Core Language Architecture

This is a prototype based object-oriented language. This means that the language is structured around a datastructure called a proto (in C: a structure, in Lisp: a frame, in Pascal: a record). This is a structure that has data items. Those data items are accessed by keys. These keys are strings. All of the data structures are 'called' when accessed. If the structure is a number, then the routine called is designed to return a number. Special: there should also be a way for a proto to override the action call and interpret it. All actions take parameters passed by value (you can even think of int's or string literals this way as well).


Proto: a prototype object
Slot:     the single association of a unique key and it's contents
Method:    A slot that activates user code
Soup:    a group of objects in a persistant state (in SimpleData format)


<Need to make this a side story>The reason for this is mostly due to Steve Dekorte pushing his view. Originally, I have always been interested in programming. I learned BASIC. Then I learned Assembly. I missed out on Forth at this time and that held back my development. Then I learned Pascal. Then I learned C. At this point, I was interested in Objective-C and Smalltalk although those languages didn't "make it." I learned C++ and Java just for survival purposes. At the same time, I became very interested in Nextstep and Objective-C (I remember reading Brad Cox's article in Byte magazine).I never learned Lua, but any ideas in it that were good were passed along by Steve. In the begining, I was just interested in doing a class based system. Towards the end, I found that the simplicity of a system like Self was just as powerful and much more dynamic (like Steve said) even though the performance hits were obvious. I was concerned with speed and efficiency, but towards the end I began to see that a proto system would be just plain neat and much more rapid for development.

This language has it's influences from Self and Smalltalk. Although, there is some heavy underground influence from Objective-C, Newtonscript, and Forth. After looking at all of the languages, I settled on modeling Bubl against Self, which is heavily derived from Smalltalk (which was influenced by even more languages, most notably lisp). The "patterns" or object systems will be heavily influenced from  Nextstep and the Mac. There will also be some Windows and Unix influences. The most significant of those will be the Nextstep environment. I have never used Self and the only Smalltalk I used was Smalltalk/V and Squeak. I programmed in neither of those versions in any real sense.

While constructing this language, I had to relearn everything I took for granted in a language. There is a lot more going on underneath the hood than I had originally learned. Many old papers on prototype systems and Self were digested over and over.


The combinations for Bubl are broken down by the kind of architecture being used. 32 bit or 64 bit. Right now, we're using 32 bit.

The name of the language: originally I was going to call it something 'self' like. Then I opted for bubl. Then I decided to use a name I had reserved for something else, Aqua. Finally, since Apple came out with that name, I'm choosing between Proto and SmallSelf and maybe Atom.


The grammar will essentially be the same as Self. The grammar is only neccessary for TEXT input. So, the grammar/text is really a way to describe the objects in the system (code and data). In other languages, they are much better at describing code. Note again, this is just for the text side of things. A visual environment can interact and build protos in their native form. If they needed to see the code for a method, they could easily ask a code object to describe itself and the decompiler could represent the code as text.

Essentially, an Aqua like module is described - a soup.

*NOTE* It will be important to be able to describe a module from this level so libraries and other special modules can be built. For example, it will be important to describe a library module so it's module information is set properly.

The key grammar, that we are mainly concerned with then is the grammer for the code slots.

Wherever there is a space, that is a WHITESPACE
WhiteSpace: '\n' SPACE TAB

Proto:  '[' Objects Code ']'
Objects: '|'  Param* Local* '|'
Param: ':'name
Local: name | LocalSet
LocalSet: name | initializedName | constantName | parentName
initializedName: name '<-' Value
constantName: name '=' Value
parentName: name'*'

Initialized/Param: These create two slots, an 'x' and an 'x:'
Code creates a separate proto and a PASSTHRU slot (see implementation)

???? What about using parentheses ? - i need to finalize the grammar.


The basic types in the system will be Protos, Numbers, Strings, Opaque/Blob, Code objects, Object references, and Dates.

Protos: (| |)
Numbers: Signed 30 bit Integer    Base'r'Number (Example: 16rff0056, 2r1011101, 12345)
Strings:  'this is a string\'s string\n'
Dates: @YYY[MM[DD[HH:MM][] ] ]   //// What about timezone? What about just specifiying a time??

Special Names and Values

nil, true, false, 1, -1, 0, ???? self


There has to be a way to specify outside modules. It has to be done in a way that allows a 'program generator' to easily identify.

