diff --git a/Lua/inject.lua b/Lua/inject.lua
index 0acaee1..c6dffea 100644
--- a/Lua/inject.lua
+++ b/Lua/inject.lua
@@ -1,6 +1,7 @@
 require = require "depend"
 require "boilerplate"
 require "addrs.init"
+require "messages"
 local assemble = require "inject.lips"
 
 local injection_points = {
@@ -67,12 +68,13 @@ function inject(fn)
     -- set up a header to handle calling our function and the original
     local header = header:format(ow_before_addr)
 
-    local inject_words = {}
+    local inject_bytes = {}
     local length = 0
-    local function add_word(line)
-        -- takes assembled code, up to 4 bytes at a time, as hex strings
+    local function add_word(pos, line)
         length = length + #line/2
-        table.insert(inject_words, tonumber(line, 16))
+        dprint(("%08X"):format(pos), line)
+        pos = pos % 0x80000000
+        inject_bytes[pos] = tonumber(line, 16)
     end
 
     -- offset assembly labels so they work properly, and assemble!
@@ -81,14 +83,15 @@ function inject(fn)
     assemble(asm_path, add_word, {unsafe=true, offset=true_offset + length})
 
     printf("length: %i words", length/4)
-    if #inject_words > inject_maxlen then
+    if #inject_bytes > inject_maxlen then
         print("Assembly too large!")
         return
     end
 
-    for i, v in ipairs(inject_words) do
-        W4(inject_addr + (i - 1)*4, v)
+    for pos, val in pairs(inject_bytes) do
+        W1(pos, val)
     end
+    print_deferred()
 
     -- finally, write our new jump over the original
     printf('%08X: %08X', ow_addr, ow_after)
diff --git a/Lua/inject/lips.lua b/Lua/inject/lips.lua
index 7bbf918..1ca76c9 100644
--- a/Lua/inject/lips.lua
+++ b/Lua/inject/lips.lua
@@ -473,9 +473,7 @@ function Dumper:init(writer, fn, options)
     self.options = options or {}
     self.labels = {}
     self.commands = {}
-    self.buff = ''
-    self.pos = 0
-    self.size = 0
+    self.pos = options.offset or 0
     self.lastcommand = nil
 end
 
@@ -586,7 +584,7 @@ function Lexer:read_number()
             return self:read_hex()
         elseif self.chr == '0' and self.chr2:find('%d') then
             self:nextc()
-            return self:read_octal_number()
+            return self:read_octal()
         else
             return self:read_decimal()
         end
@@ -670,13 +668,11 @@ end
 function Lexer:lex_string(yield)
     -- TODO: support escaping
     if self.chr ~= '"' then
-        print(self.chr, self.ord)
         self:error("expected opening double quote")
     end
     self:nextc()
     local buff = self:read_chars('[^"\n]')
     if self.chr ~= '"' then
-        print(self.chr)
         self:error("expected closing double quote")
     end
     self:nextc()
@@ -1347,15 +1343,11 @@ function Parser:parse(asm)
 end
 
 function Dumper:error(msg)
-    --error(string.format('%s:%d: Dumper Error: %s', '(code)', self.pos, msg), 2)
     error(string.format('%s:%d: Error: %s', self.fn, self.line, msg), 2)
 end
 
 function Dumper:advance(by)
     self.pos = self.pos + by
-    if self.pos > self.size then
-        self.size = self.pos
-    end
 end
 
 function Dumper:push_instruction(t)
@@ -1457,8 +1449,7 @@ function Dumper:desym(tok)
         if label == nil then
             self:error('undefined label')
         end
-        local offset = self.options.offset or 0
-        return label + offset
+        return label
     elseif tok[1] == 'LABELREL' then
         local label = self.labels[tok[2]]
         if label == nil then
@@ -1540,16 +1531,9 @@ function Dumper:valvar(tok, bits)
 end
 
 function Dumper:write(t)
-    -- this is gonna be really slow, but eh, optimization comes last
-    -- should really use a sparse table and fill in the string later
     for _, b in ipairs(t) do
-        if self.pos >= self.size then
-            error('Internal Error: pos out of range; size too small', 1)
-        end
         local s = ('%02X'):format(b)
-        local left = self.buff:sub(1, self.pos*2)
-        local right = self.buff:sub(self.pos*2 + 3)
-        self.buff = left..s..right
+        self.writer(self.pos, s)
         self.pos = self.pos + 1
     end
 end
@@ -1583,12 +1567,7 @@ function Dumper:dump_instruction(t)
 end
 
 function Dumper:dump()
-    self.pos = 0
-    self.buff = ''
-    for i=1,self.size do
-        self.buff = self.buff..'00'
-    end
-
+    self.pos = self.options.offset or 0
     for i, t in ipairs(self.commands) do
         if t.line == nil then
             error('Internal Error: no line number available')
@@ -1617,10 +1596,6 @@ function Dumper:dump()
             error('Internal Error: unknown command', 1)
         end
     end
-
-    for i=1, self.size*2 - 1, 8 do
-        self.writer(self.buff:sub(i, i + 7))
-    end
 end
 
 function assembler.assemble(fn_or_asm, writer, options)