sys/unix/nledl.c include/nle.h include/nledl.h
-            include/nleobs.h
+            include/nletypes.h include/nlernd.h src/nlernd.c
           clangFormatVersion: 12

.github/workflows/test_and_deploy.yml

@@ -127,9 +127,7 @@ jobs:
         if: github.event_name != 'release'
         uses: pypa/[email protected] # The main configuration is in pyproject.toml
         env:
-          # Build of cp312-manylinux_aarch64 wheels failing, so restrict
-          # to only building _x86_64 wheel
-          CIBW_BUILD: "cp312-manylinux_*" # Build only python 3.12 wheels for testing
+          CIBW_BUILD: "cp312-manylinux*" # Build only python 3.12 wheels for testing
           # Increase verbosity to see what's going on in the build in case of failure
           CIBW_BUILD_VERBOSITY: 3
           CIBW_REPAIR_WHEEL_COMMAND_LINUX: >

.gitignore

@@ -210,6 +210,7 @@ nle/nethackdir
 # CMake
 CMakeFiles
 CMakeCache.txt
+CMakePresets.json
 
 # cibuildwheel
 wheelhouse/

CMakeLists.txt

@@ -64,15 +64,6 @@ add_compile_definitions(
   NOCWD_ASSUMPTIONS
   NLE_USE_TILES)
 
-option(USE_SEEDING "Use seeding support in NLE" ON)
-
-if(USE_SEEDING)
-  add_compile_definitions(NLE_ALLOW_SEEDING)
-  message("Seeding enabled.")
-else()
-  message("Seeding disabled.")
-endif()
-
 set(NLE_SRC ${nle_SOURCE_DIR}/src)
 set(NLE_INC ${nle_SOURCE_DIR}/include)
 set(NLE_DAT ${nle_SOURCE_DIR}/dat)
@@ -139,7 +130,10 @@ target_link_libraries(nethack PUBLIC m fcontext bz2_static tmt)
 
 # dlopen wrapper library
 add_library(nethackdl STATIC "sys/unix/nledl.c")
-target_include_directories(nethackdl PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/include)
+target_include_directories(
+  nethackdl
+  PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/include
+         ${CMAKE_CURRENT_SOURCE_DIR}/third_party/deboost.context/include)
 target_link_libraries(nethackdl PUBLIC dl)
 
 # rlmain C++ (test) binary

doc/nle/ARCHITECTURE.md

@@ -72,7 +72,7 @@ user.
 - **part of `libnethack.so`**
 - **key directories & files** [`src/nle.c`](https://github.com/heiner/nle/blob/main/src/nle.c),
 [`include/nle.h`](https://github.com/heiner/nle/blob/main/include/nle.h),
-[`include/nleobs.h`](https://github.com/heiner/nle/blob/main/include/nleobs.h)
+[`include/nletypes.h`](https://github.com/heiner/nle/blob/main/include/nletypes.h)
 
 
 * In our case the user “playing” the game is not a terminal, but ultimately a
@@ -116,7 +116,7 @@ user.:
         * a struct that contains the stack type and the two contexts
         (`return_context`, `generator_context`) which are the user and
         nethack-stack contexts respectively.
-    * `nle_obs from include/nleobs.h`
+    * `nle_obs from include/nletypes.h`
         * a struct that contains the observations transferred between the two
         stacks/contexts
 

include/nle.h

@@ -7,44 +7,20 @@
 
 #include <fcontext/fcontext.h>
 
-#include "nleobs.h"
+#include "nletypes.h"
 
 /* TODO: Fix this. */
 #undef SIG_RET_TYPE
 #define SIG_RET_TYPE void (*)(int)
 
-typedef struct TMT TMT;
-
-typedef struct nle_globals {
-    fcontext_stack_t stack;
-    fcontext_t returncontext;
-    fcontext_t generatorcontext;
-
-    FILE *ttyrec;
-    TMT *vterminal;
-    char outbuf[BUFSIZ];
-    char *outbuf_write_ptr;
-    char *outbuf_write_end;
-
-#ifdef NLE_BZ2_TTYRECS
-    void *ttyrec_bz2;
-#endif
-
-    boolean done;
-    nle_obs *observation;
-} nle_ctx_t;
-
 /*
  * Would like to annotate this with __thread, but that causes
  * the MacOS dynamic linker to not unload the library on dlclose().
  */
 nle_ctx_t *current_nle_ctx;
 