The idea I had so far was to have a prebuilt Global that would be called 'Modules'. When that global had a slot accessed under it, Now, if
the slot already exists, the current mechanisms will work. Or we can have a 'forward:' or poseas or someother method that would allow a proto to examine the activation record and make a determination there. So when the module builds the slots for each Module, it can determine which slot name was called. Or it can set a variable and call a super.


The namespace of a proto is a hierarchy. The 'self' proto is searched first for a string.  If it isn't found, it starts going up it's parent chain. When it reaches a proto without a parent, then it is done. There is one exception to the rule, this is the Globals section. If something is refered to as Global Modules: SerialIO, then it searches the Globals OR if the search fails, it will look in the Globals proto. This is essential having a slot called "Globals" in every prototype. Without this, there would be no way to set parents to pre-built behaviors. That's all. Pretty simple, huh.

Then there is 'self' this refers tot he proto that made the call and created an activation record.


Exceptions are a good idea. The basic idea is that errors are special. They can automatically pass up through the contexts to find an 'error handler'. These errors are called exceptions and those handler's are called exception handlers. The programmer isn't forced to use this mechanism, but just about every library will probably use it.

Exceptions will be objects, just like everything else. There won't be any special syntax either. There will be a method call to a special object that will do the necessary low level work.

At this point, I haven't thought about how I will implement this completely.

Multiple-Return Values

The system will need to be able to return multiple, varargs...

The Virtual Machine

The code in the proto-environment will run in a virtual machine. The first firtual machine we have implemented is the vm32 environment. It is a simple stack based VM. It doesn't have any speicial hooks for this environment, but that may change.

Low Level Implementation

The slots in a proto have three fields. Two are the only ones visible, but it is the third that makes things interesting. This is how Forth influences this language.  In forth, each 'word' is a record in a 'dictionary'. The first three characters of the 'word' are stored there. After that, there is a link to the next word in the dictionary list, then there is a code pointer, and then there is data. The idea here is that when this word is executed, it is up to the user defined code to perform some action. The code gets a pointer to the data section for that word. In other languages, this is completely hidden or not in programmer control. Therefore, you can't do some of the things that forth can do. This is more OO in the sense that code manipulates that data, not you; and there is no difference in interface when manipulating data or activating code - it is all activating code. Some examples of the code in Forth are, variable retrieval (constants, and read write), code execution (the data portion is actual assembly), threaded-code execution (the data is an array of pointers to words in the dictionary). This is a powerful pattern and is something I decided to use in Bubl as well.

As stated earlier, Bubl uses slots with fields. You access the slot with the key string. This slot may be read/write data, read only data, or code. The other hidden field in the slot is set when the object is created and determines what kind of slot this is...

GET - Gets the value stored in type (usually a pointer)
SET - Sets the value portion of the slot
GETSTACK - The same, except the data holds the position relative to the stack frame
SETSTACK - The same, except the data holds the position relative to the stack frame
PASSTHRU - The data holds the address of the code proto it calls the proper value:value: or value: etc. method in the code proto
CODE - Execute the code stored in the 'CODE' slot in a code proto


Pointers will be to Ints, Proto's, Blobs, and Unique strings

   Bit 00 - Ints (+- 500 million signed)
         01 - Protos - (these are at least 4 bytes)
         10 - Blobs  - (these are at least 4 bytes)
         11 - Unique strings / literals (each string is at least 2 bytes,
                 string + 0 byte terminator) so this address is shifted down,
                 by one, thereby elmiminating the top 2 Gig space)

     When a module is linked in,  the first thing that happens is that the string literals all get transformed into their 32 bit quantities.

Proto Hash tables. Hash tables will have a header (described in the malloc portion of this document).The proto hash tables will have a size in the header. Each row/slot will have the key and and the data portion. The key field will be a 32 bit value and it will require 29 of the 32 bits. The remaining three bits will be used to indicate which routine will be called for this proto. Now if our pointers use 30 bits, then we are 1 bit short. So, if we kept to this, limitation, we would have to keep strings in 8 byte alignment blocks. The reality of string literals in a system like this is that the strings will occupy 4 or more bytes, a lot of the time. So, with the zero terminator, the alignment of bytes is not a bad waste. What we will do is store the left shifted pointer for literals. So, we will make sure that the strings will not occupy the upper 2 GIG of a system's virtual address space. This will be easy to guarantee since the buffer for the copies of these strings will be in heap space. When we attempt to get the value of a literal, we will have to remember to shift it right once and mask out the type to access the value.

