lisp.c

/* Distributed under Mozilla Public Licence 2.0   */
/* https://www.mozilla.org/en-US/MPL/2.0/         */
/* 2016-08-14 (C) Jonas S Karlsson, jsk@yesco.org */
/* A mini "lisp machine", main                    */

// http://stackoverflow.com/questions/3482389/how-many-primitives-does-it-take-to-build-a-lisp-machine-ten-seven-or-five

// in speed it's comparable to compiled LUA
// - simple "readline"
// - maybe 2x as slow
// - handles tail-recursion optimization
// - full closures
// - lexical binding

// RAW C esp8266 - printf("10,000,000 LOOP (100x lua) TIME=%d\r\n", tm); ===> 50ms

// Lua time 100,000 => 2s
// ----------------------
//   s=0; for i=1,100000 do s=s+i end; print(s);
//   function tail(n, s) if n == 0 then return s else return tail(n-1, s+1); end end print(tail(100000, 0))

// DEFINE(tail, (lambda (n s) (if (eq n 0) s (tail (- n 1) (+ s 1)))));
// princ(evalGC(reads("(tail 100000 0)"), env));
// -----------------------------------------------------------------
// lisp.c (tail 1,000 0)
//   all alloc/evalGC gives 5240ms with print
//   no print gc => 5040ms
//   ==> painful slow!
//   NO INT alloc => 4500ms
//   needGC() function, block mark => 1070ms
//
// lisp.c (tail 10,000 0)
//   => 9380, 9920, 10500
//   reuse() looped always from 0, made it round-robin => 3000... 4200... 4000ms!
//   mark_deep() + p->index=i ===> 1500-2000ms
//   car/cdr macro, on desktop 6.50 -> 4.5s for for 1M loop for esp => 1400-1800
//   hardcode primapply 2 parameters => 1100ms
//   hardcode primapply 1,2,3,-3 (if),-16 => 860-1100ms
//   slot alloc => 590,600 ms
//   eq => ==, remove one eval, tag test => 540,550ms
//
// lisp.c (tail 10,000 0)
//   now takes 5300ms more or less exactly every time
//   2.5x slower than lua, didn't actually measure actual lua time
//   evalGC now lookup vars too => 4780
//   bindEvalList (combined evallist + bindList) => 4040ms!
//   ... => 4010ms
//
//   slight increase if change the MAX_ALLOC to 512 but it keeps 17K free! => 4180ms

// We've slowed down some with adding the evallist/plus/times
// and trace functionality:
// > git checkout f871de73834340edb3fa5b26d49e43e373647f9b
//   ./opt
//   lisp> (time (fibo 34))
//   3 times, avg = 7015
// NOW
//   ./opt
//   lisp> (time (fibo 34))
//   3 times, avg = 7671


// 20151121: (time (fibo 24)) (5170 . xxx)
// 
// (fibo 24)
//   lua: 3s
//   esp-lisp: 5s
//
// (fibo 30)
//   x61 lisp: 4.920s
//   opt: 2.67s
//
// time echo "function fibo(n) if n < 2 then return 1; else return fibo(n-1) + fibo(n-2); end end print(fibo(35)); " | lua
//   14930352, real 0m6.006s, user 0m3.653s

// lua fibo 32 
//   3524578, real 0m1.992s, user 0m0.903s
//
// time ./a.out
//   3524578, real 0m19.353s, user 0m12.353s
//
// time ./opt
//   3524578, real 0m3.549s, user 0m2.253s
//  
// 2.5x slower than lua interpreting instead of compiling to byte code!!!
//

#include <unistd.h>
#include <ctype.h>
#include <stdarg.h>
#include <string.h>
#include <stdlib.h>
#include <stdio.h>
#include <errno.h>
#include <setjmp.h>
#ifndef UNIX
  #include "FreeRTOS.h"

  #define LOOP 99999
  #define LOOPS "99999"
  #define LOOPTAIL "(tail 99999 0)"

// #include "esp/spi.h"

#endif

#ifdef UNIX
  #define LOOP 2999999
  #define LOOPS "2999999"
  #define LOOPTAIL "(tail 2999999 0)"

#endif

// use pointer to store some tag data, use this for exteded types
// last bits (3 as it allocates in at least 8 bytes boundaries):
// ----------
// 000 = heap pointer, generic extended lisp data/struct with tag field - DONE
//  01 = integer << 2 - DONE
//  11 = inline symbol stored inside pointer! - DONE
//       32 bits = 6 chars * 5 bits = 30 bits + 11 or 3*ASCII(7)=28, if shifted
// ....fffff11 :  fffff != 11111 means 6 char atom name inline
// 00001111111 :  fffff == 11111 means 3 ascii atom name inline (names like "+" "-")
// xxxx1111111 :  xxxxx > 0 FREE! FREE! FREE!
// xxx11111111 :  24 bits left!!! == hashsymbol!!!
// TODO: 15 type of "enums" possible each with values of 3*7=21 bits 0-2,097,152
// 21+3 = 24 !!! use for hash syms!!!!
//
// -- byte[8] lispheap[MAX_HEAP]
// 010 = lispheap, cons == 8 bytes, 2 cells - DONE
// 100 = lispheap, symbol == name + primptr, not same as value
// 110 = ??

//  000 HEAP (string, symbol, prim...) - DONE
//  001 INTP 1 - DONE
//  010 CONSP 1 - heap - DONE
//  011 SYMP 1 inline pointer symbol - DONE
//  100       ??? hash symbol ??? ??? array cons???
//  101 INTP 2 - DONE
//  110 PRIMP
//  111 SYMP 2 inline pointer symbol - DONE
//
// ROM storage, issue of serializing atom and be able to execute from ROM symbols cannot change name
// so need "unique" pointer, but since symbols can be determined dynamically we cannot change
// in ROM. Let's say we use 24 bits hash (1M /usr/share/dict/words cause 290 clashes),
// 32-24 = 8 - 3 bits id => 5 (length) of following string/symbol name. len=0 means it's "unique":
// 1. a symbol can be SYMP (ROM/ROM compatible)
// 2. a symbol can be HEAP ALLOC
// 3. a symbol can be hash in ROM string (len < 2^5 = 32)
// 4. a symbol can be hash in RAM
//
// 3 & 4 needs to verify when used???? Hmmmm, harmonize number 2 with 3 & 4

// symbol hashes
// -------------
// (hash, pointer to value, pointer to string)
// (0, 0, 0) end, or use linked list

// linear list: (value, hash, inline string<32), ..., (0000, nil)
// extended cons list: (cons (symbol, value), ...)
// extended env list: next, value, symbol ("cons3")
// extended env list: next, value, hashp/string/value (var sized, "cons3+")

#include "lisp.h"
#include "compat.h"

// forwards
void gc_conses();
int kbhit();
static inline lisp callfunc(lisp f, lisp args, lisp* envp, lisp e, int noeval);
void error(char* msg);
void run(char* s, lisp* envp);

PRIM fundef(lisp f);
PRIM funenv(lisp f);
PRIM funame(lisp f);

lisp* global_envp = NULL;

// big value ok as it's used mostly no inside evaluation but outside at toplevel
#define READLINE_MAXLEN 512

// set to 1 to get GC tracing messages
// adding real debugging - http://software-lab.de/doc/tut.html#dbg
static int traceGC = 0;
static int trace = 0;

static int level = 0;
static int trace_level = 0;

static void indent(int n) {
    n *= 2;
    while (n-- > 0) putchar(' ');
}

// handle errors, break
jmp_buf lisp_break = {0};

// use for list(mkint(1), symbol("foo"), mkint(3), END);

// if non nil enables continous GC
// TODO: remove
static int dogc = 0;

typedef struct {
    char tag;
    char xx;
    short index;

    char* p; // TODO: make it inline, not second allocation
} string;

// conss name in order to be able to have a function named 'cons()'
// these are special, not stored in the allocated array
typedef struct {
    lisp car;
    lisp cdr;
} conss;

// lisp is a pointer with some semantics according to its lowest bits
// ==================================================================
// cons cells are 8 bytes, heap allocated objects are all 8 byte aligned too,
// thus, the three lowest bits aren't used so we use it for tagging/inline types.
//
// this is not ACCURATE!!! see later.... 
//
// 00000000 nil
//      000 heap allocated objects
//       01 int stored inline in the pointer
//      010 cons pointer into conses array
//       11 symbol names stored inside the pointer
//
//      100 special pointer, see below...
//     0100 UNUSED: ??? (IROM) symbol/string, n*16 bytes, zero terminated
//     1100 UNUSED: ??? (IROM) longcons (array consequtive nil terminated list) n*16 bytes (n*8 cars)
//
//     x1yy cons style things? but with other type
//     0110 UNUSED: maybe we have func/thunk/immediate
//     1110 UNUSED: 

// Espruino Javascript for esp8266 memory usage
// - http://www.esp8266.com/viewtopic.php?f=44&t=6574
// - uses 20KB, so 12 KB available for JS code + vars...

// stored using inline pointer
typedef struct {
    char tag;
    char xx;
    short index;

    int v;
} intint; // TODO: handle overflow...

// TODO: can we merge this and symbol, as all prim:s have name
typedef struct {
    char tag;
    signed char n; // TODO: maybe could use xx tag above?
    short index;

    void* f;
    char* name; // TODO: could point to symbol, however "foo" in program will always be there as string
} prim;

// Pseudo closure that is returned by if/progn and other construct that takes code, should handle tail recursion
typedef struct thunk {
    char tag;
    char xx;
    short index;

    lisp e;
    lisp env;
    // This needs be same as immediate
} thunk;

typedef struct immediate {
    char tag;
    char xx;
    short index;

    lisp e;
    lisp env;
    // This needs be same as thunk
} immediate;

typedef struct func {
    char tag;
    char xx;
    short index;

    lisp e;
    lisp env;
    lisp name; // TODO: recycle
} func;

int tag_count[MAX_TAGS] = {0};
int tag_bytes[MAX_TAGS] = {0};
int tag_freed_count[MAX_TAGS] = {0};
int tag_freed_bytes[MAX_TAGS] = {0};

char* tag_name[MAX_TAGS] = { "total", "string", "cons", "int", "prim", "symbol", "thunk", "immediate", "func", 0 };
int tag_size[MAX_TAGS] = { 0, sizeof(string), sizeof(conss), sizeof(intint), sizeof(prim), sizeof(thunk), sizeof(immediate), sizeof(func) };

int gettag(lisp x) {
    return TAG(x);
}

// essentially total number of cons+symbol+prim
// TODO: remove SYMBOL since they are never GC:ed! (thunk are special too, not tracked)
//#define MAX_ALLOCS 819200 // (fibo 22)
//#define MAX_ALLOCS 8192
//#define MAX_ALLOCS 1024 // keesp 15K free
//#define MAX_ALLOCS 512 // keeps 17K free
//#define MAX_ALLOCS 256 // keeps 21K free
//#define MAX_ALLOCS 128 // keeps 21K free
//#define MAX_ALLOCS 128 // make slower!!!
#define MAX_ALLOCS 256 // make faster???

int allocs_count = 0; // number of elements currently in used in allocs array
int allocs_next = 0; // top of allocations in allocs array
void* allocs[MAX_ALLOCS] = { 0 };
unsigned int used[MAX_ALLOCS/32 + 1] = { 0 };
    
#define SET_USED(i) ({int _i = (i); used[_i/32] |= 1 << _i%32;})
#define IS_USED(i) ({int _i = (i); (used[_i/32] >> _i%32) & 1;})

// any slot with no value/nil can be reused
int reuse_pos = 0;
int reuse() {
    int n = allocs_next;
    while(n--) {
        if (!allocs[reuse_pos]) return reuse_pos;
        reuse_pos++;
        if (reuse_pos >= allocs_next) reuse_pos = 0;
    }
    return -1;
}

// total number of things in use
int used_count = 0;
int used_bytes = 0;

// TODO: maybe not good idea, just pre allocate 12 and 16 bytes stuff 4 at a time and put on free list?
// permanent malloc, no way to give back
#define PERMALLOC_CHUNK 64

// used_count=72 cons_count=354 free=19280 USED=12 bytes
// used_count=72 cons_count=354 free=19580 USED=16 bytes
// saved: (- 19368 19580) = -122
// 84 mallocs = (* 84 4) 336 bytes "saved"?

// used_count=72 cons_count=354 free=19288 USED=12 bytes
// used_count=72 cons_count=354 free=19580 USED=16 bytes
// used_count=72 cons_count=354 free=18892 USED=12 bytes

// unmodified: used_count=72 cons_count=354 free=18900 USED=12 bytes startMem=34796 
// this:       used_count=72 cons_count=354 free=19540 USED=16 bytes startMem=34796
// (- 34796 19540) = 15256

void* perMalloc(int bytes) {
    if (bytes > PERMALLOC_CHUNK) return malloc(bytes);

    static char* heap = NULL;
    static int used = 0;
    if (used + bytes > PERMALLOC_CHUNK) {
//        printf("[perMalloc.more waste=%d]\n", PERMALLOC_CHUNK - used);
        // TODO: how to handle waste? if too much, then not worth it...
        heap = NULL;
        used = 0;
    }
    if (!heap) heap = malloc(PERMALLOC_CHUNK);
    void* r = heap;
//    printf("[perMalloc %d]", bytes); fflush(stdout);
    heap += bytes;
    used += bytes;
    return r;
}

// SLOT salloc/sfree, reuse mallocs of same size instead of free, saved 20% speed
#define SALLOC_MAX_SIZE 32
void* alloc_slot[SALLOC_MAX_SIZE] = {0}; // TODO: probably too many sizes...

void sfree(void** p, int bytes, int tag) {
    if (IS((lisp)p, symboll) || CONSP((lisp)p)) {
        error("sfree.ERROR: symbol or cons!\n");
    }
    if (bytes >= SALLOC_MAX_SIZE) {
        used_bytes -= bytes;
        return free(p);
    }
    /// TODO: use sfree?
    if (IS((lisp)p, string)) {
        free(((string*)p)->p);
    }

    // store for reuse
    void* n = alloc_slot[bytes];
    *p = n;
    alloc_slot[bytes] = p;
    // stats
    if (tag > 0) {
        tag_freed_count[tag]++;
        tag_freed_bytes[tag] += bytes;
    }
    tag_freed_count[0]++;
    tag_freed_bytes[0] += bytes;
}

static void* salloc(int bytes) {
    void** p = alloc_slot[bytes];
    if (bytes >= SALLOC_MAX_SIZE) {
        used_bytes += bytes;
        return malloc(bytes);
    } else if (!p) {
        used_bytes += bytes;
        return malloc(bytes);
        int i = 8;
        char* n = malloc(bytes * i);
        while (i--) {
            sfree((void*)n, bytes, -1);
            n += bytes;
        }
        p = alloc_slot[bytes];
    }
    alloc_slot[bytes] = *p;
    return p;
}

// call this malloc using ALLOC(typename) macro
// if tag < 0 no GC on these (don't keep pointer around)
void* myMalloc(int bytes, int tag) {
    ///printf("MALLOC: %d %d %s\n", bytes, tag, tag_name[tag]);

    if (1) { // 830ms -> 770ms 5% faster if removed, depends on the week!?
        if (tag > 0) {
            tag_count[tag]++;
            tag_bytes[tag] += bytes;
            used_count++;
        }

        tag_count[0]++;
        tag_bytes[0] += bytes;
    }

    // use for heap debugging, put in offending addresses
    //if (allocs_next == 269) { printf("\n==============ALLOC: %d bytes of tag %s ========================\n", bytes, tag_name[tag]); }
    //if ((int)p == 0x08050208) { printf("\n============================== ALLOC trouble pointer %d bytes of tag %d %s ===========\n", bytes, ag, tag_name[tag]); }

    void* p = salloc(bytes);

    // immediate optimization, only used transiently, so given back fast, no need gc.
    // symbols and prims are never freed, so no need keep track of or GC
    if (tag <= 0 || tag == immediate_TAG || tag == symboll_TAG || tag == prim_TAG) {
        ((lisp)p)->index = -1;
        return p;
    }

    int pos = reuse();
    if (pos < 0) pos = allocs_next++;

    allocs[pos] = p;
    allocs_count++;
    ((lisp)p)->index = pos;

    if (allocs_next >= MAX_ALLOCS) {
        report_allocs(2);
        error("Exhausted myMalloc array!\n");
    }
    return p;
}

static void mark_clean() {
    memset(used, 0, sizeof(used));
}

static int blockGC = 0;

PRIM gc(lisp* envp) {
    if (blockGC) {
        printf("\n%% [warning: GC called with blockGC=%d]\n", blockGC);
        return nil;
    }

    // mark
    syms_mark();

    //if (envp) { printf("ENVP %u=", (unsigned int)*envp); princ(*envp); terpri();}
    if (envp) mark(*envp);

    // sweep
    gc_conses();
    
    int count = 0;
    int i ;
    for(i = 0; i < allocs_next; i++) {
        lisp p = allocs[i];
        if (!p) continue;
        if (INTP(p) || CONSP(p)) {
            printf("GC.erronious pointer stored: %u, tag=%d\n", (int)p, TAG(p));
            printf("VAL="); princ(p); terpri();
            exit(1);
        }
        
