diff --git a/.gitmodules b/.gitmodules
index bb67b08..199449c 100644
--- a/.gitmodules
+++ b/.gitmodules
@@ -1,6 +1,3 @@
 [submodule "unicope"]
 	path = unicope
 	url = https://github.com/bumbread/unicope.git
-[submodule "ryu"]
-	path = ryu
-	url = https://github.com/bumbread/ryu.git
diff --git a/bake.py b/bake.py
index 7c35ce6..09e9a77 100644
--- a/bake.py
+++ b/bake.py
@@ -1,9 +1,29 @@
 
 import os
+import sys
 import subprocess
 import runpy
 from pathlib import Path
 
+for arg in sys.argv[1:]:
+    if arg == 'test':
+        test = os.getenv('test');
+        if test == None:
+            test = 'assert'
+        test_file = 'test/test_' + test + '.c'
+        subprocess.run([
+            'clang',
+            test_file,
+            '-Iinc',
+            '-g',
+            '-luser32',
+            '-lkernel32',
+            '-lshell32',
+            '-lDbghelp',
+            '-lciabatta.lib',
+        ])
+        sys.exit(0)
+
 # Build dependencies rq
 if not Path("ryu/ryu.lib").exists():
     os.chdir('ryu')
@@ -23,7 +43,8 @@ do_cuik = False
 
 inc_folders = [
     'inc',
-    os.path.join('unicope', 'inc'),
+    'unicope/inc',
+    'ryu',
 ]
 
 definitions = [
@@ -101,5 +122,5 @@ compile(os.path.normpath('src/code'), compile_map)
 for dir, _, f in os.walk('bin'):
     if len(f) != 0:
         obj_paths.append(os.path.join(dir, '*.obj'))
-subprocess.run(['llvm-ar', 'rc', 'ciabatta.lib'] + obj_paths)
+subprocess.run(['lib', '/out:ciabatta.lib'] + obj_paths)
 print('*.obj => ciabatta.lib')
\ No newline at end of file
diff --git a/src/code/fmt/gen_fmt.h b/src/code/fmt/gen_fmt.h
index 0ca9a95..81ce1a7 100644
--- a/src/code/fmt/gen_fmt.h
+++ b/src/code/fmt/gen_fmt.h
@@ -3,7 +3,11 @@
 #include <stdlib.h>
 #include <inttypes.h>
 #include <limits.h>
+#include <string.h>
 #include <math.h>
+#include <ctype.h>
+
+#include <ryu/ryu.h>
 
 typedef int (pfx(cbfn))(void *ctx, ctype ch);
 
@@ -42,6 +46,24 @@ static inline int pfx(_dtoh)(
     int width,
     int flags
 );
+static inline int pfx(_dtoa)(
+    int w,
+    void *ctx,
+    pfx(cbfn) cb,
+    double value,
+    int prec,
+    int width,
+    int flags
+);
+static inline int pfx(_etoa)(
+    int w,
+    void *ctx,
+    pfx(cbfn) cb,
+    double value,
+    int prec,
+    int width,
+    int flags
+);
 static inline int pfx(_infnantoa)(
     int w,
     void *ctx,
@@ -229,7 +251,38 @@ static int pfx(vprintfcb)(
         case 'F':
         case 'g':
         case 'G':
-
+            if(*fmt == 'E' || *fmt == 'F' || *fmt == 'G') flags |= FLAG_UPPER;
+            char conv = tolower(*fmt);
+            ++fmt;
+            double value = va_arg(va, double);
+            if(conv == 'f') {
+                w = pfx(_dtoa)(w, ctx, cb, value, prec, width, flags);
+            }
+            else if(conv == 'e') {
+                w = pfx(_etoa)(w, ctx, cb, value, prec, width, flags);
+            }
+            else {
+                int P = prec;
+                if(!(flags & FLAG_PREC)) P = 6;
+                if(prec == 0) P = 1;
+                union {
+                    uint64_t bits;
+                    double value;
+                } _ = {.value = value};
+                uint64_t bits = _.bits;
+                int E = (int)((bits >> 52) & 0x7ff) - 1023;
+                int class = fpclassify(value);
+                if(class == FP_SUBNORMAL || class == FP_ZERO) {
+                    E = 0;
+                }
+                if(P > E && E >= -4) {
+                    w = pfx(_dtoa)(w, ctx, cb, value, P-(E+1), width, flags);
+                }
+                else {
+                    w = pfx(_etoa)(w, ctx, cb, value, P-1, width, flags);
+                }
+            }
+            if(w < 0) return -1;
             break;
         }
     }
@@ -516,4 +569,78 @@ static inline int pfx(_dtoh)(
     return w;
 }
 