-nle_ctx_t *nle_start(nle_obs *, FILE *, nle_seeds_init_t *, nle_settings *);
+nle_ctx_t *nle_start(nle_obs *, FILE *, nle_settings *);
 nle_ctx_t *nle_step(nle_ctx_t *, nle_obs *);
 void nle_end(nle_ctx_t *);
 
-void nle_set_seed(nle_ctx_t *, unsigned long, unsigned long, boolean);
-void nle_get_seed(nle_ctx_t *, unsigned long *, unsigned long *, boolean *);
-
 #endif /* NLE_H */

include/nledl.h

@@ -7,7 +7,7 @@
 
 #include <stdio.h>
 
-#include "nleobs.h"
+#include "nletypes.h"
 
 typedef struct nledl_ctx {
     char dlpath[1024];
@@ -17,12 +17,10 @@ typedef struct nledl_ctx {
     FILE *ttyrec;
 } nledl_ctx;
 
-nledl_ctx *nle_start(const char *, nle_obs *, FILE *, nle_seeds_init_t *,
-                     nle_settings *);
+nledl_ctx *nle_start(const char *, nle_obs *, FILE *, nle_settings *);
 nledl_ctx *nle_step(nledl_ctx *, nle_obs *);
 
-void nle_reset(nledl_ctx *, nle_obs *, FILE *, nle_seeds_init_t *,
-               nle_settings *);
+void nle_reset(nledl_ctx *, nle_obs *, FILE *, nle_settings *);
 void nle_end(nledl_ctx *);
 
 void nle_set_seed(nledl_ctx *, unsigned long, unsigned long, char);

include/nlernd.h

@@ -0,0 +1,17 @@
+/*
+
+Set of functions to manipulate NetHack's Random Number Generators
+
+*/
+
+#ifndef NLERND_H
+#define NLERND_H
+
+
+#include "nletypes.h"
+
+
+void nle_set_seed(nle_ctx_t *, unsigned long, unsigned long, boolean);
+void nle_get_seed(nle_ctx_t *, unsigned long *, unsigned long *, boolean *);
+
+#endif

include/nleobs.h → include/nletypes.h

@@ -1,6 +1,10 @@
 
-#ifndef NLEOBS_H
-#define NLEOBS_H
+#ifndef NLETYPES_H
+#define NLETYPES_H
+
+#include <stdio.h>
+#include <stdbool.h>
+#include <fcontext/fcontext.h>
 
 #define NLE_MESSAGE_SIZE 256
 #define NLE_BLSTATS_SIZE 27
@@ -42,9 +46,19 @@
 #define NLE_BL_CONDITION 25 /* condition bit mask */
 #define NLE_BL_ALIGN 26
 
-/* #define NLE_ALLOW_SEEDING 1 */ /* Set in CMakeLists.txt if not disabled. */
 /* #define NLE_USE_TILES 1 */     /* Set in CMakeLists.txt. */
 
+/* NetHack defines boolean as follows:
+    typedef xchar boolean;      (global.h:80)
+    typedef schar xchar;        (global.h:73)
+    typedef signed char schar;  (config.h:420)
+
+So we'll do the same to avoid having to include all of NetHack's types
+*/
+typedef signed char boolean;
+
+typedef struct TMT TMT;
+
 typedef struct nle_observation {
     int action;
     int done;
@@ -72,14 +86,30 @@ typedef struct nle_observation {
 } nle_obs;
 
 typedef struct {
-#ifdef NLE_ALLOW_SEEDING
     unsigned long seeds[2]; /* core, disp */
     char reseed; /* boolean: use NetHack's anti-TAS reseed mechanism? */
-#else
-    int _dummy; /* empty struct has size 0 in C, size 1 in C++ */
-#endif
+    bool use_init_seeds; /* bool to tell NLE if seeds were provided */
 } nle_seeds_init_t;
 