        // USE FOR DEBUGGING SPECIFIC PTR
        //if ((int)p == 0x0804e528) { printf("\nGC----------------------%d ERROR! p=0x%x  ", i, p); princ(p); terpri(); }

        if (TAG(p) > 8 || TAG(p) == 0) {
            printf("\nGC----------------------%d ILLEGAL TAG! %d p=0x%x  ", i, TAG(p), (unsigned int)p); princ(p); terpri();
        }
        int u = (used[i/32] >> i%32) & 1;
        if (u) {
            // printf("%d used=%d  ::  ", i, u); princ(p); terpri();
        } else {
            count++;
            if (1) {
                sfree((void*)p, tag_size[TAG(p)], TAG(p));;
            } else {
                printf("FREE: %d ", i); princ(p); terpri();
                // simulate free
                p->tag = 66;
            }
            allocs[i] = NULL;
            allocs_count--;
            used_count--;
        }
    }
    mark_clean();

    return mem_usage(count);
}


////////////////////////////////////////////////////////////////////////////////
// string

// only have this in order to keep track of allocations
char* my_strndup(char* s, int len) {
    int l = strlen(s);
    if (l > len) l = len;
    char* r = myMalloc(len + 1, -1);
    strncpy(r, s, len);
    r[len] = 0;
    return r;
}

// make a string from POINTER (inside other string) by copying LEN bytes
// if len < 0 then it's already malloced elsewhere
PRIM mklenstring(char* s, int len) {
    if (!s) return nil;
    string* r = ALLOC(string);
    if (len >= 0) {
        r->p = my_strndup(s, len); // TODO: how to deallocate?
    } else {
        r->p = s;
    }
    return (lisp)r;
}

PRIM mkstring(char* s) {
    return mklenstring(s, strlen(s));
}

char* getstring(lisp s) {
    return IS(s, string) ? ATTR(string, s, p) : "";
}
    
// TODO:
// (string-ref s 0index)
// (string-set! s 0index char) -- maybe not modify
// string=? (substring=? s1 start s2 start end)
// cmp instead of string-compare
// string-hash, string-upcase/string-downcase
// (substring s start end)
// (string-search-forward pattern string)
// (substring-search-forward pattern string start end)
// (string-search-all pattern start end) => (pos1 pos2...) / nil

// srfi.schemers.org/srfi-13/srfi-13.html -- TOO BOORING!!!
//   predicate: string? string-null? string-every string-any
//   construct: make-string string string-tabulate
//   convertn: string->list list->string reverse-list->string string-join
//   select: string-length string-ref string-copy substring/shared string-copy!
//     string-take string-take-right string-drop- string-drop-right
//     string-pad string-pad-right string-trim string-trim-both
//   modify: string-set! string-fill!
//   compare: string-compare string<> string= string< string> string<= string>=
//     string-hash
//   prefix: string-prefix-length string-suffix-length string-prefix? string-suffix?
//   searching: string-index string-index-right string-skip string-skip-right string-count string-contains
//   reverse & append: string-reverse string-append string-concatenate
//   modify: string-replace string-delete

////////////////////////////////////////////////////////////////////////////////
// CONS

//#define MAX_CONS 137 // allows allocation of (list 1 2)
#define MAX_CONS 2048
//#define MAX_CONS 512
#define CONSES_BYTES (MAX_CONS * sizeof(conss))

conss* conses = NULL;
unsigned int cons_used[MAX_CONS/32 + 1] = { 0 };
lisp free_cons = 0;
int cons_count = 0; // TODO: remove, used for GC indication

#define CONS_SET_USED(i) ({int _i = (i); cons_used[_i/32] |= 1 << _i%32;})
#define CONS_IS_USED(i) ({int _i = (i); (cons_used[_i/32] >> _i%32) & 1;})

// TODO: as alternative to free list we could just use the bitmap
// this would allow us to allocate adjacent elements!
void gc_conses() {
    // make first pointer point to first position
    free_cons = nil;
    cons_count = 0;
    int i;
    for(i = MAX_CONS - 1; i >= 0; i--) {
        if (!CONS_IS_USED(i)) {
            conss* c = &conses[i];
            //if (c->car && c->car == _FREE_) continue; // already in free list...
            c->car = _FREE_;
            c->cdr = free_cons;
            free_cons = MKCONS(c);
            //printf("%u CONS=%u CONSP=%d\n", (int)c, (int)free_cons, CONSP(free_cons));
            cons_count++;
        }
    }
    memset(cons_used, 0, sizeof(cons_used));
}

void gc_cons_init() {
    conses = malloc(CONSES_BYTES); // we malloc, so it's pointer starts at xxx000
    memset(conses, 0, CONSES_BYTES);
    memset(cons_used, 0, sizeof(cons_used));
    gc_conses();
}

PRIM cons(lisp a, lisp b) {
    conss* c = GETCONS(free_cons);
    cons_count--;
    if (!c) {
        error("Run out of conses\n");
    }
    if (cons_count < 0) {
        error("Really ran out of conses\n");
    }
    if (c->car != _FREE_) {
        printf("Conses corruption error %u ... %u CONSP=%d\n", (int)c, (int)free_cons, CONSP(free_cons));
        printf("CONS="); princ((lisp)c); terpri();
        exit(1);
    }
    // TODO: this is updating counter in myMalloc stats, maybe refactor...
    if (0) { // TOOD: enable this and it becomes very slow!!!!??? why compared to myMalloc shouldn't????
    used_count++; // not correct as cons are different...
    tag_count[conss_TAG]++;
    tag_bytes[conss_TAG] += sizeof(conss);
    tag_count[0]++;
    tag_bytes[0] += sizeof(conss);
    }

    free_cons = c->cdr;

    c->car = a;
    c->cdr = b;
    return MKCONS(c);
}

PRIM recons(lisp a, lisp b, lisp ab) {
    if (a == car(ab) && b == cdr(ab)) return ab;
    return cons(a, b);
}

// inline works on both unix/gcc and c99 for esp8266
inline PRIM car(lisp x) { return CONSP(x) ? GETCONS(x)->car : nil; }
inline PRIM cdr(lisp x) { return CONSP(x) ? GETCONS(x)->cdr : nil; }

PRIM nthcdr(lisp n, lisp l) {
    int in = getint(n);
    while (in-- > 0) l = cdr(l);
    return l;
}

PRIM nth(lisp n, lisp l) {
    if (getint(n) < 0) return nil;
    return car(nthcdr(n, l));
}

int cons_count; // forward

// however, on esp8266 it's only inlined and no function exists,
// so we need to create them for use in lisp
#ifdef UNIX
  #define car_ car
  #define cdr_ cdr
#else
  PRIM car_(lisp x) { return car(x); }
  PRIM cdr_(lisp x) { return cdr(x); }
#endif

PRIM setcar(lisp x, lisp v) { return IS(x, conss) ? GETCONS(x)->car = v : nil; }
PRIM setcdr(lisp x, lisp v) { return IS(x, conss) ? GETCONS(x)->cdr = v : nil; }

PRIM list(lisp first, ...) {
    va_list ap;
    lisp r = nil;
    // points to cell where cdr is next pos
    lisp last = r;
    lisp x;

    va_start(ap, first);
    for (x = first; x != (lisp)-1; x = va_arg(ap, lisp)) {
        lisp nw = cons(x, nil);
        if (!r) r = nw;
        setcdr(last, nw);
        last = nw;
    }
    va_end(ap);

    return r;
}

void report_allocs(int verbose) {
    int i;

    if (verbose) terpri();
    if (verbose == 2)
        printf("--- Allocation stats ---\n");

    if (verbose == 1) {
        printf("\nAllocated: ");
        for(i = 0; i<16; i++)
            if (tag_count[i] > 0) printf("%d %s=%d bytes, ", tag_count[i], tag_name[i], tag_bytes[i]);
        
        printf("\n    Freed: ");
        for(i = 0; i<16; i++)
            if (tag_freed_count[i] > 0) printf("%d %s=%d bytes, ", tag_count[i], tag_name[i], tag_bytes[i]);

        printf("\n     Used: ");
    }

    for(i = 0; i<16; i++) {
        if (tag_count[i] > 0 || tag_freed_count[i] > 0) {
            int count = tag_count[i] - tag_freed_count[i];
            int bytes = tag_bytes[i] - tag_freed_bytes[i];
            if (verbose == 2) 
                printf("%12s: %3d allocations of %5d bytes, and still use %3d total %5d bytes\n",
                       tag_name[i], tag_count[i], tag_bytes[i], count, bytes);
            else if (verbose == 1 && (count > 0 || bytes > 0))
                printf("%d %s=%d bytes, ", count, tag_name[i], bytes);
        }
    }

    if (verbose == 2) {
        for(i = 0; i<16; i++) {
            if (tag_count[i] > 0 || tag_freed_count[i] > 0) {
                int count = tag_count[i] - tag_freed_count[i];
                int bytes = tag_bytes[i] - tag_freed_bytes[i];

                if (verbose == 1 && (tag_count[i] != count || tag_bytes[i] != bytes) && (tag_count[i] || tag_bytes[i]))
                    printf("churn %d %s=%d bytes, ", tag_count[i], tag_name[i], tag_bytes[i]);
            }
        }
    }
    
    for(i = 0; i<16; i++) {
        tag_count[i] = 0;
        tag_bytes[i] = 0;
        tag_freed_count[i] = 0;
        tag_freed_bytes[i] = 0;
    }

    // print static sizes...
    if (verbose) {
        int tot = 0, b;
        printf("\nSTATICS ");
        b = sizeof(tag_name); printf("tag_name: %d ", b); tot += b;
        b = sizeof(tag_size); printf("tag_size: %d ", b); tot += b;
        b = sizeof(tag_count); printf("tag_count: %d ", b); tot += b;
        b = sizeof(tag_bytes); printf("tag_bytes: %d ", b); tot += b;
        b = sizeof(tag_freed_count); printf("tag_freed_count: %d ", b); tot += b;
        b = sizeof(tag_freed_bytes); printf("tag_freed_bytes: %d ", b); tot += b;
        b = sizeof(allocs); printf("allocs: %d ", b); tot += b;
        b = sizeof(alloc_slot); printf("alloc_slot: %d ", b); tot += b;
        b = CONSES_BYTES; printf("conses: %d ", b); tot += b;
        b = sizeof(cons_used); printf("cons_used: %d ", b); tot += b;
        printf(" === TOTAL: %d\n", tot);
    }

    // TODO: this one doesn't make sense?
    if (verbose) {
        printf("\nused_count=%d cons_count=%d ", used_count, cons_count);
        fflush(stdout);
    }
}

lisp mkint(int v) {
    return MKINT(v);

    // TODO: add "bigint" or minusint...
    intint* r = ALLOC(intint);
    r->v = v;
    return (lisp)r;
}

int getint(lisp x) {
    return INTP(x) ? GETINT(x) : 0;

    // TODO: add "bigint" or minusint...
    return IS(x, intint) ? ATTR(intint, x, v) : 0;
}

PRIM eq(lisp a, lisp b);

PRIM member(lisp e, lisp r) {
    while (r) {
        if (eq(e, car(r))) return r;
        r = cdr(r);
    }
    return nil;
}

PRIM out(lisp pin, lisp value) {
    gpio_enable(getint(pin), GPIO_OUTPUT);
    gpio_write(getint(pin), getint(value));
    return value;
}

PRIM in(lisp pin) {
    gpio_enable(getint(pin), GPIO_INPUT);
    return mkint(gpio_read(getint(pin)));
}

PRIM dht(lisp pin) {
    int t, h;
    if (dht_read(pin, &t, &h)) return nil;
    return cons(mkint(t), mkint(h));
}

// CONTROL INTERRUPTS:
// -------------------
// (interrupt PIN 0)  : disable
// (interrupt PIN 1)  : EDGE_POS
// (interrupt PIN 2)  : EDGE_NEG
// (interrupt PIN 3)  : EDGE_ANY
// (interrupt PIN 4)  : LOW
// (interrupt PIN 5)  : HIGH
//
// TODO: this has 200 ms, "ignore" if happen within 200ms, maybe add as parameter?
//
// CALLBACK API:
// -------------
// If any interrupt is enabled it'll call intXX where XX=pin if the symbol exists.
// This is called from the IDLE loop, no clicks counts will be lost, clicks is the
// new clicks since last invokation/clear. It will only be invoked once if any was
// missed, and only the last time in ms is retained.
//
//   (define (int00 pin clicks count ms)
//     (printf " [button %d new clicks=%d total=%d last at %d ms] " pin clicks count ms))
// 
// POLLING API:
// ------------
// (interrupt PIN)    : get count
// (interrupt PIN -1) : get +count if new, or -count if no new
// (interrupt PIN -2) : get +count if new, or 0 otherwise
// (interrupt PIN -3) : get ms of last click
PRIM interrupt(lisp pin, lisp changeType) {
    if (!pin && !changeType) return nil;
    int ct = getint(changeType);
    if (changeType && ct >= 0) {
        interrupt_init(getint(pin), ct);
        return pin;
    } else {
        return mkint(getInterruptCount(getint(pin), changeType ? ct : 0));
    }
}

// wget functions...
// echo '
// (wget "yesco.org" "http://yesco.org/index.html" (lambda (t a v) (princ t) (cond (a (princ " ") (princ a) (princ "=") (princ v)(terpri)))))
// ' | ./run

static void f_emit_text(lisp callback, char* path[], char c) {
    maybeGC();

    apply(callback, list(mkint(c), END)); // more effcient with "integer"/char
}

static void f_emit_tag(lisp callback, char* path[], char* tag) {
    maybeGC();
    
    apply(callback, list(symbol(tag), END));
}

static void f_emit_attr(lisp callback, char* path[], char* tag, char* attr, char* value) {
    maybeGC();
    
    apply(callback, list(symbol(tag), symbol(attr), mkstring(value), END));
}

// TODO: http://www.gnu.org/software/mit-scheme/documentation/mit-scheme-ref/XML-Input.html#XML-Input
// https://www.gnu.org/software/guile/manual/html_node/Reading-and-Writing-XML.html
// https://www.gnu.org/software/guile/manual/html_node/sxml_002dmatch.html
PRIM wget_(lisp server, lisp url, lisp callback) {
    wget_data data;
    memset(&data, 0, sizeof(data));
    data.userdata = callback;
    data.xml_emit_text = (void*)f_emit_text;
    data.xml_emit_tag = (void*)f_emit_tag;
    data.xml_emit_attr = (void*)f_emit_attr;

    wget(&data, getstring(url), getstring(server));
    return nil;
}

// web functions...

// Generalize, similarly to xml stuff, with userdata etc, in order to handle several servers
static lisp web_callback = NULL;

static void header(char* buff, char* method, char* path) {
    buff = buff ? buff : "";
    maybeGC();

    apply(web_callback, list(symbol("header"), mkstring(buff), symbol(method), mkstring(path), END));
}

static void body(char* buff, char* method, char* path) {
    buff = buff ? buff : "";
    maybeGC();

    apply(web_callback, list(symbol("body"), mkstring(buff), symbol(method), mkstring(path), END));
}

static void response(int req, char* method, char* path) {
    maybeGC();

    lisp ret = apply(web_callback, list(nil, mkint(req), symbol(method), mkstring(path), END));
    printf("RET="); princ(ret); terpri();

    // TODO: instead redirect output to write!!! 
    char* s = getstring(ret);

    do {
        int r = write(req, s, strlen(s));
        if (r < 0) { printf("%%Error on writing response, errno=%d\n", errno); break; }
        s += r;
    } while (*s);

    maybeGC();
}

// echo '
// (web 8080 (lambda (r w s m p) (princ w) (princ " ") (princ s) (princ " ") (princ m) (princ " ") (princ p) (terpri) "FISH-42"))
// ' | ./run

int web_socket = 0;

int web_one() {
    int r = -1;
    if (!web_socket) return 0;
    if (setjmp(lisp_break) == 0) {
        r = httpd_next(web_socket, header, body, response);
    } else {
        printf("\n%%web_one.error... recovering...\n");
    }
    // disable longjmp
    memset(lisp_break, 0, sizeof(lisp_break));
    return r;
}

PRIM web(lisp* envp, lisp port, lisp callback) {
    //wget_data data;
    //memset(&data, 0, sizeof(data));
    //data.userdata = callback;
    //data.xml_emit_text = (void*)f_emit_text;
    //data.xml_emit_tag = (void*)f_emit_tag;
    //data.xml_emit_attr = (void*)f_emit_attr;

    // store a pointer in global env to the function so it doesn't get gc:ed
    web_callback = evalGC(callback, envp);
    _define(envp, list(symbol("webcb"), reads("web_callback"), END));

    int s = httpd_init(getint(port));
    if (s < 0) { printf("ERROR.errno=%d\n", errno); return nil; }

    web_socket = s;

    web_one();
    return mkint(s);
}


// lookup binding of symbol variable name (not work for int names)
PRIM assoc(lisp name, lisp env) {
    while (env) {
        lisp bind = car(env);
        // only works for symbol
        if (car(bind)==name) return bind; 
        // TODO: this is required for for example integer if BIGINT
        // if (eq(car(bind), name)) return bind;
        env = cdr(env);
    }
    return nil;
}