+static inline int pfx(_dtoa)(
+    int w,
+    void *ctx,
+    pfx(cbfn) cb,
+    double value,
+    int prec,
+    int width,
+    int flags
+) {
+    int class = fpclassify(value);
+    if(class == FP_INFINITE || class == FP_NAN) {
+        return pfx(_infnantoa)(w, ctx, cb, value, prec, width, flags);
+    }
+    // This guy does memory allocation which makes it pretty cringe
+    if(!(flags & FLAG_PREC)) prec = 6;
+    char *buf = d2fixed(value, prec);
+    int len = (int)strlen(buf);
+    int pad = width - len;
+    // Left pad
+    if(!(flags & FLAG_LEFT) && !(flags & FLAG_ZERO)) while(pad-- > 0) {
+        out(' ');
+    }
+    {
+        char *str = buf;
+        while(*str) {
+            out(*str);
+            ++str;
+        }
+    }
+    free(buf);
+    // Right pad
+    if(flags & FLAG_LEFT) while(pad-- > 0) {
+        out(' ');
+    }
+    return w;
+}
+
+static inline int pfx(_etoa)(
+    int w,
+    void *ctx,
+    pfx(cbfn) cb,
+    double value,
+    int prec,
+    int width,
+    int flags
+) {
+    int class = fpclassify(value);
+    if(class == FP_INFINITE || class == FP_NAN) {
+        return pfx(_infnantoa)(w, ctx, cb, value, prec, width, flags);
+    }
+    // This guy does memory allocation which makes it pretty cringe
+    if(!(flags & FLAG_PREC)) prec = 6;
+    char *buf = d2exp(value, prec);
+    int len = (int)strlen(buf);
+    int pad = width - len;
+    // Left pad
+    if(!(flags & FLAG_LEFT) && !(flags & FLAG_ZERO)) while(pad-- > 0) {
+        out(' ');
+    }
+    {
+        char *str = buf;
+        while(*str) {
+            out(*str);
+            ++str;
+        }
+    }
+    free(buf);
+    // Right pad
+    if(flags & FLAG_LEFT) while(pad-- > 0) {
+        out(' ');
+    }
+    return w;
+}
+
 #undef out
diff --git a/src/code/math/cordic/cordic_data.c b/src/code/math/cordic/cordic_data.h
similarity index 100%
rename from src/code/math/cordic/cordic_data.c
rename to src/code/math/cordic/cordic_data.h
diff --git a/src/code/math/cordic/cordic_dataf.c b/src/code/math/cordic/cordic_dataf.h
similarity index 100%
rename from src/code/math/cordic/cordic_dataf.c
rename to src/code/math/cordic/cordic_dataf.h
diff --git a/src/code/math/cordic/cordic_datal.c b/src/code/math/cordic/cordic_datal.h
similarity index 100%
rename from src/code/math/cordic/cordic_datal.c
rename to src/code/math/cordic/cordic_datal.h
diff --git a/src/code/math/cordic/maketab.py b/src/code/math/cordic/maketab.py
index d46700e..055d754 100644
--- a/src/code/math/cordic/maketab.py
+++ b/src/code/math/cordic/maketab.py
@@ -9,9 +9,9 @@ dname = os.path.dirname(abspath)
 os.chdir(dname)
 
 data=[
-    ('cordic_dataf.c', 'float', 'f',       32),
-    ('cordic_data.c',  'double', '',       64),
-    ('cordic_datal.c', 'long double', 'l', 64),
+    ('cordic_dataf.h', 'float', 'f',       32),
+    ('cordic_data.h',  'double', '',       64),
+    ('cordic_datal.h', 'long double', 'l', 64),
 ]
 
 for f in data:
diff --git a/src/code/math/gen_math.c b/src/code/math/gen_math.c
index 206d8cc..0ef25bb 100644
--- a/src/code/math/gen_math.c
+++ b/src/code/math/gen_math.c
@@ -14,21 +14,21 @@ static double LOG2E     = 1.442695040888963407359924681001892;
 
 #define ftype float
 #define suffix(name) name ## f
-#include "cordic/cordic_dataf.c"
+#include "cordic/cordic_dataf.h"
 #include "gen_math.h"
 #undef ftype
 #undef suffix
 
 #define ftype double
 #define suffix(name) name
-#include "cordic/cordic_data.c"
+#include "cordic/cordic_data.h"
 #include "gen_math.h"
 #undef ftype
 #undef suffix
 
 #define ftype long double
 #define suffix(name) name ## l
-#include "cordic/cordic_datal.c"
+#include "cordic/cordic_datal.h"
 #include "gen_math.h"
 #undef ftype
 #undef suffix
diff --git a/src/win/win_heap.c b/src/win/win_heap.c
index ece3e54..28a67c6 100644
--- a/src/win/win_heap.c
+++ b/src/win/win_heap.c
@@ -64,7 +64,8 @@ void free(void *ptr) {
 }
 
 void *malloc(size_t size) {
-    return aligned_alloc(8, size);
+    if(size == 0) return NULL;
+    return HeapAlloc(heap_handle, 0, size);
 }
 
 void *realloc(void *ptr, size_t size) {
diff --git a/test/test_printf.c b/test/test_printf.c
index 134a57e..39afd6d 100644
--- a/test/test_printf.c
+++ b/test/test_printf.c
@@ -4,14 +4,7 @@
 #include <limits.h>
 
 int main() {
-    printf("%a\n", 0.0);
-    printf("%a\n", 2.0);
-    printf("%#a\n", 128.0);
-    printf("%a\n", 128.0);
-    printf("%16a|\n", 128.0);
-    printf("%16.2a|\n", 128.0);
-    printf("%-16.2a|\n", 128.0);
-    printf("%016.2a|\n", 128.0);
-    printf("%.2A\n", 3.141592);
+    printf("%g\n", 1.2345e+10);
+    printf("%g\n", 123.45);
     return 0;
 }
diff --git a/todo b/todo
index 1215597..9833d1b 100644
--- a/todo
+++ b/todo
@@ -31,15 +31,21 @@ stdio.h:
 *   The only reason to keep a linked list of all streams is to make sure they
     are flushed after main() returns. I wonder if only remembering the files
     with a buffer would increase performance.
-*   Formatted print, scan
+*   Formatted scan
+*   The float printing should include width specifiers
+*   %g should use decimal exponent
 
 stdlib.h:
 *   Strtod base 16 must be correctly rounded
 *   Multibyte string functions
 *   Better PRNG
+*   Fix aligned_malloc
 
 threads.h:
 *   TODO: add todo items
 
 wchar.h:
 *   Basically everything
+
+entry:
+*   Pretty sure there is a bug in counting wide chars for parsing argv