Each associate is 8 bytes - the minimum for malloc is 16 bytes...

Chaining vs. Not. Hash vs. Not.

Chaining: chaining requires that the associations can store pointers to the next entry in the hash table. This allows the hash table to grow without having to move the table around often. (reallocs). If a table gets realloc'd, the pointers to it must be changed.

When a slot is added, the key is applied with a hashing function, this then determines the position into the table.

Memory Types

Header in malloc chunks:

 In Use
   bit 0 - 0                  Previous Free
           1                  Previous in Use/Alloced
   Bit 1 - 0                  Non-GC
           1                  GC
                Automatic (GC)             Manual
   Bit 2 - 0    Blob/Alloc                 Alloc
           1    Proto                      Small Alloc? Some other type?
   The Rest are only for GC objects
   Bit 3 - 0    Not Marked                 Unused
           1    Marked                     Unused
   Bit 4 - 0    Generation                 Unused
           1    Marked                     Unused
   Size         Blob, Free - Easy -
                Blob (Size (only so many bits used, and it stands for 16 byte groups), Bytes),
                Free (Size (only so many bits used and it stands for 16 byte groups), PTRS, Footer)
Also, must not forget 'forwarders' - a structure that points to a proto's actual location (due to a size increase). Once the garbage collector
hits a forwarder through two cycles, all references should be gone.

Code Protos

Code protos are like any other proto, except they aren't created by the programmer directly. They are created indirectly when code is compiled. It was nice to treat blocks and code as the same thing and have the resulting objects respond to their proper 'value' message. They are special in that they have their 'self' and their 'parent' slots mapped differently. Their parent slot is stack mapped and their self points to their parent proto.


All code is in VM bytecode and is in position-independent code. When code or names are called for external routines there will be a small table that the code can position independent refer to.

The String Literals

All the protos need to be accessed by strings. Since doing string compares is expensive and accessing protos is a common thing most systems (Smalltalk or Self), usually have what is called a "String Literal". A string literal gets transformed into a number upon linking or compiling of a module. That is, a string is added to a global table. If it is in the table, that the origninal ID is used, otherwise it is added. All strings are unique and case sensitive. Now, when the Bubl runtime needs to compare strings, it can just compare these numbers.

Implementation. So, when strings are added, we need a quick way to determine if the string is in the global table (access path). Also, since strings will be added to the system when the system becomes linked, it is possible for this table to grow. That means that it is better if the strings stayed in place and the table of pointers just moved. So, we will need a hash table that will have an association of the hash of the string and a pointer to it's place in the pointer table. Or we could use the hash table as storage and when a string already exists, we use that value?? I think when I implement this, it will become clear as to how to do it best.

The Activation Record

Activation Record Documentation

Initial Bootstrap

To initially get going. I will need to use Objective-C to build the compiler and assembler. I would like to build one using Bubl, but that just isn't possible. So the lexer,parser,compiler,assembler, and packager each take a specification of objects and use SD format between each step. The end result is a finished module. Once the proto environment can start reading files and generating files, it should be enhanced to do this process.

Intermediate Forms

Once the code is typed in by the programmer for a Bubl soup (a program/application/whatever), the compilers need to make it in SimpleData format. The first step is to place as many of the standard types into that form as possible. This will leave the code sections undescribed. The next step is to pass this simple data module to the lexer/scanner. It will tokenize the code slots in the protos with code and return a resulting SimpleData module. This process continues until a finished module is produced. (Lexer/Scanner, Parser, Optimizer, Code Generator, Assembler)

7/8/98 - First creation, basically everything.
7/9/98 - Morning, a little work
7/10/98 - Morning, add a little about the types, grammar, and (32 bit) vs.(64 bit) transcription of notes. ?n
 7/11/98 - Exceptions, Multiple return types
7/12/98 - Filled out exceptions some more., globals, and lots of low-level implementation details.
7/19/98 - Modified the storage section, put a better description of the string literal storing/processing.
7/10/99 - Moved activation record to it's own document