PRIM evallist(lisp e, lisp* envp) {
    if (!e) return e;

    lisp r = nil;
    lisp last = r;
    lisp ee = cons(symbol("*GC"), r);
    lisp safeenv = cons(ee, *envp);
    while (e) {
      setcdr(ee, r);
      lisp a = evalGC(car(e), &safeenv);
      lisp nw = cons(a, nil);
      if (!r) r = nw;
      setcdr(last, nw);
      last = nw;
      e = cdr(e);
    }
    return r;
}

static inline int tracep(lisp f);

// dummy function that doesn't eval, used instead of eval
static PRIM noEval(lisp x, lisp* envp) { return x; }

PRIM primapply(lisp ff, lisp args, lisp* envp, lisp all, int noeval) {
    //printf("PRIMAPPLY "); princ(ff); princ(args); terpri();
    int n = GETPRIMNUM(ff);
    lisp (*e)(lisp x, lisp* envp) = (noeval && n > 0) ? noEval : evalGC;
    int an = abs(n);

    int dotrace = tracep(ff);
    if (!dotrace) {
        // For optimization these special cases improve speed a lot!
        if (n == 2) { // eq/plus etc
            lisp (*fp)(lisp,lisp) = GETPRIMFUNC(ff);
            return (*fp)(e(car(args), envp), e(car(cdr(args)), envp)); // safe!
        }
        if (n == -3) { // if...
            lisp (*fp)(lisp*,lisp,lisp,lisp) = GETPRIMFUNC(ff);
            return (*fp)(envp, car(args), car(cdr(args)), car(cdr(cdr(args))));
        }
        if (n == -1) { // quote...
            lisp (*fp)(lisp*,lisp) = GETPRIMFUNC(ff);
            return (*fp)(envp, car(args));
        }
        if (n == 1) {
            lisp (*fp)(lisp) = GETPRIMFUNC(ff);
            return (*fp)(e(car(args), envp));
        }
        if (n == 3) {
            lisp (*fp)(lisp,lisp,lisp) = GETPRIMFUNC(ff);
            return (*fp)(e(car(args), envp), e(car(cdr(args)), envp), e(car(cdr(cdr(args))),envp));
        }
        if (n == -7) { // lambda, quite uncommon
            lisp (*fp)(lisp*,lisp,lisp) = GETPRIMFUNC(ff);
            return (*fp)(envp, args, all);
        }

        // don't do evalist, but allocate array, better for GC
        if (1 && an > 0 && an <= 6) {
            if (n < 0) an++; // add one for neval and initial env
            lisp argv[an];
            int i;
            for(i = 0; i < an; i++) {
                // if noeval, put env first
                if (i == 0 && n < 0) { 
                    argv[0] = (lisp)envp;
                    continue;
                }
                lisp a = car(args);
                if (a && n > 0) a = e(a, envp);
                argv[i] = a;
                args = cdr(args);
            }
            lisp (*fp)() = GETPRIMFUNC(ff);
            switch (an) {
            case 1: return fp(argv[0]);
            case 2: return fp(argv[0], argv[1]);
            case 3: return fp(argv[0], argv[1], argv[2]);
            case 4: return fp(argv[0], argv[1], argv[2], argv[3]);
            case 5: return fp(argv[0], argv[1], argv[2], argv[3], argv[4]);
            case 6: return fp(argv[0], argv[1], argv[2], argv[3], argv[4], argv[5]);
            }
        }
    }
    // above is all optimiziations
    // this is the old fallback solution, simple and works but expensive

    // prepare arguments
    if (n >= 0) {
        args = noeval ? args : evallist(args, envp);
    } else if (n > -7) { // -1 .. -7 no-eval lambda, put env first
        // TODO: for NLAMBDA this may not work...  may need a new lisp type
        args = cons(*envp, args);
    }

    if (dotrace) {
        indent(trace_level++); printf("---> "); princ(funame(ff));
        prin1(args); terpri();
    }

    lisp r;
    if (abs(n) == 7) {
        lisp (*fp)(lisp*, lisp, lisp) = GETPRIMFUNC(ff);
        r = fp(envp, args, all);
    } else {
        lisp a = args, b = cdr(a), c = cdr(b), d = cdr(c), e = cdr(d), f = cdr(e), g = cdr(f), h = cdr(g), i = cdr(h), j = cdr(i);
        // with C calling convention it's ok, but maybe not most efficient...
        lisp (*fp)(lisp, lisp, lisp, lisp, lisp, lisp, lisp, lisp, lisp, lisp) = GETPRIMFUNC(ff);
        r = fp(car(a), car(b), car(c), car(d), car(e), car(f), car(g), car(h), car(i), car(j));
    }

    if (dotrace) {
        indent(--trace_level); printf("<--- ");
        prin1(funame(ff)); printf(" ==> "); princ(r); terpri();
    }

    return r;
}

// TODO: not used??? this can be used to implement generators
static inline lisp mkthunk(lisp e, lisp env) {
    thunk* r = ALLOC(thunk);
    r->e = e;
    r->env = env;
    return (lisp)r;
}

// an immediate is a continuation returned that will be called by eval directly to yield a value
// this implements continuation based evaluation thus maybe allowing tail recursion...
// these are used to avoid stack growth on self/mutal recursion functions
// if, lambda, progn etc return these instead of calling eval on the tail
static inline lisp mkimmediate(lisp e, lisp env) {
    immediate* r = ALLOC(immediate); //(thunk*)mkthunk(e, env); // inherit from func_TAG
    r->e = e;
    r->env = env;
    return (lisp)r;
}

// these are formed by evaluating a lambda
PRIM mkfunc(lisp e, lisp env) {
    func* r = ALLOC(func);
    r->e = e;
    r->env = env;
    r->name = nil;
    return (lisp)r;
}

PRIM fundef(lisp f) {
    if (IS(f, func)) return ((func*)f)->e;
    return nil;
}

PRIM funenv(lisp f) {
    if (IS(f, func)) return ((func*)f)->env;
    return nil;
}

PRIM funame(lisp f) { // fun-name lol (6 char) => funame
    if (IS(f, func)) return ((func*)f)->name;
    if (IS(f, prim)) return *(lisp*)GETPRIM(f);
    return nil;
}

////////////////////////////// GC

#define FLASHP(x) ((unsigned int)next >= (unsigned int)flash_memory && ((unsigned int)next <= (unsigned int)(flash_memory + SPI_FLASH_SIZE_BYTES - FS_ADDRESS)))

void mark_deep(lisp next, int deep) {
    while (next) {
        // -- pointer to FLASH? no follow...
        if (FLASHP(next)) {
            printf("[mark.flash %x]\n", (unsigned int)next);
            return;
        }
        // -- pointer contains tag
        if (INTP(next)) return;
        if (SYMP(next)) return;
        if (PRIMP(next)) return;
	if (CONSP(next)) {
	    int i = (conss*)next - &conses[0];
            if (i >= MAX_CONS) { // pointing to other RAM/FLASH not allocated to cons array
                printf("mark.cons.badcons i=%d    %u max=%d\n", i, (int)next, MAX_CONS);
                exit(1);
            }
            if (CONS_IS_USED(i)) return; // already checked!
  	    CONS_SET_USED(i);
  	    mark_deep(car(next), deep+1);
	    next = cdr(next);
	    continue;
	}
        // -- generic pointer to heap allocated object with type tag
        // optimization, we store index position in element
        int index = next->index;
        if (index < 0) return; // no tracked here
        if (index >= MAX_ALLOCS) {
            printf("\n--- ERROR: mark_deep - corrupted data p=%u   index=%d\n", (unsigned int)next, index);
            printf("VALUE="); princ(next); terpri();
        }

        lisp p = allocs[index];
        if (!p || p != next) {
            printf("\n-- ERROR: mark_deep - index %d doesn't contain pointer. p=%u\n", index, (unsigned int)next);
            //printf("pppp = 0x%u and next = 0x%u \n", p, next);
            //princ(next);
            //return;
        } 
        
        if (IS_USED(index)) return;

        SET_USED(index);
        //printf("Marked %i deep %i :: p=%u ", index, deep, p); princ(p); terpri();
            
        int tag = TAG(p);
        if (tag == thunk_TAG || tag == immediate_TAG || tag == func_TAG) {
            mark_deep(ATTR(thunk, p, e), deep+1);
            next = ATTR(thunk, p, env);
        }
    }
}

inline void mark(lisp x) {
    mark_deep(x, 1);
}

///--------------------------------------------------------------------------------
// Primitives
// naming convention:
// -- lisp foo()   ==   arguments like normal lisp function can be called from c easy
// -- lsp foo_()   ==   same as lisp but name clashes with stanrdard function read_
// -- lsp _foo()   ==   special function, order may be different (like it take env& as first)

// first one liners
PRIM nullp(lisp a) { return a ? nil : t; }
PRIM consp(lisp a) { return IS(a, conss) ? t : nil; }
PRIM atomp(lisp a) { return IS(a, conss) ? nil : t; }
PRIM stringp(lisp a) { return IS(a, string) ? t : nil; }
PRIM symbolp(lisp a) { return IS(a, symboll) ? t : nil; } // rename struct symbol to symbol?
PRIM numberp(lisp a) { return IS(a, intint) ? t : nil; } // TODO: extend with float/real
PRIM integerp(lisp a) { return IS(a, intint) ? t : nil; }
PRIM funcp(lisp a) { return IS(a, func) || IS(a, thunk) || IS(a, prim) ? t : nil; }

PRIM iota(lisp count, lisp start, lisp step) {
    int c = getint(count);
    if (c <= 0) return nil;
    return cons(start?start:mkint(0),
                iota(mkint(c-1), mkint(getint(start) + (step?getint(step):1)), step));
}

PRIM plus(lisp *envp, lisp x) {
    int s = 0;
    while (x) {
        s += getint(evalGC(car(x), envp));
        x = cdr(x);
    }
    return mkint(s);
}
// because evallist is expensive, these can be quite expensive
// better not build up intermidate structure... thats why -7 (no eval)
PRIM times(lisp *envp, lisp x) {
    int p = 1;
    while (x) {
        p *= getint(evalGC(car(x), envp));
        x = cdr(x);
    }
    return mkint(p);
}

PRIM minus(lisp a, lisp b) { return b ? mkint(getint(a) - getint(b)) : mkint(-getint(a)); }
PRIM divide(lisp a, lisp b) { return mkint(getint(a) / getint(b)); }
PRIM mod(lisp a, lisp b) { return mkint(getint(a) % getint(b)); }
PRIM random_(lisp a, lisp b) {
  // randomize
  int ia = getint(a);
  if ((!a && !b) || ia < 0) {
    if (!a) ia = randomized() % 0xffffff;
    int seed = abs(ia);
    srand(seed);
    return mkint(-seed);
  }
  if (b) return mkint(rand() % (getint(b)-ia+1) + ia);
  if (!ia) return a;
  return mkint(rand() % ia);
}

// TODO: http://www.gnu.org/software/emacs/manual/html_node/elisp/Input-Functions.html#Input-Functions
// http://www.lispworks.com/documentation/HyperSpec/Body/f_rd_rd.htm#read
PRIM read_(lisp s) {
    if (s) {
        return reads(getstring(s));
    } else {
        char* str = readline(">", READLINE_MAXLEN);
        lisp r = str ? reads(str) : nil;
        if (str) free(str);
        return r;
    }
}

PRIM edit(lisp s, lisp title) {
    return mkstring(editor(getstring(s), getstring(title)));
}

// TODO: consider http://picolisp.com/wiki/?ArticleQuote
PRIM _quote(lisp* envp, lisp x) { return x; }
PRIM quote(lisp x) { return list(symbol("quote"), x, END); } // TODO: optimize
PRIM _env(lisp* e, lisp all) { return *e; }

// longer functions
PRIM eq(lisp a, lisp b) {
    if (a == b) return t;
    // TODO: revisit, if no values on the heap then just return false;
    char ta = TAG(a);
    char tb = TAG(b);
    if (ta != tb) return nil;
    // only int needs to be eq with other int even if on heap...
    if (ta != intint_TAG) return nil;
    if (getint(a) == getint(b)) return t;
    return nil;
}

// We implement a generlized <=> compare function for all types, including "list"
// this means we don't need to duplicate code for equal, <, string<, (=, <=, ==, >=, etc)
// -1 means <
//  0 means equal
// +1 means >
// -2 means tagtype < 
// +2 means tagtype >
// -3 means not cons/can't compare
int cmp(lisp a, lisp b) {
    while (a != b) {
        lisp r = eq(a, b);
        if (r) return 0;
        char taga = TAG(a), tagb = TAG(b);
        if (taga != tagb) return taga < tagb ? -2 : +2;
        if (taga == intint_TAG) return getint(a) < getint(b) ? -1 : a > b ? +1 : 0;
        if (taga == string_TAG) return strcmp(getstring(a), getstring(b));
        if (SYMP(a) && SYMP(b)) {
          char as[7] = {0}, bs[7] = {0}, *ap = NULL, *bp = NULL;
          ap = HSYMP(a) ? symbol_getString(a) : sym2str(a, as);
          bp = HSYMP(b) ? symbol_getString(b) : sym2str(b, bs);
          return strcmp(ap, bp);
        }          
        if (taga != conss_TAG) return -3;
        // cons, iterate
        int v = cmp(car(a), car(b));
        if (v) return v;
        a = cdr(a); b = cdr(b);
    }
    return 0;
}

PRIM cmp_(lisp a, lisp b) {
    return mkint(cmp(a, b));
}

PRIM equal(lisp a, lisp b) {
    return cmp(a, b) ? nil : t;
}

PRIM lt (lisp a, lisp b) { return cmp(a, b) <  0 ? t : nil; }
PRIM lte(lisp a, lisp b) { return cmp(a, b) <= 0 ? t : nil; }
PRIM gt (lisp a, lisp b) { return cmp(a, b) >  0 ? t : nil; }
PRIM gte(lisp a, lisp b) { return cmp(a, b) >= 0 ? t : nil; }

inline lisp getBind(lisp* envp, lisp name, int create) {
    //printf("GETBIND: envp=%u global_envp=%u ", (unsigned int)envp, (unsigned int)global_envp); princ(name); terpri();

    // if we're at top level create a global binding and entry in the hashtable
    if (create && envp && global_envp && *envp == *global_envp) return hashsym(name, NULL, 0, create);
    if (create) return nil;

    // first search local lexical env
    lisp bind = assoc(name, *envp);
    if (bind) return bind;

    // second search global env (stored in a hashtable)
    return hashsym(name, NULL, 0, 0); // not create, read only
}
// magic, this "instantiates" an inline function!
lisp getBind(lisp* envp, lisp name, int create);

// TODO: measure overhead! (compare to: f871de73834340edb3fa5b26d49e43e373647f9b)
// (time (fibo 34))
// - before: 6973
// - after but nothing in list: 7729
// - 3 items check make it 600 ms slower
// - 6 items -> 1300 ms slower...
//
// - this function adds maybe 10% overhead
// - if function return 0 directly still 5% overhead
// ==> other sourced of changes that slowed down... (like evallist)
static inline int tracep(lisp f) {
    static lisp vb = 0;
    if (!vb && global_envp) vb = getBind(global_envp, symbol("*TR"), 1);
    lisp x = cdr(vb);
    if (!vb || !x) return 0;
    lisp fn = funame(f);
    // inlined member 3 items 3000 ms overhead -> 600 ms!
    while (x) {
        if (fn == car(x)) return 1;
        x = cdr(x);
    }
    return 0;
}

// like setqq but returns binding, used by setXX
// 1. define, de - create binding in current environment
// 2. set! only modify existing binding otherwise give error
// 3. setq ??? (allow to define?)
inline lisp _setqqbind(lisp* envp, lisp name, lisp v, int create) {
    lisp bind = getBind(envp, name, create);
    //    printf("SETQBIND: found "); princ(name); putchar(' '); princ(bind); terpri();
    if (!bind) {
        //printf("SETQBIND: "); princ(name); putchar(' '); princ(bind); terpri();
        bind = cons(name, nil);
        *envp = cons(bind, *envp);
    }
    setcdr(bind, v);
    return bind;
}
// magic, this "instantiates" an inline function!
lisp _setqqbind(lisp* envp, lisp name, lisp v, int create);
    
inline PRIM _setqq(lisp* envp, lisp name, lisp v) {
    _setqqbind(envp, name, nil, 0);
    return v;
}
// magic, this "instantiates" an inline function!
PRIM _setqq(lisp* envp, lisp name, lisp v); 

// next line only needed because C99 can't get pointer to inlined function?
PRIM _setqq_(lisp* envp, lisp name, lisp v) { return _setqq(envp, name, v); }

inline PRIM _setbang(lisp* envp, lisp name, lisp v) {
    if (!symbolp(name)) { printf("set! of non symbol="); prin1(name); terpri(); error("set! of non atom: "); }
    v = eval(v, envp);
    lisp bind = _setqqbind(envp, name, nil, 0);
    // TODO: evalGC? probably safe as steqqbind changed an existing env
    // eval using our own named binding to enable recursion
    setcdr(bind, v);

    return v;
}

inline PRIM _set(lisp* envp, lisp name, lisp v) {
    // TODO: evalGC? probably safe as steqqbind changed an existing env
    return _setbang(envp, eval(name, envp), v);
}
// magic, this "instantiates" an inline function!
PRIM _set(lisp* envp, lisp name, lisp v);