+typedef struct nle_globals {
+    fcontext_stack_t stack;
+    fcontext_t returncontext;
+    fcontext_t generatorcontext;
+
+    FILE *ttyrec;
+    TMT *vterminal;
+    char outbuf[BUFSIZ];
+    char *outbuf_write_ptr;
+    char *outbuf_write_end;
+
+#ifdef NLE_BZ2_TTYRECS
+    void *ttyrec_bz2;
+#endif
+
+    boolean done;
+    nle_obs *observation;
+} nle_ctx_t;
+
 typedef struct nle_settings {
     /*
      *  Path to NetHack's game files.
@@ -98,6 +128,10 @@ typedef struct nle_settings {
      * Filename for nle's ttyrec*.bz2.
      */
     char ttyrecname[4096];
+
+    /* Initial seeds for the RNGs */
+    nle_seeds_init_t initial_seeds;
+
 } nle_settings;
 
-#endif /* NLEOBS_H */
+#endif /* NLETYPES_H */

nle/tests/test_envs.py

@@ -259,8 +259,6 @@ def test_rollout_no_archive(self, env_name, rollout_len):
 
     def test_seed_interface_output(self, env_name, rollout_len):
         """Tests whether env.seed output can be reused correctly."""
-        if not nethack.NLE_ALLOW_SEEDING:
-            return  # Nothing to test.
         if env_name.startswith("NetHackChallenge"):
             pytest.skip("Not running seed test on NetHackChallenge")
 
@@ -277,8 +275,6 @@ def test_seed_interface_output(self, env_name, rollout_len):
 
     def test_seed_rollout_seeded(self, env_name, rollout_len):
         """Tests that two seeded envs return same step data."""
-        if not nethack.NLE_ALLOW_SEEDING:
-            return  # Nothing to test.
         if env_name.startswith("NetHackChallenge"):
             pytest.skip("Not running seed test on NetHackChallenge")
 
@@ -302,8 +298,6 @@ def test_seed_rollout_seeded(self, env_name, rollout_len):
 
     def test_seed_rollout_seeded_int(self, env_name, rollout_len):
         """Tests that two seeded envs return same step data."""
-        if not nethack.NLE_ALLOW_SEEDING:
-            return  # Nothing to test.
         if env_name.startswith("NetHackChallenge"):
             pytest.skip("Not running seed test on NetHackChallenge")
 
@@ -468,9 +462,6 @@ def test_no_seed_setting(self):
             RuntimeError, match="NetHackChallenge doesn't allow seed changes"
         ):
             env.unwrapped.seed()
+
         with pytest.raises(RuntimeError, match="Should not try changing seeds"):
             env.unwrapped.nethack.set_initial_seeds(0, 0, True)
-
-        if not nethack.NLE_ALLOW_SEEDING:
-            with pytest.raises(RuntimeError, match="Seeding not enabled"):
-                env.unwrapped.nethack._pynethack.set_initial_seeds(0, 0, True)

nle/tests/test_nethack.py

@@ -119,8 +119,6 @@ def test_several_nethacks(self, game):
             game1.close()
 
     def test_set_initial_seeds(self):
-        if not nethack.NLE_ALLOW_SEEDING:
-            return  # Nothing to test.
         game = nethack.Nethack(copy=True)
         game.set_initial_seeds(core=42, disp=666)
         obs0 = game.reset()
@@ -135,9 +133,6 @@ def test_set_initial_seeds(self):
             game.close()
 
     def test_set_seed_after_reset(self, game):
-        if not nethack.NLE_ALLOW_SEEDING:
-            return  # Nothing to test.
-
         game.reset()
         # Could fail on a system without a good source of randomness:
         assert game.get_current_seeds()[2] is True

nle/tests/test_profile.py

@@ -55,8 +55,6 @@ def test_run_1k_steps(self, observation_keys, benchmark):
         actions = np.random.choice(env.action_space.n, size=steps)
 
         def seed():
-            if not nle.nethack.NLE_ALLOW_SEEDING:
-                return
             nonlocal seeds
             seeds += 1
             env.unwrapped.seed(seeds, 2 * seeds)