PRIM de(lisp* envp, lisp namebody);

PRIM _define(lisp* envp, lisp args) {
    if (SYMP(car(args))) { // (define a 3)
        lisp name = car(args);

        // like _setq but with create == 1
        lisp bind = _setqqbind(envp, name, nil, 1);
        lisp r = eval(car(cdr(args)), envp);
        setcdr(bind, r);

        if (IS(r, func)) ((func*)r)->name = name;
        return r;
    } else { // (define (a x) 1 2 3)
        lisp name = car(car(args));
        lisp fargs = cdr(car(args));
        return de(envp, cons(name, cons(fargs, cdr(args))));
    }
}

PRIM de(lisp* envp, lisp namebody) {
    return _define(envp, cons(car(namebody), cons(cons(symbol("lambda"), cdr(namebody)), nil)));
}

lisp reduce_immediate(lisp x);

PRIM apply(lisp f, lisp args) {
    // TODO: for now, block GC as args could have been built out of thin air!
    blockGC++; 

    lisp e = nil; // dummy
    // TODO: like eval push on stack so can GC safely?
    lisp x = callfunc(f, args, &e, nil, 1);
    x = reduce_immediate(x);

    blockGC--;
    return x;
}

PRIM mapc(lisp f, lisp r) {
    while (r && consp(r) && funcp(f)) {
        apply(f, cons(car(r), nil));
        r = cdr(r);
    }
    return nil;
}

PRIM filter(lisp p, lisp l) {
    if (!l || !p) return l;
    lisp a = car(l); l = cdr(l);
    if (p && apply(p, cons(a, nil))) return cons(a, filter(p, l));
    return filter(p, l);
}
PRIM mapcar(lisp m, lisp l) {
    if (!l || !m) return l;
    lisp a = car(l); l = cdr(l);
    a = apply(m, cons(a, nil));
    return cons(a, mapcar(m, l));
}
PRIM reduce(lisp r, lisp l) {
    if (!l || !r) return l;
    lisp a = car(l); l = cdr(l);
    if (!l) return a;
    lisp b = car(l); l = cdr(l);
    lisp v = apply(r, cons(a, cons(b, nil)));
    return reduce(r, cons(v, l));
}

// efficent implementation of filtermapfilterreduce that doesn't build
// intermidiate lists...
// TODO: do it by merging, filter+mapc+filter+reduce into one function!
PRIM filtermapfilterreduce(lisp p, lisp m, lisp mp, lisp r, lisp l) {
    return reduce(r, filter(mp, mapcar(m, filter(p, l))));
}

PRIM length(lisp r) {
    if (IS(r, string)) return mkint(strlen(getstring(r)));
    if (!IS(r, conss)) return mkint(0);
    int c = 0;
    while (r) {
        c++;
        r = cdr(r);
    }
    return mkint(c);
}

// scheme string functions - https://www.gnu.org/software/guile/manual/html_node/Strings.html#Strings
// common lisp string functions - http://www.lispworks.com/documentation/HyperSpec/Body/f_stgeq_.htm
PRIM concat(lisp* envp, lisp x) {
    // calcualate len
    int len = 1;
    lisp i = x;
    while (i) {
        lisp v = car(i);
        if (INTP(v)) {
            int iv = getint(v);
            // negative
            if (iv < 0) len += 1;
            iv = abs(iv);
            // each digit
            while ((iv /= 10) > 0) len++; 
            // last digit
            len++; 
        } else {
            char* s = getstring(v);
            char ss[7] = {0};
            if (HSYMP(v)) s = symbol_getString(v);
            else if (SYMP(v)) s = sym2str(v, ss);
            else if (INTP(v)) 
            len += strlen(s);
        }
        i = cdr(i);
    }
    // build the string
    char* r = malloc(len), *p = r;
    *r = 0;
    i = x;
    while (i) {
        lisp v = car(i);
        if (INTP(v)) {
            p += snprintf(p, 20, "%d", getint(v));
        } else {
            char* s = getstring(v);
            char ss[7] = {0};
            if (HSYMP(v)) s = symbol_getString(v);
            else if (SYMP(v)) s = sym2str(v, ss);
            p = strcat(p, s);
        }
        i = cdr(i);
    }
    return mklenstring(r, -len);
}

PRIM char_(lisp i) {
    if (IS(i, string)) {
        return mkint(getstring(i)[0]);
    } else if (IS(i, intint)) {
        char s[2] = {0};
        s[0] = getint(i);
        return mkstring(s);
    } else 
        error("char.unsupported type");
    return nil;
}

// optional n, number of entries to return (1..x), 0, <0, nil, non-number => all
PRIM split(lisp s, lisp d, lisp n) {
    int i = getint(n);
    char* src = getstring(s);
    char* delim = getstring(d);
    int len = strlen(delim);
    lisp r = nil;
    lisp p = r;
    lisp last = nil;
    while (src && (i > 0 || i <= 0)) {
        char* where = strstr(src, delim);
        if (!where) {
            where = src + strlen(src);
            i = 1; // will terminate after add
        }
        lisp m = mklenstring(src, where - src);

        // add conscell at end and add value
        p = cons(nil, nil);
        if (!r)
            r = p;
        else
            setcdr(last, p);
        setcar(p, m);

        last = p;
        p = cdr(p);

        src = where + len;
        i--;
        if (!i) break;
    }
    return r;
}

///--------------------------------------------------------------------------------
// lisp reader

static char *input = NULL;
static char nextChar = 0;

static char nextx() {
    if (nextChar != 0) {
        char c = nextChar;
        nextChar = 0;
        return c;
    }
    if (!input) return 0;
    if (!*input) {
        input = NULL;
        return 0;
    }
    return *(input++);
}

static char next() {
    // for debuggin
    if (0) {
        char c = nextx();
        printf(" [next == %d] ", c);
        return c;
    } else {
        return nextx();
    }
}

static void skipSpace() {
    char c = next();
    while (c && isspace((int)c)) c = next();
    nextChar = c;
}

static int readInt(int v) {
    unsigned char c = next();
    while (c && isdigit(c)) {
        v = v*10 + c-'0';
        c = next();
    }
    nextChar = c;
    return v;
}

static lisp readString() {
    char* start = input;
    char c = next();
    while (c && c != '"') {
        if (c == '\\') c = next();
        c = next();
    }
    if (!c) error("string.not_terminated");
    int len = input - start - 1;

   // we modify the copied string as the string above may be constant
    lisp r = mklenstring(start, len);
    // remove '\', this may waste a byte or two if any
    char* from = getstring(r);
    char* to = from;
    while (*from) {
        // TODO: \n \t ...
        if (*from == '\\') from++;
        *to = *from; // !
        from++; to++;
    }
    *from = 0;
    return r;
}

static lisp readSymbol(char c, int o) {
    // TODO: cleanup, ugly
   if (!input) {
       char s[2] = {'-', 0};
       return symbol_len(s, 1);
    }
    char* start = input - 1 + o;
    int len = 0;
    while (c && c!='(' && c!=')' && !isspace((int)c) && c!='.') {
        len++;
	c = next();
    }
    nextChar = c;
    return symbol_len(start, len-o);
}

static lisp readx();

static lisp readList() {
    skipSpace();

    char c = next();
    if (!c) return nil;
    if (c == ')') return nil;
    if (c == '(') {
        lisp ax = readList();
        lisp dx = readList();
        return cons(ax, dx);
    }
    nextChar = c;

    lisp a = readx();
    skipSpace();
    c = next();
    lisp d = nil;
    if (c == '.') {
        d = readx();
        c = next();
        if (c != ')') error("Dotted pair expected ')'!");
        c = next();
    } else {
        nextChar = c;
        d = readList();
        c = next();
    }
    nextChar = c;
    return cons(a, d);
}

static lisp readx() {
    skipSpace();
    unsigned char c = next();
    if (!c) return NULL;
    if (c == '\'') return quote(readx());
    if (c == '(') return readList();
    if (c == ')') return nil;
    if (isdigit(c)) return mkint(readInt(c - '0'));
    if (c == '-') {
        unsigned char n = next();
        if (isdigit(n))
            return mkint(-readInt(n - '0'));
        else
            return readSymbol(n, -1);
    }
    if (c == '"') return readString();
    return readSymbol(c, 0);
}

PRIM reads(char *s) {
    input = s;
    nextChar = 0;
    return readx();
}

///////////////////////////////////////////////////////////////////////////////
// output, with capturing
typedef int putcf(int c);
putcf *writeputc, *origputc;

// TODO: more efficient...
//typedef int writef(char* s, int len);
//putcf *writewrite;


PRIM with_putc(lisp* envp, lisp args) {
    lisp fn = car(args);
    putcf *old = writeputc;

    int recurse = 0;
    int myputc(int c) {
        recurse++;
        if (recurse > 1) error("with-putc called with function that calls putc - prohibited!");
        lisp r= callfunc(fn, cons(mkint(c), nil), envp, nil, 1);
        recurse--;
        return getint(r);
    }

    lisp r = nil;
    writeputc = myputc; {
        // need catch errors and restore... (setjmp/error?)
        r = progn(envp, cdr(args));

    } writeputc = old;
    return r;
}

PRIM with_fd(lisp* envp, lisp args) {
    int fd = getint(eval(car(args), envp));
    putcf *old = writeputc;

    int myputc(int c) {
        char cc = c;
        return write(fd, &cc, 1);
    }

    lisp r = nil;
    writeputc = myputc; {
        // need catch errors and restore... (setjmp/error?)
        r = reduce_immediate(progn(envp, cdr(args)));
    } writeputc = old;
    return r;
}

// TODO: move out to an ide.c file? consider with ide-www

PRIM with_fd_json(lisp* envp, lisp args) {
    int fd = getint(eval(car(args), envp));
    putcf *old = writeputc;

    int myputc(int c) {
        char cc = c;
        origputc(c);
        // TODO: specific for jsonp?
        if (c == '\n' || c == '\r')
            return write(fd, "\\n", 2);
        else if (c == '"')
            return write(fd, "\\\"", 2);
        else if (c == '\'')
            return write(fd, "\\\'", 2);
        else
            return write(fd, &cc, 1);
    }

    lisp r = nil;
    writeputc = myputc; {
        // need catch errors and restore... (setjmp/error?)
        r = reduce_immediate(progn(envp, cdr(args)));
    } writeputc = old;
    return r;
}

int writec(int c) {
    if (!writeputc) origputc = writeputc = &putchar;
    return writeputc(c);
} 

int writes(char* s) {
    int n = 0;
    while (*s && ++n) writec(*s++);
    return n;
} 

int writef(const char* fmt, ...) {
    va_list args;
    va_start(args, fmt);
    int len;

    if (writeputc == &putchar) { // more efficient
        len = vprintf(fmt, args);
    } else {
        // capture output of printf by first faking it, just counting
        // then, write to buffer, then call writec...
        char* tst = NULL;
        len = vsnprintf(tst, 0, fmt, args) + 1;
        char out[len]; // TODO: this may not be safe...
        memset(out, 0, len);
        vsnprintf(out, len, fmt, args);
        writes(out);
    }

    va_end(args);
    return len;
}

#undef putchar
#undef printf
#undef puts

#define printf writef
#define putchar writec
#define puts writes

// TODO: prin1, princ, print, pprint/pp by calling
//    (write o output-stream= base= escape= level= circle= lines= pretty= readably=)
//
// http://www.gnu.org/software/emacs/manual/html_node/elisp/Output-Functions.html
// http://www.lispworks.com/documentation/HyperSpec/Body/f_wr_pr.htm
//
// (prin1 object output-stream)
//     ==  (write object :stream output-stream :escape t)
// (princ object output-stream)
//     ==  (write object stream output-stream :escape nil :readably nil)
// (print object output-stream)
//     ==  (progn (terpri output-stream)
//                (write object :stream output-stream :escape t)
//                (write-char #\space output-stream))
// (pprint object output-stream)
//     ==  (write object :stream output-stream :escape t :pretty t)
lisp princ_hlp(lisp x, int readable) {
//    printf(" [:: %u ::] ");
    int tag = TAG(x);
    // simple one liners
    if (!tag) printf("nil");
    else if (tag == intint_TAG) printf("%d", getint(x));
    else if (tag == prim_TAG) { putchar('#'); princ_hlp(*(lisp*)GETPRIM(x), readable); }
    // for now we have two "symbolls" one inline in pointer and another heap allocated
    else if (HSYMP(x)) writes(symbol_getString(x));
    else if (SYMP(x)) { char s[7] = {0}; sym2str(x, s); writes(s); }
    // can't be made reable really unless "reveal" internal pointer
    else if (tag == thunk_TAG) { printf("#thunk["); princ_hlp(ATTR(thunk, x, e), readable); putchar(']'); }
    else if (tag == immediate_TAG) { printf("#immediate["); princ_hlp(ATTR(thunk, x, e), readable); putchar(']'); }
    else if (tag == func_TAG) { putchar('#'); princ_hlp(ATTR(func, x, name), readable); }
    // string
    else if (tag == string_TAG) {
        if (readable) putchar('"');
        char* c = ATTR(string, x, p);
        while (*c) {
            if (readable && *c == '\"') putchar('\\');
            putchar(*c); //printf("[%d]", *c);
            c++;
        }
        if (readable) putchar('"');
    }
    // cons
    else if (tag == conss_TAG) {
        putchar('(');
        princ_hlp(car(x), readable);
        lisp d = cdr(x);
        while (d && gettag(d) == conss_TAG) {
            putchar(' ');
            princ_hlp(car(d), readable);
            d = cdr(d);
        }
        if (d) {
            putchar(' '); putchar('.'); putchar(' ');
            princ_hlp(d, readable);
        }
        putchar(')');
    } else {
        printf("*UnknownTag:%d;%u;%x*", tag, (unsigned int)x, (unsigned int)x);
    }
    // is need on esp, otherwise it's buffered and comes some at a time...
    // TODO: check performance implications
    fflush(stdout); 
    return x;
}

PRIM princ(lisp x) {
    return princ_hlp(x, 0);
}

// print in readable format
PRIM prin1(lisp x) {
    return princ_hlp(x, 1);
}

PRIM print(lisp x) {
    terpri();
    lisp r = prin1(x);
    putchar(' ');
    return r;
}

PRIM terpri() { putchar('\n'); return nil; }

// format conforms to - http://www.gnu.org/software/emacs/manual/html_node/elisp/Formatting-Strings.html
// TODO: make it return the numbers of characters printed?
// TODO: make a sprintf, or call it "format".
// which essentially is printf for lisp, they call it format in elisp
// TODO: format - http://www.gnu.org/software/mit-scheme/documentation/mit-scheme-ref/Format.html#Format
PRIM printf_(lisp *envp, lisp all) {
    char* f = getstring(car(all));
    all = cdr(all);
    while (*f) {
        if (*f == '%') {
            char fmt[16] = {0};
            fmt[14] = 1; fmt[15] = 1;
            char* p = &(fmt[0]);
            while (*f && !isalpha((int)*f) && !*p)
                *p++ = *f++;
            char type = *p++ = *f++;
            switch (type) {
            case '%':
                putchar('%'); break;
            case 's':
                princ(car(all)); break;
            case 'S':
                prin1(car(all)); break;
            case 'o': case 'd': case 'x': case 'X': case 'c':
                printf(fmt, getint(car(all))); break;
            case 'e': case 'f': case 'g': break;
                // printf(fmt, getfloat(car(x))); break;
            }
            if (type != '%')
                all = cdr(all);
        } else {
            putchar(*f++);
        }
    }
    return nil;
}

lisp pp_hlp(lisp e, int indent) {
    void nl() { terpri(); int i = indent*3; while(i--) putchar(' '); };
    void print_list(lisp e) {
        indent++;
        while (e) {
            nl();
            pp_hlp(car(e), indent+1);
            e = cdr(e);
        }
        putchar(')');
        indent--; nl();
    }

    if (!e) return prin1(e);
    if (!consp(e)) return prin1(e);
    lisp a1 = car(e), l2 = cdr(e),
        a2 = car(l2), l3 = cdr(l2),
        a3 = car(l3), l4 = cdr(l3),
        a4 = car(l4);
    if (funame(a1)) a1 = funame(a1); // decompile! haha
    if (symbolp(a1)) {
        if (a1 == symbol("if")) {
            printf("(if "); indent++; pp_hlp(a2, indent+1); nl();
            pp_hlp(a3, indent+2); indent--;
            if (l4) { nl(); pp_hlp(a4, indent+1); }
            putchar(')');
        //} else if (a1 == symbol("lambda")) {
        } else if (a1 == symbol("cond")) {
            printf("(cond ");
            print_list(l3);
        } else if (a1 == symbol("case")) {
            printf("(case "); pp_hlp(a2, indent+1);
            print_list(l3);
        } else if (a1 == symbol("define")) {
            prin1(a2);
            print_list(l3);
        } else if (a1 == symbol("de")) {
            printf("(de "); prin1(a2); putchar(' '); prin1(a3);
            print_list(l4);
        } else {
            printf("(");
            pp_hlp(a1, indent+1);
            lisp r = l2;
            while (r) {
                putchar(' ');
                pp_hlp(car(r), indent+1);
                r = cdr(r);
            }
            printf(")");
        }
    } else if (consp(a1)) { // list of list
        putchar('('); pp_hlp(a1, indent+2);
        print_list(l2);
    } else { // a list of some kind
        // TODO: try figure out complexity of list and if to do print_list on it...
        // use length and "estimated" chars for output (recursive length?)
        prin1(e); nl();
    }
    return nil;
}

PRIM pp(lisp e) {
    if (funcp(e))
        e = cons(symbol("de"), cons(funame(e), fundef(e)));
    pp_hlp(e, 0);
    return symbol("");
}

static lisp funcapply(lisp f, lisp args, lisp* envp, int noeval);

// get value of var, or complain if not defined
static inline lisp getvar(lisp e, lisp env) {
    lisp v = getBind(&env, e, 0);
    if (v) return cdr(v);

    printf("\n-- ERROR: Undefined symbol: "); princ(e); terpri();
    error("%% Undefined symbol");
    return nil;
}

// don't call directly, call evalGC() or eval()
static inline lisp eval_hlp(lisp e, lisp* envp) {
    if (!e) return e;
    char tag = TAG(e);
    if (tag == symboll_TAG) return getvar(e, *envp);
    if (tag != conss_TAG) return e;

    // find function
    lisp orig = car(e);
    lisp f = orig;
    tag = TAG(f);
    lisp last = nil; // if eval to exactly same value return, can still loop
    while (f && f!=last && tag!=prim_TAG && tag!=thunk_TAG && tag!=func_TAG) {
        last = f;
        f = evalGC(f, envp);
        tag = TAG(f);
    }

    // This may return a immediate, this allows tail recursion evalGC will reduce it.
    lisp r = callfunc(f, cdr(e), envp, e, 0);

    // we replace it after as no error was generated...
    if (f != orig) {
        // "macro expansion" lol (replace with implementation)
        // TODO: not safe if found through variable (like all!)
        // TODO: keep on symbol ptr to primitive function/global, also not good?
        // DEFINE(F,...) will then break many local passed variables
        // maybe must search all list till find null, then can look on symbol :-(
        // but that's everytime? actually, not it's a lexical scope!
        // TODO: only replace if not found in ENV and is on an SYMBOL!
        if (SYMP(orig) && eq(funame(f), orig))
            setcar(e, f);
    }

    return r;
}

// inline this is essential to not have stack grow!
inline lisp reduce_immediate(lisp x) {
    while (x && IS(x, immediate)) {
        lisp tofree = x;

        if (trace > 0) // make it visible
            x = evalGC(ATTR(thunk, x, e), &ATTR(thunk, x, env));
        else
            x = eval_hlp(ATTR(thunk, x, e), &ATTR(thunk, x, env));

        // immediates are immediately consumed after evaluation, so they can be free:d directly
        tofree->tag = 0;
        sfree((void*)tofree, sizeof(thunk), immediate_TAG);
        used_count--;
    }
    return x;
}

static void mymark(lisp x) {
    if (dogc) mark(x);
}

static void mygc() {
    if (dogc) gc(NULL);
}

// this is a safe eval to call from anywhere, it will not GC
// but it may blow up the stack, or the heap!!!
lisp eval(lisp e, lisp* envp) {
    blockGC++;
    lisp r = evalGC(e, envp);
    blockGC--;
    return r;
}

PRIM _eval(lisp e, lisp env) {
    // taking pointer creates a new "scrope"
    // if no env given, use the global env (so define will be on top level)
    return evalGC(e, env ? &env : global_envp);
}

lisp mem_usage(int count) {
    // TODO: last number conses not correct new useage
    if (traceGC) printf(" [GC freed %d used=%d bytes=%d conses=%d]\n", count, used_count, used_bytes, MAX_CONS - cons_count);
    return nil;
}

inline int needGC() {
    if (cons_count < MAX_CONS * 0.2) return 1;
    return (allocs_count < MAX_ALLOCS * 0.8) ? 0 : 1;
}
// magic, this "instantiates" an inline function!
int needGC();


// (de rec (n) (print n) (rec (+ n 1)) nil) // not tail recursive!
// === optimized:
// ...
// #767 0x0804d8b3 in evalGC ()
//
// #768 0x08051818 in progn ()
// #770 0x0804d8b3 in evalGC ()
//
// #771 0x08051e67 in run ()
// #772 0x08052d58 in readeval ()
// #773 0x08048b57 in main ()

#define MAX_STACK 80

static struct stack {
    lisp e;
    lisp* envp;
} stack[MAX_STACK];

// TODO: because of tail call optimization, we can't tell where the error occurred as it's not relevant on the stack???
PRIM print_detailed_stack() {
    int l;
    // TODO: DONE but too much: using fargs of f can use .envp to print actual arguments!
    for(l = 0; l < level + 5; l++) {
        if (!stack[l].e && !stack[l].envp) break;

        if (!l) terpri();
        printf("%4d : ", l);
        prin1(stack[l].e); printf(" ==> ");

        lisp f = car(stack[l].e);
        lisp* envp = stack[l].envp;
        lisp env = envp ? *envp : nil; // env before
        while (f && IS(f, symboll) && !IS(f, func) && !IS(f, thunk) && !IS(f, prim) && !IS(f, immediate)) {
            f = getvar(f, env); // eval?
        }

        putchar('['); princ(f);
        if (f && IS(f, func)) {
            // get env AFTER it has been extended... (using next expression)
            lisp *nenvp = (l+1 < MAX_STACK) ? stack[l+1].envp : NULL;
            lisp nenv = nenvp ? *nenvp : nil;

            if (nenv) {
                lisp def = ATTR(thunk, f, e); // get definition
                lisp fargs = car(def);
                //printf("\nFARGS="); princ(fargs); printf("  ENV="); princ(nenv); terpri();
                while (fargs && nenv) {
                    putchar(' ');
                    prin1(car(car(nenv))); putchar('='); prin1(cdr(car(nenv)));
                    env = cdr(nenv);
                    fargs = cdr(fargs);
                }
            }
            putchar(']');
        } else if (f && IS(f, prim)) {
            printf(" ... ] ");
        } else {
            printf(" ... ] ");
            printf(" ...car did not evaluate to a function... (it's an %s)\n", f ? tag_name[TAG(f)] : "nil");
            break;
        }
        terpri();
    }
    return nil;
}

void print_stack() {
    int l;
    lisp last = nil;
    int count = 0;
    // TODO: DONE but too much: using fargs of f can use .envp to print actual arguments!
    for(l = 0; l < level; l++) {
        if (!stack[l].e && !stack[l].envp) break;
        if (!l) printf(" @ ");
        lisp f = car(stack[l].e);
        if (f == last) {
            count++;
            continue;
        }

        if (count > 1) {
            printf("%d ", count);
            count = 0;
        }
        if (last) {
            princ(last);
            printf(" -> ");
        }

        last = f;
    }
    if (count > 1) printf("%d ", count);
    if (last) princ(last);
}

// prints env and stops at (nil . nil) binding (hides "globals")
void print_env(lisp env) {
    indent(level+1);
    printf(" ENV ");
    lisp xx = env;
    while (xx && car(car(xx))) {
        lisp b = car(xx);
	princ(car(b)); putchar('='); princ(cdr(b)); putchar(' ');
	xx = cdr(xx);
    }
    terpri();
}

// prints env and stops at (nil . nil) binding (hides "globals")
void print_args(lisp env, lisp f) {
    printf(" (");
    princ(funame(f));
    lisp fargs = car(fundef(f));
    // find varargs name
    lisp varargs = fargs;
    while (varargs && !symbolp(varargs)) varargs = cdr(varargs);

    lisp xx = env;
    while (xx && car(car(xx))) {
        lisp b = car(xx);
        lisp name = car(b);
        if ((name && member(name, fargs)) || name == varargs) {
            putchar(' ');
            princ(name); putchar('='); princ(cdr(b));
        }
	xx = cdr(xx);
    }
    printf(") ");
}

PRIM evalGC(lisp e, lisp* envp) {
    if (!e) return e;
    char tag = TAG(e);
    // look up variable
    if (tag == symboll_TAG) return getvar(e, *envp); 
    if (tag != symboll_TAG && tag != conss_TAG && tag != thunk_TAG) return e;

    if (level >= MAX_STACK) { error("Stack blowup!"); exit(3); }

    stack[level].e = e;
    stack[level].envp = envp;

    // TODO: move this to function
    if (!blockGC && needGC()) {
        mymark(*envp);
        if (trace > 2) printf("%d STACK: ", level);
        int i;
        for(i=0; i<64; i++) {
            if (!stack[i].e) break;
            mymark(stack[i].e);
            mymark(*stack[i].envp);
            if (trace > 2) {
                printf(" %d: ", i);
                princ(stack[i].e);
            }
        }
        if (trace > 2) terpri();
        mygc();
        // TODO: address growth?
        if (needGC()) {
            printf("\n[We GC:ed but after GC we need another GC - expect slowdowns!!!]\n");
        }
        // check ctlr-t and maybe at queue (GC issue needs resolve first)
        kbhit();
    }

    if (trace > 0) { indent(level); printf("---> "); princ(e); }
    level++;
    //if (trace) { indent(level+1); printf(" ENV= "); princ(env); terpri(); }
    if (trace > 0) { print_args(*envp, car(e)); terpri(); }

    lisp r = eval_hlp(e, envp);
    r = reduce_immediate(r);

    --level;
    if (trace > 0) { indent(level); princ(r); printf(" <--- "); princ(e); }
    if (trace > 0) { print_args(*envp, car(e)); terpri(); terpri(); }

    stack[level].e = nil;
    stack[level].envp = NULL;
    return r;
}

PRIM if_(lisp* envp, lisp exp, lisp thn, lisp els) {
    // evalGC is safe here as we don't construct any structes, yet
    // TODO: how did we get here? primapply does call evallist thus created something...
    // but we pass ENV on so it should be safe..., it'll mark it!
    return evalGC(exp, envp) ? mkimmediate(thn, *envp) : mkimmediate(els, *envp);
}

PRIM cond(lisp* envp, lisp all) {
    while (all) {
        lisp nxt = car(all);
        lisp e = car(nxt);
        e = evalGC(e, envp);
        lisp thn = cdr(nxt);
        if (e) return thn ? progn(envp, thn) : e;
        all = cdr(all);
    }
    return nil;
}

PRIM case_(lisp* envp, lisp all) {
    lisp x = evalGC(car(all), envp);
    all = cdr(all);
    while (all) {
        lisp ths = car(all);
        lisp m = member(x, car(ths));
        if (m) return progn(envp, cdr(ths));
        all = cdr(all);
    }
    return nil;
}

PRIM and(lisp* envp, lisp all) {
    lisp r = nil;
    while(all) {
        // TODO: tail call on last?
        r = evalGC(car(all), envp);
        if (!r) return nil;
        all = cdr(all);
    }
    return r;
}

PRIM or(lisp* envp, lisp all) {
    lisp r = nil;
    while(all) {
        // TODO: tail call on last?
        r = evalGC(car(all), envp);
        if (r) return r;
        all = cdr(all);
    }
    return nil;
}

PRIM not(lisp x) {
    return x ? nil : t;
}

// essentially this is a quote but it stores the environment so it's a closure!
PRIM lambda(lisp* envp, lisp all) {
    return mkfunc(all, *envp);
}

PRIM nlambda(lisp* envp, lisp all) {
    return mkfunc(all, nil);
}

PRIM progn(lisp* envp, lisp all) {
    while (all && cdr(all)) {
        evalGC(car(all), envp);
        all = cdr(all);
    }
    // only last form needs be tail recursive..., or if have "return"?
    return mkimmediate(car(all), *envp);
}

static inline lisp letevallist(lisp args, lisp* envp, lisp extend);
static inline lisp letstarevallist(lisp args, lisp* envp, lisp extend);

PRIM let(lisp* envp, lisp all) {
    lisp lenv = letevallist(car(all), envp, *envp);
    return progn(&lenv, cdr(all));
}

PRIM let_star(lisp* envp, lisp all) {
    lisp lenv = letstarevallist(car(all), envp, *envp);
    return progn(&lenv, cdr(all));
}

// use bindEvalList unless NLAMBDA
static inline lisp bindList(lisp fargs, lisp args, lisp env) {
    // TODO: not recurse!
    if (!fargs) return env;
    if (SYMP(args)) return cons(cons(fargs, args), env);
    lisp b = cons(car(fargs), car(args));
    // self tail recursion is "goto" - efficient
    return bindList(cdr(fargs), cdr(args), cons(b, env));
}

static inline lisp bindEvalList(lisp fargs, lisp args, lisp* envp, lisp extend) {
    while (fargs) {
        if (SYMP(fargs)) return cons(cons(fargs, evallist(args, envp)), extend);
        // This eval cannot be allowed to GC! (since it's part of building a cons structure
        // TODO: protect ala evallist
        lisp b = cons(car(fargs), eval(car(args), envp));
        extend = cons(b, extend);
        fargs = cdr(fargs);
        args = cdr(args);
    }
    return extend;
}

static inline lisp letevallist(lisp args, lisp* envp, lisp extend) {
    while (args) {
        lisp one = car(args);
        // This eval cannot be allowed to GC! (since it's part of building a cons structure
        // TODO: protect ala evallist
        lisp r = eval(car(cdr(one)), envp);
        extend = cons(cons(car(one), r), extend);
        args = cdr(args);
    }
    return extend;
}

static inline lisp letstarevallist(lisp args, lisp* envp, lisp extend) {
    while (args) {
        lisp one = car(args);
        // This eval cannot be allowed to GC! (since it's part of building a cons structure
        // TODO: protect ala evallist
        lisp r = eval(car(cdr(one)), &extend);
        extend = cons(cons(car(one), r), extend);
        args = cdr(args);
    }
    return extend;
}

static inline lisp funcapply(lisp f, lisp args, lisp* envp, int noeval) {
    lisp lenv = ATTR(thunk, f, env);
    lisp l = ATTR(thunk, f, e);
    //printf("FUNCAPPLY:"); princ(f); printf(" body="); princ(l); printf(" args="); princ(args); printf(" env="); princ(lenv); terpri();
    lisp fargs = car(l);

    if (!lenv) { // !lenv) {
        //printf("[funcapply NLAMBDA: "); prin1(f); putchar(' '); prin1(args);
        //printf(" LENV="); prin1(lenv);
        //printf(" ENV="); prin1(*envp);
        //printf("]\n");
        noeval = 1;
        // put env of caller in front, similary to PRIM -7 function
        args = cons(*envp, args);
    }

    // TODO: check if NLAMBDA!
    lenv = noeval ? bindList(fargs, args, lenv) : bindEvalList(fargs, args, envp, lenv);
    //printf("NEWENV: "); princ(lenv); terpri();

    int dotrace = tracep(f);
    if (dotrace) {
        indent(trace_level++); printf("--->"); print_args(lenv, f); terpri();
    }

    if (!dotrace) return progn(&lenv, cdr(l)); // tail recurse on rest

    lisp r = reduce_immediate(progn(&lenv, cdr(l)));

    if (dotrace) {
        indent(--trace_level); printf("<--- ");
        prin1(funame(f)); printf(" ==> "); princ(r); terpri();
    }
    return r;
}

// TODO: prim apply/funcapply may return immediate... (so users should call apply instead)
static inline lisp callfunc(lisp f, lisp args, lisp* envp, lisp e, int noeval) {
    int tag = TAG(f);
    if (tag == prim_TAG) return primapply(f, args, envp, e, noeval);
    if (tag == func_TAG) return funcapply(f, args, envp, noeval);
    if (tag == thunk_TAG) return f; // ignore args

    printf("%% "); princ(f); printf(" did not evaluate to a function in: "); princ(e ? e : cons(f, args));
    printf(" evaluated to "); princ(cons(f, args)); terpri();
    error("Not a function");
    return nil;
}

static PRIM test(lisp*);

// ticks are counted up in idle() function, as well as this one, they are semi-unique per run
static long lisp_ticks = 0;
PRIM ticks() { return mkint(lisp_ticks++ & 0xffffffff); } // TODO: mklong?

PRIM clock_() { return mkint(clock_ms()); }

PRIM time_(lisp* envp, lisp exp) {
    int start = clock_ms();
    lisp ret = evalGC(exp, envp);
    int ms = clock_ms() - start;
    return cons(mkint(ms), ret);
}

PRIM delay(lisp ms) {
  return mkint(delay_ms(getint(ms)));
}

// TODO: time functions... http://naggum.no/lugm-time.html

// different verbosity levels are differently chatty
// 0 = only results of actions like print (it doesn't suppress prints)
// 1 = 0 + print file header, echo result of each expression
// 2 = 1 + print expression, arrow, result
// 3 = 2 + print header with line number before each chunk. Good for debugging

PRIM load(lisp* envp, lisp name, lisp verbosity) {
    lisp* savedenvp = global_envp;
    global_envp = envp;

    char* filename = getstring(evalGC(name, envp));
    verbosity = evalGC(verbosity, envp);
    int v = getint(verbosity);
    if (v > 0) printf("\n========================= %s\n", filename);

    // no gcc style innner functions with outer variables.. .:-(
    int evalIt(void* p, char* s, char* filename, int startno, int endno, int v) {
        if (!s || !s[0] || s[0] == ';') return;
        if (v > 1) printf("\n========================= %s :%d-%d>\n%s\n", filename, startno, endno, s);
        lisp* envp = p;
        jmp_buf saved;
        memcpy(&saved, &lisp_break, sizeof(saved));
        if (setjmp(lisp_break) == 0) {
            lisp e = reads(s);
            lisp r = evalGC(e, envp);
            if (v > 0) {
                prin1(r);
                terpri();
            }
            memcpy(&lisp_break, &saved, sizeof(saved));
        } else {
            fprintf(stderr, "\n%%======ERROR IN SCRIPT/LOAD====== %s :%d-%d>\n%s", filename, startno, endno, s);
            memcpy(&lisp_break, &saved, sizeof(saved));
            return -1;
        }
        return 0;
    }

    int r = process_file(envp, filename, evalIt, v);
    global_envp = savedenvp;

    if (v > 0 && !r) printf("\n==========DONE=========== %s\n", filename);
    if (r < 0) error("%%failure running script, aborted...");

    return r ? nil : name;
}

PRIM cat(lisp fn) {
    char* filename = getstring(fn);
    FILE* f = fopen(filename, "r");
    if (!f) {
        perror(filename);
        return nil;
    }

    char line[257];
    do {
        if (!fgets(line, sizeof(line)-1, f)) break;
        puts(line);
    } while (1);

    fclose(f);
    return t;
}

PRIM dir(lisp name) {
    char* nm = getstring(name);
    if (!name || !nm || !*nm) nm = "./";

    DIR *dp;
    struct dirent *ep;

    dp = opendir(nm);
    if (!dp) {
        perror("Couldn't open directory");
        return nil;
    }

    lisp r = nil;
    while ((ep = readdir(dp)))
        r = cons(mkstring(ep->d_name), r);
    (void)closedir(dp);
    return r;
}

PRIM at(lisp* envp, lisp spec, lisp f) {
    int c = clock_ms();
    int w = getint(spec);
    lisp r = cons(mkint(c + abs(w)), cons(spec, evalGC(f, envp)));
    lisp bind = getBind(envp, ATSYMBOL, 0);
    // TOOD: insert sort should be easy, only problem is the first
    // so, we could prefix by atom QUEUE.
    setcdr(bind, cons(r, cdr(bind)));
    return r;
}

// we allow stopping original scheduled event, or any repeated
PRIM stop(lisp* envp, lisp at) {
    lisp att = evalGC(at, envp);
    lisp bind = getBind(envp, ATSYMBOL, 0);
    lisp lst = cdr(bind);
    lisp prev = bind;
    while (lst) {
        lisp entry = car(lst);
        if (entry == att || cdr(entry) == cdr(att)) {
            setcdr(prev, cdr(lst)); // remove
            return symbol("*stop*");
        }
        prev = lst;
        lst = cdr(lst);
    }
    return nil;
}

PRIM atrun(lisp* envp) {
    lisp prev = getBind(envp, ATSYMBOL, 0);
    lisp lst = cdr(prev);
    // TODO: sort? now we're checking all tasks all the time.
    // for now don't care as we do it as idle.
    while (lst) {
        int c = clock_ms();
        lisp entry = car(lst);
        int when = getint(car(entry));
        if (when && when < c) {
            int spec = getint(car(cdr(entry)));
            lisp exp = cdr(cdr(entry));
            //printf("[ @%d :: ", when, c); princ(exp); printf(" ");
            //lisp ret =
            apply(exp, nil);
            // TODO: add a way by special return value to unschedule itself if its a repeating task
            //printf(" => "); princ(ret); printf(" ]\n");
            if (spec > 0)
                setcdr(prev, cdr(lst)); // remove
            else
                setcar(entry, mkint(c + abs(spec)));
        }
        prev = lst;
        lst = cdr(lst);
        //if (!lst) terpri();
    }
    return nil;
}

//lisp imacs_(lisp name) {
//    return mkint(imacs_main(0, NULL));
//}

#ifdef UNIX
int xPortGetFreeHeapSize() { return -1; }
#endif

// test stack usage on simple function
// a simple function call takes 32 bytes (8 stack entry as reported by uxTaskGetStackHighWaterMark)
// This means we can recurse about 233 times!
//extern unsigned long uxTaskGetStackHighWaterMark(void*);

PRIM rec(lisp i) {
    int a = getint(i);
//    printf("%d - %u ::: stackUsed=%lu\n", a, (unsigned int)&a, uxTaskGetStackHighWaterMark(NULL));
    return mkint(1 + getint(rec(mkint(a - 1))));
}

// test function: eat the heap
PRIM heap() {
    void* a[60];
    int c = 0;

    int t = 0;
    int z = 16384;
    // TODO: keep allocated in free list and deallocate?
    while (z > 0) {
        //while (xPortGetFreeHeapSize() < z) z = z/2; // enable to make allocation overflow "safer"
        //void* p = pvPortMalloc(z);
        void* p = malloc(z);
        if (!p) {
            z = z/2;
            continue;
        }
        printf("%u %d %d %d\n", (unsigned int)p, z, t, xPortGetFreeHeapSize());
        t += z;
        // test is writable
        int* x = p;
        *x = 4711;
        if (*x != 4711) printf("--- Write failed! %d\n", *x);
        a[c++] = p;
        if (c == 60) break; // before it crashes!
    }
    int i;
    for (i = 0; i < c; i++) {
        free(a[i]);
        printf("%u %d\n", (unsigned int)a[i], xPortGetFreeHeapSize());
    }
    return mkint(t);
}


////////////////////////////////////////////////////////////////////////////////
// flash file access
// 
// - https://blog.cesanta.com/esp8266_using_flash
// ~/GIT/Espruino-on-ESP8266/user/user_main.c 

// TODO: use header
// "FF FF 1A 5H" --- 'FFFFLASH' // GOOD
// "BA DF 1A 5H" --- 'BADFLASH'
// "DE 1F 1A 5H" --- 'DELFLASH'
// "E0 FF 1A 5H" --- 'EOFFLASH'

// record:
// 'FF FF FF FF' - free to write
// 'yy cc xx xx' - yy != 00/ff, active len xx xx, of type yy = 1..254
// '00 cc xx xx' - deleted len xx xx
//
// cc: CRC?

// type yy:
// 0 - (deleted)
// ff - (free)
// bit 7: in USE
// bit 6: printable C string(s) (0=binary)
// bit 5: lisp (serialized string, or binary)
// bit 4: key/value
// bit 0-3: 16 sub types app defined for content

#ifdef UNIX

// Highlevel

// (save '(http boot) (lambda (x) (init-web-server)) (lambda (x) (print "Done")))
//lisp save(lisp key, lisp data, lisp cb) {
//}

// tail-recurses on: cb(key, info, cb)
//lisp dir(lisp matchkey, lisp cb) {
//}

// tail-recurses on: cb(key, info, data, cb, end?)
//lisp load(lisp matchkey, lisp cb) {
//}

#else

// how lua can compile functions to flash file and then call it
// - http://www.esp8266.com/viewtopic.php?f=19&t=1940

#include <stdint.h>
#include <stdbool.h>
#include <stdlib.h>
#include <stdio.h>
#include <string.h>

#include <espressif/spi_flash.h>
#include <espressif/esp_system.h>

#include <FreeRTOS.h>
#include <task.h>

// Wrote 41984 bytes at 0x00000000 in 4.1 seconds (81.6 kbit/s)...
// Wrote 211968 bytes at 0x00020000 in 20.9 seconds (81.0 kbit/s)...
// 
// spi (0-...) address:
// 0x00000000 42 KB
// 0x00020000 (128 KB) 211 KB start ---
// 0x0003C000 ESP private area? (16KB)
// 0x00054C00 end --- 211 KB
// 0x00100000 MINE?
// 0x00400000 end 4MB

// memory mapped range:
// 0x40200000
// 0x40600000 4MByte

/* (swrite "myfile" (list 1 2 3)) */
/* (swrite "abc" "foobar") */
/* (swrite "myfile" 1235) */

/* (sreset "myfile") -> */
/* (sread "myfile") -> (1 2 3) */
/* (sread "myfile") -> 12345 */

/* FLASH: */
/* ("myfile" . (1 2 3))\0 */
/* ("abc" . "foobar")\0 */
/* ("myfile" . 1235)\0 */

/* (point x y c) */
/* (line x y x y c) */
/* (circle ....) */
/* (text ...) */

/* (zap) */
/* (ping) */

#endif

int writeBytesToFlash(char* code, int len, int offset) {
    if (offset % 4 !=0) {
        printf("Illegal offset %d!", offset);
        error("Illegal offset");
        return -1;
    }
    int addr = FS_ADDRESS + offset;
    int sector = addr/SPI_FLASH_SEC_SIZE;
    int to = addr + len;
    int error;
    while (addr < to) {
        // only erase if start write at boundary, otherwise erase already written stuff
        // it will break at next sector and while loop back here and erase next one
        if (offset % SPI_FLASH_SEC_SIZE == 0) sdk_spi_flash_erase_sector(sector);
        char buff[SPI_FLASH_SEC_SIZE] = {0xff};
        int sz = strlen(code) + 1;
        if (sz + offset > SPI_FLASH_SEC_SIZE) {
            sz = SPI_FLASH_SEC_SIZE - offset;
            offset -= SPI_FLASH_SEC_SIZE;
        }
        memcpy(&buff[0], code, sz);
        sz = (sz + 3) & ~3; // round up to 4byte boundary
        if (SPI_FLASH_RESULT_OK != (error = sdk_spi_flash_write(addr, (uint32 *)buff, sz))) {
            printf("\nwriteBytesToFlash error %d\n", error);
        }
        addr += sz;
        code += sz;
        sector++;
    }

    // write and overwritable end marker
//    char c = 0xff;
//    if (SPI_FLASH_RESULT_OK != (error = sdk_spi_flash_write(addr + offset, (uint32 *)&c, 1))) {
//        printf("\nwriteBytesToFlash error %d\n", error);
//    }

    return to - FS_ADDRESS;
}

// TODO: connect to lisp writer?
int writeStringToFlash(char* code, int offset) {
    int len = strlen(code) + 1;
    return writeBytesToFlash(code, len, offset);
}

// TODO: connect to lisp reader!
// if called with NULL just count otherwise write
// returns len in bytes, to read next, round offset up to 4b boundary and call again
// see findLastFlash()

// readFromFlash(NULL, 0, offset) --> count of bytes including 0 at end
// readFromFlash(buff, maxbytes, offset) --> read into buff, length including 0 at end
// readFromFlash(buff, -bytes, offset) --> read bytes into buff, allow 0 // TODO: use sdk_spi_flash_read directly?
// returns -1 if in zeroterm mode and read 0xff
int readFromFlash(char* buff, int maxlen, int offset) {
    int zeroterm = (maxlen >= 0 || !buff);
    maxlen = maxlen >= 0 ? maxlen : -maxlen;

    int error;
    unsigned char c;
    int i;
    for (i = 0; (i < maxlen) || !buff; i++) {
        // TODO: can it really read one char at a time?
        // suspect to cast to uint32 pointer for &(char c)
        if (SPI_FLASH_RESULT_OK != (error = sdk_spi_flash_read(FS_ADDRESS + i + offset, (uint32 *)&c, 1))) {
            printf("\nreadFromFlash.error %d\n", error);
            break;
        }

        // if zeroterm mode end at 0, or at ff, as that's unwritten (?)
        if (zeroterm) {
            if (i == 0 && c == 0xff) return -1;
        }

        if (buff) buff[i] = c;
        if (zeroterm && !c) break;

        if (!buff && i > 100) { // TODO: change
            printf("...enough...!\n");
            return -1;
        }
                                
    }
    if (zeroterm) {
        // make sure zero terminated
        if (buff) buff[maxlen-1] = 0;
        return i + 1;
    } else {
        return i;
    }
}

int findLastFlash() {
    int offset = 0;
    int len;
    while (1) {
        len = readFromFlash(NULL, 0, offset);
        if (len < 0) break;
        offset += (len + 3) & ~3;
    }
    return offset;
}

PRIM scan(lisp s) {
    int maxlen =  SPI_FLASH_SIZE_BYTES - FS_ADDRESS;
    int i;

    if (1) {
//      // includes program ROM
//      uint32* a = (uint32*)0x40200000;
//      a += FS_ADDRESS/4;

        uint32* a = (uint32*)flash_memory;
        int s = 0;
        int c0 = 0;
        int cffffffff = 0;
        //int stop = 0;
        for(i = 0; i < maxlen/4; i++) {
            uint32 v = *a;
            a++;
            //if ((i % (16*1024/4) == 0)) { putchar('.'); fflush(stdout); }
            s += v;
            if (v == 0xffffffff) cffffffff++;
            if (v == 0) c0++;
            //while (v && !stop) {
            if (v) {
                int j;
                for(j = 4; j; j--) {
                    char c = v & 0xff;
                    if (c >= 32 && c <= 125) { putchar(v & 0xff); fflush(stdout); }
                    else if (1) { printf("[%d]", (unsigned char)c); fflush(stdout); }
                    // if (v && 0xff == 0xff) stop = 1;
                    v >>= 8;
                }
            }
            //putchar('/');
        }
        printf("\n");
        return cons(mkint(c0), mkint(cffffffff));
    }

    int error;
    char c;
    int c0 = 0;
    int cff = 0;
    for (i = 0; i < maxlen; i++) {
        // TODO: can it really read one char at a time?
        // suspect to cast to uint32 pointer for &(char c), probably works because alignment allows overwrite
        if (SPI_FLASH_RESULT_OK != (error = sdk_spi_flash_read(FS_ADDRESS + i, (uint32 *)&c, 1))) {
            printf("\nerror %d\n", error);
            break;
        }
        if ((i % (16*1024) == 0)) { putchar('.'); fflush(stdout); }

        if (c == 0) c0++;
        if (c == 0xff) cff++;
    }
    return cons(mkint(c0), mkint(cff));
}

// returns lisp pointer into buffer
lisp serializeLisp(lisp x, lisp* buffer, int *n) {
    printf("\n------ Serializelisp "); print(x);
    if (*n <= 2) return symbol("*FULL*");
    //if (HSYMP(x)) {
        // for now just "pray" - collisions in english language are 190/99K!
        // TODO: serialize by putting first HSYM then string directly after...
        // maybe make all symbol strings in linked list as local symbol table
        // This will be known for flash memory, maybe need bit to indicate?
    //}

    // self representive inside pointer
    if (!x || INTP(x) || SYMP(x)) return x;

    if (IS(x, string)) {
        // string is simple, just serialize a "heap" object with, pointer (to next cells)
        int sz = sizeof(string);
        memcpy(buffer, x, sz);
        // point to next cell
        sz = (sz + 3) / 4;
        *n -= sz;
        lisp* p = buffer + sz;
        lisp s = (lisp) p;
        buffer[1] = s;
        int len = strlen(getstring(x));

        int ilen = (len + 4) & ~3;
        int iz = ilen / 4;
        if (*n <= 2 + iz) return symbol("*FULL*");
        strncpy((char*)s, getstring(x), len);
        *n -= 2 + iz;
        return (lisp)buffer;
    }
    if (CONSP(x)) {
        // TODO: what if buffer not aligned? cons need be lisp[2] (8 bytes boundary)
        if ((unsigned int)buffer & 7) {
            printf("serializeLisp: not aligned\n");
            *n -= 1;
            return serializeLisp(x, buffer + 1, n);
        }
        *n -= 2;
        lisp* cr = buffer+2; int beforecar = *n;
        buffer[0] = serializeLisp(car(x), cr, n);
        int sz = beforecar - *n;
        buffer[1] = serializeLisp(cdr(x), cr + sz, n);
        printf("CONSP! %x  ", (unsigned int)MKCONS(buffer)); prin1(MKCONS(buffer)); terpri();
        return MKCONS(buffer);
    }

    printf("flashit.ERROR: unsupported type: %d %s\n", TAG(x), tag_name[TAG(x)]);
    return symbol("*ERROR*");
}

#define MAXFLASHPRIM 256

PRIM flashArray(lisp *serialized, int len) {
    if (!CONSP(serialized) && !IS((lisp)serialized, string)) {
        printf("flashArray.ERROR: wrong type %d\n", TAG((lisp)serialized));
        return nil;
    }

    // TODO: move to relocate function, that relocates a RAM region array of lisp*

    // patch up internal references
    lisp* where = malloc(len * sizeof(lisp));
    int i;

    lisp* from = (lisp*)GETCONS(serialized);
    for(i = 0; i < len; i++) {
        lisp p = from[i];
        // TODO: not safe... as it could overlap with some bitrepresentation of other type...
        int o = (lisp)(((unsigned int)p) & ~3) - (lisp)from; // index from from[0]
        //printf("%2d : %d [%x] : ", i, o, (unsigned int)p); prin1(p); terpri();

        where[i] = p;
        if (INTP(p) || SYMP(p))
            ; // TODO: skip over inline symbol...
        else if (stringp(p))
            ; // TODO: skip over inline string...
        else if (o >= 0 && o < MAXFLASHPRIM) // pointers are generated for cons, string (maybe symbol) and some heap types
            where[i] = (lisp)((unsigned int)p - (unsigned int)from + (unsigned int)where);
        //printf("%u %u %d %s\n", (unsigned int) p, (unsigned int) where[i], TAG(where[i]), tag_name[TAG(where[i])]);
    }

    // TODO: actually copy to flash
    int offset = 0; // TODO: find free space to store
    writeBytesToFlash((char*)where, len * sizeof(lisp), offset);
    int flashaddr = (unsigned int)flash_memory + offset;

    free(where);
    if (CONSP(*serialized)) {
        return MKCONS(flashaddr);
    } else {
        // TODO: pointer stupidity, works fine for CONS, string/heap, but not symboll?
        unsigned int us = (unsigned int)serialized;
        return (lisp)((unsigned int)flashaddr | (us & 2)); // 2 ???
    }

    // TODO: remove
    // return pointer to memory, first attemtp
    if (CONSP(*serialized)) return MKCONS(where);

    // TODO: pointer stupidity, works fine for CONS, string/heap, but not symboll?
    unsigned int us = (unsigned int)serialized;
    return (lisp)((unsigned int)where | (us & 2)); // 2 ???
}

PRIM flashit(lisp x) {
    lisp* buffer = malloc(MAXFLASHPRIM * sizeof(lisp)); // align as conss
    memset(buffer, 0, MAXFLASHPRIM * sizeof(lisp));
    int n = MAXFLASHPRIM;
    lisp ret = serializeLisp(x, (lisp*)buffer, &n);
    int len = MAXFLASHPRIM - n;
    if (!len) return x;

  printf("flashit.serialized [len=%d]: ", len); prin1(ret); terpri();
    lisp f = flashArray((lisp*)ret, len);
  printf("flashit.read using lisp[len=%d]: ", len); prin1(f); terpri();
    free(buffer);
    return f;
}

// (flash) -> read all (print), return count
// (flash 3) -> read 3rd entry (1..n), return string, or nil
// (flash -30) -> read at offset 30 (0..), return string, or nil
// (flash "foo") -> append one string
// (flash "foo" 0) -> force write at offset (EOR will mess thins up) (offset = 0 will format)
PRIM flash(lisp s, lisp o) {
    if (!s) { // read all
        int offset = 0;
        int n = 0;
        while (1) {
            int len = readFromFlash(NULL, 0, offset);
            //printf("[LEN=%d]...\n", len);
            if (len < 0) break;

            char buff[len];
            readFromFlash(buff, len, offset);
            n++;
            printf("%d @%d :: [%d] '%s'\n", n, offset, len, buff);

            offset += (len + 3) & ~3;
        }
        return mkint(n);
    } else if (INTP(s)) {
            int n = getint(s);
            int len = -3;
            int offset = n <= 0 ? -n : 0;
            n = n <= 0 ? 1 : n;
            while (n--) {
                offset += (len + 3) & ~3;
                len = readFromFlash(NULL, 0, offset);
                if (len < 0) return nil;
            }
            char buff[len];
            readFromFlash(buff, len, offset);
            return mklenstring(buff, len);
    } else if (IS(s, string)) {
        int offset = integerp(o) ? getint(o) : findLastFlash();
        writeStringToFlash(getstring(s), offset);
        return mkint(findLastFlash());
    } else {
        return nil;
    }
}

////////////////////////////////////////////////////////////////////////////////
// stuff

PRIM fibb(lisp n);

// returns an env with functions
lisp lisp_init() {
    nil = 0;
    int verbose;

    init_symbols();

    // enable to observer startup sequence
    if (1) {
        char* f = readline("PRESS RETURN>", 2);
        if (f) {
            verbose = (f[0] != 0);
            free(f);
        } else {
            verbose = 0;
        }
    }
    print_memory_info( verbose ? 2 : 0 ); // init by first call

    lisp env = nil;
    lisp* envp = &env;

    // free up and start over...
    dogc = 0;

    // TODO: this is a leak!!!
    allocs_next = 0;

    // need to before gc_cons_init()...
    _FREE_ = symbol("*FREE*");

    // setup gc clean slate
    mark_clean();
    gc_cons_init();
    gc(NULL);

    global_envp = envp;

    t = symbol("t");
    DEFINE(t, 1);
    ATSYMBOL = symbol("*at*");
    DEFINE(ATSYMBOL, nil);

    DEFPRIM(lambda, -7, lambda);
    DEFPRIM(nlambda, -7, nlambda);

    // types
    DEFPRIM(null?, 1, nullp);
    DEFPRIM(cons?, 1, consp);
    DEFPRIM(atom?, 1, atomp);
    DEFPRIM(string?, 1, stringp);
    DEFPRIM(symbol?, 1, symbolp);
    DEFPRIM(number?, 1, numberp);
    DEFPRIM(integer?, 1, integerp);
    DEFPRIM(func?, 1, funcp);

    // mathy stuff
    DEFPRIM(iota, 3, iota);
    DEFPRIM(+, -7, plus);
    DEFPRIM(-, 2, minus);
    DEFPRIM(*, -7, times);
    DEFPRIM(/, 2, divide);
    DEFPRIM(%, 2, mod);

    DEFPRIM(eq, 2, eq);
    DEFPRIM(cmp, 2, cmp_);
    DEFPRIM(equal, 2, equal);
    DEFPRIM(=, 2, equal);
    DEFPRIM(<, 2, lt);
    DEFPRIM(<=, 2, lte);
    DEFPRIM(>, 2, gt);
    DEFPRIM(>=, 2, gte);

    // all other <= > >= can be made from cmp

    DEFPRIM(random, 2, random_);

    // https://www.gnu.org/software/guile/manual/html_node/Pattern-Matching.html#Pattern-Matching
    DEFPRIM(if, -3, if_);
    DEFPRIM(cond, -7, cond);
    DEFPRIM(case, -7, case_);
    DEFPRIM(and, -7, and);
    DEFPRIM(or, -7, or);
    DEFPRIM(not, 1, not);

    // output
    // TODO: write procedures? - http://www.gnu.org/software/mit-scheme/documentation/mit-scheme-ref/Output-Procedures.html
    // TODO: make these take a "fd" or output stream?
    DEFPRIM(terpri, 0, terpri);
    DEFPRIM(princ, 1, princ);
    DEFPRIM(prin1, 1, prin1);
    DEFPRIM(print, 1, print);
    DEFPRIM(printf, 7, printf_);
    DEFPRIM(pp, 1, pp); // TODO: pprint?
    DEFPRIM(with-putc, -7, with_putc);
    DEFPRIM(with-fd, -7, with_fd);
    DEFPRIM(with-fd-json, -7, with_fd_json);

    // cons/list
    DEFPRIM(cons, 2, cons);
    DEFPRIM(recons, 3, recons);
    DEFPRIM(car, 1, car_);
    DEFPRIM(cdr, 1, cdr_);
    DEFPRIM(set-car!, 2, setcar);
    DEFPRIM(set-cdr!, 2, setcdr);
    DEFPRIM(nth, 2, nth);
    DEFPRIM(nthcdr, 2, nthcdr);

    DEFPRIM(list, 7, _quote);
    DEFPRIM(length, 1, length);
    DEFPRIM(concat, 7, concat); // scheme: string-append/string-concatenate
    DEFPRIM(char, 1, char_); // scheme: integer->char
    DEFPRIM(split, 3, split); // scheme: string-split
    DEFPRIM(assoc, 2, assoc);
    DEFPRIM(member, 2, member);
    DEFPRIM(mapcar, 2, mapcar);
    DEFPRIM(mapc, 2, mapc);
    DEFPRIM(filter, 2, filter);
    DEFPRIM(reduce, 2, reduce);
    DEFPRIM(filtermapfilterreduce, 5, filtermapfilterreduce);
    DEFPRIM(quote, -1, _quote);
    // DEFPRIM(quote, -7, quote); // TODO: consider it to quote list?
    // DEFPRIM(list, 7, listlist);

    // TODO: consider some functions - http://srfi.schemers.org/srfi-1/srfi-1.html
    DEFPRIM(let, -7, let);
    DEFPRIM(let*, -7, let_star);
    DEFPRIM(progn, -7, progn);
    DEFPRIM(eval, 2, _eval);
    DEFPRIM(evallist, 2, evallist);
    DEFPRIM(apply, 2, apply);
    DEFPRIM(env, -7, _env);

    DEFPRIM(read, 1, read_);
    DEFPRIM(edit, 1, edit);

    // TODO: consider introducting these that will create local bindings if no global exists, hmm bad?
    DEFPRIM(set, -2, _set);
    //DEFPRIM(setq, -2, _setq);
    //DEFPRIM(setqq, -2, _setqq_);
    DEFPRIM(set!, -2, _setbang);

    DEFPRIM(define, -7, _define);
    DEFPRIM(de, -7, de);

    DEFPRIM(fundef, 1, fundef);
    DEFPRIM(funenv, 1, funenv);
    DEFPRIM(funame, 1, funame);

    // define
    // defun
    // defmacro
    // while
    // gensym

    // network
    DEFPRIM(wget, 3, wget_);
    DEFPRIM(web, -2, web);

    // hardware
    DEFPRIM(out, 2, out);
    DEFPRIM(in, 1, in);
    DEFPRIM(dht, 1, dht);
    DEFPRIM(interrupt, 2, interrupt);

    // system stuff
    DEFPRIM(gc, -1, gc);
    DEFPRIM(test, -7, test);

    DEFPRIM(ticks, 1, ticks);
    DEFPRIM(clock, 1, clock_);
    DEFPRIM(delay, 1, delay);
    DEFPRIM(time, -1, time_);

    DEFPRIM(load, -3, load); // -3 to get env?
    DEFPRIM(dir, 1, dir);
    DEFPRIM(cat, 1, cat);

/* TODO: http://clhs.lisp.se/Front/X_Mast_W.htm

    // DEFPRIM(make-string-input-stream, 1, );
    // DEFPRIM(with-input-from-string, 1, );
    // DEFPRIM(with-open-file, 1, );
    // DEFPRIM(with-open-stream, 1, );
    // DEFPRIM(with-output-to-string, 1, );
    // DEFPRIM(with-output-to-string, 1, );
    DEFPRIM(open, 2, open_); // r, r+ w, w+, a, a+
    // SCHEME
    DEFPRIM(open-input-file, );
    DEFPRIM(open-output-file, );
    DEFPRIM(display);
    DEFPRIM(newline);

    DEFPRIM(close, 1, close_);
    DEFPRIM(flush, 1, flush_);
    //DEFPRIM(delete, 1, delete_);
    //DEFPRIM(rename, 2, rename_);
    DEFPRIM(seek, 2, seek_);    
    DEFPRIM(pos, 1, pos_);    
    DEFPRIM(fsinfo, 2, fsinfo);    

    DEFPRIM(read-line, 1, readline_);
    // maybe one write function? char/byte/lisp object
    DEFPRIM(write-char, 1, writechar_);
    DEFPRIM(write-byte, 1, writebyte_);
    DEFPRIM(write-line, 1, writeline_);
    DEFPRIM(write-string, 1, writeline_);
    DEFPRIM(write-to-string, 1, write-to-string_);
    DEFPRIM(read, 1, read_);
*/

    // debugging - http://www.gnu.org/software/mit-scheme/documentation/mit-scheme-user/Debugging-Aids.html 
    // http://www.gnu.org/software/mit-scheme/documentation/mit-scheme-user/Command_002dLine-Debugger.html#Command_002dLine-Debugger
    // TODO: set-trace! set-break! http://www.lilypond.org/doc/v2.19/Documentation/contributor/debugging-scheme-code
    DEFPRIM(pstack, 0, print_detailed_stack); 

    // flash stuff - experimental
    DEFPRIM(flash, 2, flash);
    DEFPRIM(flashit, 1, flashit);
    DEFPRIM(scan, 2, scan);
    // TODO: consider integrating with - https://www.gnu.org/software/guile/manual/html_node/Symbol-Props.html
    // 2d-!!! https://groups.csail.mit.edu/mac/ftpdir/scheme-7.4/doc-html/scheme_12.html#SEC105
    // 2d-put! 2d-remove! 2d-get 2d-get-alist-x 2d-get-alist-y

    // scheduling
    DEFPRIM(at, -2, at); // TODO: eval at => #stop?!?!??!
    DEFPRIM(stop, -1, stop);
    // DEFPRIM(atrun, -1, atrun); // no reason for user to call (yet)

    // DEFPRIM(imacs, -1, imacs_); // link in the imacs?
    DEFPRIM(syms, 0, syms); // TODO: rename to apropos?
    DEFPRIM(fib, 1, fibb);

    //DEFPRIM(readit, 0, readit);
    DEFPRIM(heap, 1, heap);
    DEFPRIM(rec, 1, rec);

    // another small lisp in 1K lines
    // - https://github.com/rui314/minilisp/blob/master/minilisp.c

    // neat "trick" - limit how much 'trace on' will print by injecting nil bound to nil
    // in evalGC function the print ENV stops when arriving at this token, thusly
    // will not show any variables defined above...
    env = cons( cons(nil, nil), env );

    dogc = 1;

    print_memory_info( verbose ? 2 : 0 ); // summary of init usage

    return env;
}

void about() {
    printf("\n2016 (c) Jonas S Karlsson under MPL 2.0\n");
    printf("Read more on https://github.com/yesco/esp-lisp/\n");
    printf("Welcome to esp-lisp! Type 'help' for help.\n");
}

void help(lisp* envp) {
    about();
    printf("\nGlobal/SYMBOLS: ");
    PRINT((syms (lambda (x) (princ x) (princ " "))));
    printf("\nCOMMANDS: help/trace on/trace off/gc on/gc off/wifi SSID PSWD/wget SERVER URL/mem EXPR/quit/exit\n\n");
    printf("CTRL-C: to break execution, CTRL-T: shows current time/load status, CTRL-D: to exit\n\n");
    printf("Type 'help' to get this message again\n");
}

// TODO: make it take one lisp parameter?
// TODO: https://groups.csail.mit.edu/mac/ftpdir/scheme-7.4/doc-html/scheme_17.html#SEC153
void error(char* msg) {
    jmp_buf empty = {0};
    static int error_level = 0;

    // restore output to stdout, if error inside print function we're screwed otherwise!
    writeputc = origputc;

    terpri();
    if (error_level == 0) {
        error_level++;
        if (level) { printf("%%%s\nBacktrace: ", msg); print_stack(); terpri(); }
        print_detailed_stack();
        printf("%s -- see above!\n", msg);
        error_level--;
    } else {
        error_level = 0;
        printf("%% error(): error inside error... recovering...\n");
    }

    if (memcmp(lisp_break, empty, sizeof(empty))) { // contains valid value
        // reset stack
        level = 0;
        stack[0].e = nil;
        stack[0].envp = NULL;

        longjmp(lisp_break, 1);
        // does not continue!
    } else {
        printf("%%%% error(): NOT inside setjmp, continuing, ...possibly bad\n");
        // exit(77);
    }
}

void run(char* s, lisp* envp) {
    if (setjmp(lisp_break) == 0) {
        lisp r = reads(s);
        prin1(evalGC(r, envp)); terpri();
        // TODO: report parsing allocs separately?
        // mark(r); // keep history?
    } else {
        // escaped w ctrl-c (longjmp)

        // enableGC and kill stack
        blockGC = 0;
        level = 0;
        trace_level = 0;
        printf("%%%% type 'help' to get help\n");
    }
    // disable longjmp
    memset(lisp_break, 0, sizeof(lisp_break));

    gc(envp); // TODO: maybe move out!
}

// it would not be completely safe to run multiple threads of lisp at the
// same time, problems are GC and allocations. Therefore we provide a tasker
// that handles "actors". To drive this, we only do it while idle, i.e,
// when waiting for keyboard input. This is similar to NodeMCU.
//
// The system actors that are checked are:
// - TODO: web server
// - TODO: wget 

void maybeGC() {
    if (blockGC || !global_envp) return;
    if (needGC()) gc(global_envp);
}

void handleInterrupts() {
    // TODO: maybe cache the symbols? not matter in idle, but if called elsewhere
    int checkpin(int pin, uint32_t clicked, uint32_t count, uint32_t last) {
        char name[16] = {0};
        snprintf(name, sizeof(name), "int%02d", pin);
        // this will define the symbol, but only for interrupts enabled
        lisp handler = getvar(symbol(name), *global_envp);
        if (!handler) return -666;
        //printf("BUTTON: %d count %d last %d\n", pin, count, last);
        lisp r = apply(handler, list(mkint(pin), mkint(clicked), mkint(count), mkint(last), END));
        return getint(r);
    }
                  
    checkInterrupts(checkpin);
}

PRIM atrun(lisp* envp);

PRIM idle(int lticks) {
    // 1 000 000 == 1s for x61 laptop
    //if (lticks % 1000000 == 0) { putchar('^'); fflush(stdout); }

    // polling tasks, invoking callbacks
    web_one();
    atrun(global_envp);
    handleInterrupts();

    // gc
    maybeGC(); // TODO: backoff, can't do all the time???

    // clean stats
    print_memory_info(0);

    return nil;
}

// tweenex style ctrl-t process status
// freebsd10$ sleep 5 ... ctrl-t
// load: 0.67  cmd: sleep 90628 [nanslp] 0.92r 0.00u 0.00s 0% 1464k
// sleep: about 4 second(s) left out of the original 5
// Could print "load" (ticks last second, keep time last tick)
void print_status(long last_ticks, long ticks, int last_ms, int ms, int max_ticks_per_ms) {
    int s = ms / 1000;
    int m = s / 60; s -= m * 60;
    // we approximate the load with ticks seen since "last"
    int ld = 100 - (100 * (ticks - last_ticks)) / max_ticks_per_ms;
    if (ld < 0) ld = 0;
    if (ld > 99) ld = 99;
    // %3.2f doesn't work on esp-open-rtos
    printf("[%% %d:%02d load: 0.%02d ", m, s, ld);
    // printf("dticks: %ld, last: %ld, ticks: %ld mtps = %d",
    //    ld, ticks - last_ticks, last_ticks, ticks, max_ticks_per_ms);
    print_stack();
    printf("]\n");
}

// make an blocking mygetchar() that waits for kbhit() to be true 
// and meanwhile calls idle() continously.
static int thechar = 0;

// called from idle in tight loop, also called each time we gc() during execution
int kbhit() {
    static int last_ms = 0;
    static int last_ticks = 0;
    static int max_ticks_per_ms = 1;

    int ms = clock_ms();
    int update = 0;
    // update every s
    if (ms - last_ms > 1000) {
        int tms = (lisp_ticks - last_ticks) / (ms - last_ms);
        if (tms > max_ticks_per_ms) max_ticks_per_ms = tms;
        update = 1;
    }

    // TODO: "bug", if there is a ungotten char that won't be read, ctrl-c ctrl-t will not be seen :-(
    // alt 1) could dispose of characters :-(
    // alt 2) have a "buffer" (but how long?) may be unbounded and trouble for piped strings to program
    if (!thechar) {
        thechar = nonblock_getch();
        if (thechar < 0) thechar = 0;
        if (thechar == 'T'-64) {
            print_status(last_ticks, lisp_ticks++, last_ms, ms, max_ticks_per_ms);
            thechar = 0;
        }
        if (thechar == 'C'-64) {
            int c = thechar;
            thechar = 0;
            error("CTRL-C");
            // error only returns if couln't longjmp to setjmp position, so keep the ctrl-c
            thechar = c;
        }
    }

    if (update) {
        last_ms = ms;
        last_ticks = lisp_ticks;
    }

    return thechar;
}

int mygetchar() {
    while (!kbhit()) idle(lisp_ticks++); 
    int c = thechar;
    thechar = 0;
    return c;
}

int lispreadchar(char *chp) {
    int c = mygetchar();
    if (c >= 0) *chp = c;
    return c < 0 ? -1 : 1;
}

void readeval(lisp* envp) {
    run("(load \"init.lsp\")", envp);
    about();

    while(1) {
        global_envp = envp; // allow idle to gc
        char* ln = readline_int("lisp> ", READLINE_MAXLEN, lispreadchar);
        global_envp = NULL;

        if (!ln) {
            break;
        } else if (strncmp(ln, ";", 1) == 0) {
            ; // comment - ignore
        } else if (strcmp(ln, "help") == 0 || ln[0] == '?') {
            help(envp);
        } else if (strcmp(ln, "gc on") == 0) {
            traceGC = 1;
        } else if (strcmp(ln, "gc off") == 0) {
            traceGC = 0;
        } else if (strcmp(ln, "trace on") == 0) {
            trace = 1;
        } else if (strcmp(ln, "trace off") == 0) {
            trace = 0;
        } else if (strncmp(ln, "wifi ", 5) == 0) {
            strtok(ln, " "); // skip wifi
            char* ssid = strtok(NULL, " "); if (!ssid) ssid = "dsl";
            char* pass = strtok(NULL, " "); if (!pass) pass = "0xdeadbeef";
            connect_wifi(ssid, pass);
        } else if (strncmp(ln, "wget ", 5) == 0) {
            strtok(ln, " "); // skip wget
            char* server = strtok(NULL, " "); if (!server) server = "yesco.org";
            char* url = strtok(NULL, " ");    if (!url) url = "http://yesco.org/index.html";
            printf("SERVER=%s URL=%s\n", server, url);
            int r = http_get(url, server);
            printf("GET=> %d\n", r);
        } else if (strncmp(ln, "web ", 4) == 0) {
            strtok(ln, " "); // skip web
            char* ports = strtok(NULL, " ");
            int port = ports ? atoi(ports) : 8080;
            printf("WEBSERVER on port=%d\n", port);
            int s = httpd_init(port);
            if (s < 0) printf("WEBSERVER.errno=%d: s=%d\n", errno, s);
            while(1) {
                httpd_loop(s);
            }
        } else if (strncmp(ln, "mem ", 4) == 0) {
            char* e = ln + 4;
            if (*e) {
                print_memory_info(0);
                run(e, envp);
            }
            print_memory_info(2);
        } else if (strcmp(ln, "mem") == 0) {
            print_memory_info(1);
        } else if (strcmp(ln, "exit") == 0 || strcmp(ln, "quit") == 0 || strcmp(ln, "bye") == 0) {
            exit(0);
        } else if (strlen(ln) > 0) { // lisp
            global_envp = envp; // allow idle to gc
            print_memory_info(0);
            run(ln, envp);
            global_envp = NULL;
            //print_memory_info(1);
        }

        free(ln);
    }

    printf("OK, bye!\n");
}

void treads(char* s) {
    printf("\nread-%s: ", s);
    princ(reads(s));
    terpri();
}
    
int fib(int n) {
    if (n < 2) return 1;
    else return fib(n-1) + fib(n-2);
}

PRIM fibb(lisp n) { return mkint(fib(getint(n))); }

// lisp implemented library functions hardcoded
void init_library(lisp* envp) {
  //DEFINE(fibo, (lambda (n) (if (< n 2) 1 (+ (fibo (- n 1)) (fibo (- n 2))))));
  DE((fibo (n) (if (< n 2) 1 (+ (fibo (- n 1)) (fibo (- n 2))))));

// POSSIBLE encodings to save memory:
    // symbol: fibo
    // "fibo" 
// 0: (lambda
// 1:  (n)
// 2: (if
// 3:  (<
// 4:   n
// 5:   2)
// 6:  1
// 7:  (+
// 8:   (fibo
//10:    (-
//11:     n
//12:      1))
//14:    (fibo
//15:     (-
//16:      n
//17:      2))))));
// 18 lines + 9 paranthesises, a cons 8 bytes
// (* (+ 18 9) 8) = 216 bytes
//
// possible storage strategies:
// - 64 bytes: (lambda (n) (if (< n 2) 1 (+ (fibo (- n 1)) (fibo (- n 2)))))
// - 52 bytes: (lambda(n)(if(< n 2)1(+(fibo(- n 1))(fibo(- n 2)))))
//
// - 24 conses! (* 24 8) = 192 bytes
// - 25: L3 lambda L1 n L4 if L3 < n 2 1 L2 + L2 fibo L3 - n 1 L2 fibo L3 - n 2
//   25 slots 25*4 = 100 bytes, can't detect end of list by "cdr"
// - 34: L3 lambda L1 n \0 L4 if L3 < n 2 1 \0 L2 + L2 fibo L3 - n 1 \0 \0 L2 fibo L3 - n 2 \0 \0 \0 \0 \0
//   34 slots (* 34 4) = 136 bytes, 
// - L:fibo  lambda L:P L:IF \0  4
//   L:P     n \0                2
//   L:IF    if L:X 1 L:E \0     4
//   L:X     < n 2 \0            4
//   L:E     + L:F1 L:F2 \0      4
//   L:F1    fibo L:M1 \0        3
//   L:M1    - n 1 \0            4
//   L:F2    fibo L:M2 \0        3
//   L:M2    - n 2 \0            4
//                            =====
//                              32 = 128 bytes
//
// - skip list with end and compact
//  #34 lambda #3 n \0 #26 if #5 < n 2 \0 1 #18 + #8 fibo #5 - n 1 \0 \0 #8 fibo #5 - n 2 \0 \0 \0 \0 \0
//  skip marker 1-2 bytes, end marker \0 1 bytes
//  9 skips, 9 ends = 18 bytes
//  8 functions, 6 are basic (2b), 2 user (fibo) (4b) =  (+ (* 6 2) (* 2 4)) = 20 bytes
//  4 numbers: 2, 1, 1, 2 - 2 byte x 4 = 8 bytes
//  TOTAL: (+ 18 20 8) = 46 bytes compact!
//  but: can execute? need a new "VM", or decode on the fly to "cons"
//
// - forth stack style: : fibo 1 ref 2 < 0skip? jmp:+3 1 ret 1 ref 2 - fibo 1 ref 2 - fibo + ret ;
//   20 instructions: 40 bytes
// 
// HOWEVER: If put in flash, maybe we don't need to care?
// 
//   picolisp: Only 20 of the 196 bytes stay in RAM (for the fibo symbol), the rest is moved off to ROM.
//   esp-lisp: all conses can move "moved" to FLASH! (192 bytes!) + 4 bytes for the "function", must be RAM, plus symbol (none)
}

void testc(lisp* envp , char* whats, lisp val, lisp expect) {
    printf("TEST: %s\n=> ", whats);
    lisp r = val;
    princ(r);
    printf("\nexpected: "); princ(expect); terpri();
    printf("status: "); printf("%s\n\n", equal(r, expect) ? "passed" : "failed");
}

void testee(lisp* envp , lisp what, lisp expect) {
    printf("TEST: "); princ(what); printf("\n=> ");
    lisp r = eval(what, envp);
    princ(r);
    printf("\nexpected: "); princ(expect); terpri();
    printf("status: "); printf("%s\n\n", equal(r, expect) ? "passed" : "failed");
}

void testss(lisp* envp , char* what, char* expect) {
    testee(envp, reads(what), reads(expect));
}

// TODO: implement, (port 8080) => p, (listen p) (http @) (close @)
//   https://github.com/SuperHouse/esp-open-rtos/commit/147257efa472307608019f04f38f8ebadadd7c01
//   http://john.freml.in/teepeedee2-vs-picolisp
//   http://picolisp.com/wiki/?ErsatzWebApp

static PRIM test(lisp* e) {

// removing the body of this function gives: (- 42688 41152) 1536
// gives 25148 bytes, 19672k
#ifdef TEST_REMOVED
// enabling this take 1K from RAM :-(
//   also: -rw-r--r-- 1 knoppix knoppix  43712 Nov  6 20:12 0x00000.bin 
//   becomes 43712 instead of 42688
//static IROM 
//const char const xxx[] = "===============================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================\n";
//    printf(&xxx[0]);

    lisp env = *e;
    lisp* envp = &env; // make local, don't leak out!

    terpri();

    TEST(nil, nil);
    TEST(nil, (car (quote (nil . nil)))); // lol (was an issue)
    TEST(nil, (cdr (quote (nil . nil)))); // lol

    TEST(42, 42);

    // make sure have one failure
    TEST(t, fail);

    // equal
    //testee(list(symbol("eq"), mkint(42), mkint(42), END), t);

    TEST((equal nil nil), t);
    TEST((equal 42 42), t);
    TEST((equal (list 1 2 3) (list 1 2 3)), t);
    TEST((equal "foo" "bar"), nil);
    TEST((equal "foo" "foo"), t);
    TEST((equal equal equal), t);

    // list & read function
    testee(envp, list(nil, mkstring("fihs"), mkint(1), symbol("fish"), mkint(2), mkint(3), mkint(4), nil, nil, nil, END),
           reads("(nil fihs 1 fish 2 3 4 nil nil nil)"));

    // misc
    testss(envp, "123", "123");
    TEST("()", "nil");
    TEST("(1)", "(1)");
    TEST("(1 2)", (1 2));
    treads("(1 2 3)");
    treads("((1) (2) (3))");
    treads("(A)");
    treads("(A B)");
    treads("(A B C)");
    printf("\n3=3: "); princ(eq(mkint(3), mkint(3)));
    printf("\n3=4: "); princ(eq(mkint(3), mkint(4)));
    printf("\na=a: "); princ(eq(symbol("a"), symbol("a")));
    printf("\na=b: "); princ(eq(symbol("a"), symbol("b")));
    printf("\na=a: "); princ(eq(symbol("a"), symbol("a")));
    printf("\n");

    treads("(lambda (n) if (eq n 0) (* n (fac (- n 1))))");

    // read
    TEST((read "foo"), foo);
    TEST((read "(+ 3 4)"), (+ 3 4));
    TEST((number? (read "42")), t);
    
    // set, setq, setqq
    TEST((define a (+ 3 4)), 7);
    //TEST((setqq b a), a);
    TEST((set! b (quote a)));
    TEST(b, a);
    //TEST((set b 3), 3);
    //TEST(a, 3);
    //TEST(b, a);
       
    // if
    lisp IF = mkprim("if", -3, if_);
    testee(envp, IF, IF);
    testee(envp, cons(IF, cons(mkint(7), cons(mkint(11), cons(mkint(22), nil)))), mkint(11));
    testee(envp, cons(IF, cons(nil, cons(mkint(11), cons(mkint(22), nil)))), mkint(22));
    TEST((if 7 11 22), 11);
    TEST((if nil 11 22), 22);
    TEST((if nil 11 22 33), 22);

    // thunk
    lisp th = mkthunk(mkint(14), NULL);
    testee(envp, th, th); // eval(thunk)
    testee(envp, cons(th, nil), th); // (eval (thunk))

    // lambda
    TEST((func? (lambda (n) 37)), t);
    TEST(((lambda (n) 37) 99), 37);
    TEST(((lambda (n) n) 99), 99);
    TEST(((lambda (a) ((lambda (n) (+ n a)) 33)) 66), 99); // lexical scoping

    // recursion
    DEFINE(fac, (lambda (n) (if (= n 0) 1 (* n (fac (- n 1))))));
    TEST((fac 6), 720);
    TEST((fac 21), 952369152);

    // tail recursion optimization test (don't blow up stack!)
    DEFINE(bb, (lambda (b) (+ b 3)));
    DEFINE(aa, (lambda (a) (bb a)));
    TEST((aa 7), 10);

    DEFINE(tail, (lambda (n s) (if (eq n 0) s (tail (- n 1) (+ s 1)))));
    TEST(tail, xyz);
    testss(envp, LOOPTAIL, LOOPS);

    // progn, progn tail recursion
    TEST((progn 1 2 3), 3);
    TEST((set! a nil), nil);
    TEST((progn (set! a (cons 1 a)) (set! a (cons 2 a)) (set! a (cons 3 a))),
         (3 2 1));

    // implicit progn in lambda
    DEFINE(f, (lambda (n) (set! n (+ n 1)) (set! n (+ n 1)) (set! n (+ n 1))));
    TEST((f 0), 3);

//    PRINT((define tailprogn (lambda (n) (progn 3 2 1 (if (= n 0) (quote ok) (tailprogn (- n 1)))))));
//    TEST(tailprogn, 3);
//    TEST((tailprogn 10000), ok);

    // cond
    TEST((cond), nil);
    TEST((cond (7)), 7);
    TEST((cond (1 2 3)), 3);
    TEST((cond (nil 7)), nil);
    TEST((cond (nil 7)(2 3 4)(7 99)), 4);
    TEST((cond (nil 7)((eq 3 5) 9)((eq 5 5) 77)), 77);

    TEST((and), nil);
    TEST((and 1 2 3), 3);
    TEST((and 1 nil 3), nil);
    TEST((and 1 (eq 3 3) 7), 7);
    TEST((and 1 (eq 3 4) 7), nil);

    TEST((or), nil);
    TEST((or 1 2 3), 1);
    TEST((or nil 1 3), 1);
    TEST((or (eq 3 3) 7), t);
    TEST((or (eq 3 4) 7), 7);

    // mapcar
    TEST((mapcar (lambda (x) (+ 5 x)) (list 1 2 3)), (6 7 8));
    TEST((mapcar car (list (cons 1 2) (cons 3 4) (cons 5 6))), (1 3 5));
    TEST((mapcar cdr (list (cons 1 2) (cons 3 4) (cons 5 6))), (2 4 6));

    TEST((set! a 2));
    TEST((list 1 2 (let ((a (+ 1 a)) (b a)) (list a (+ b b))) 5 (+(+ a (+ a a))), (1 2 (3 4) 5 6)));
    TEST(a, 2);

#else
    printf("%%Tests have been commented out.\n");
#endif

    return nil;
}

void lisp_run(lisp* envp) {
    init_library(envp);
    readeval(envp);
    return;
}

// find this display for chinese price - http://digole.com/index.php?productID=1208 (17.89 USD)
//
// esp8266 st7735 lcd/tft display driver ucglib
//
// https://github.com/spapadim/ESPClock
//
// https://github.com/bblanchon/ArduinoJson
//
// http://www.pjrc.com/teensy/td_libs_Time.html
//
// https://github.com/spapadim/ESPClock
//
// https://gist.github.com/spapadim/a4bc258df47f00831006

// http://stackoverflow.com/questions/8751264/memory-mapped-using-linker
// https://github.com/esp8266/esp8266-wiki/wiki/Memory-Map
// http://esp8266-re.foogod.com/wiki/Memory_Map#Data_RAM_.280x3FFE8000_-_0x3FFFFFFF.29